freebsd-dev/sys/kern/subr_smp.c

1160 lines
28 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2001, John Baldwin <jhb@FreeBSD.org>.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/*
* This module holds the global variables and machine independent functions
* used for the kernel SMP support.
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
1997-08-25 21:28:08 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
#include <sys/ktr.h>
#include <sys/proc.h>
#include <sys/bus.h>
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/pcpu.h>
#include <sys/sched.h>
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <machine/cpu.h>
#include <machine/smp.h>
#include "opt_sched.h"
#ifdef SMP
MALLOC_DEFINE(M_TOPO, "toponodes", "SMP topology data");
2011-05-18 15:50:12 +00:00
volatile cpuset_t stopped_cpus;
volatile cpuset_t started_cpus;
volatile cpuset_t suspended_cpus;
2011-05-18 15:50:12 +00:00
cpuset_t hlt_cpus_mask;
cpuset_t logical_cpus_mask;
2002-03-19 21:25:46 +00:00
void (*cpustop_restartfunc)(void);
#endif
static int sysctl_kern_smp_active(SYSCTL_HANDLER_ARGS);
/* This is used in modules that need to work in both SMP and UP. */
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
cpuset_t all_cpus;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
int mp_ncpus;
/* export this for libkvm consumers. */
int mp_maxcpus = MAXCPU;
volatile int smp_started;
u_int mp_maxid;
static SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD|CTLFLAG_CAPRD, NULL,
"Kernel SMP");
SYSCTL_INT(_kern_smp, OID_AUTO, maxid, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_maxid, 0,
"Max CPU ID.");
SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_maxcpus,
0, "Max number of CPUs that the system was compiled for.");
SYSCTL_PROC(_kern_smp, OID_AUTO, active, CTLFLAG_RD|CTLTYPE_INT|CTLFLAG_MPSAFE,
NULL, 0, sysctl_kern_smp_active, "I",
"Indicates system is running in SMP mode");
int smp_disabled = 0; /* has smp been disabled? */
SYSCTL_INT(_kern_smp, OID_AUTO, disabled, CTLFLAG_RDTUN|CTLFLAG_CAPRD,
&smp_disabled, 0, "SMP has been disabled from the loader");
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
int smp_cpus = 1; /* how many cpu's running */
SYSCTL_INT(_kern_smp, OID_AUTO, cpus, CTLFLAG_RD|CTLFLAG_CAPRD, &smp_cpus, 0,
"Number of CPUs online");
int smp_topology = 0; /* Which topology we're using. */
SYSCTL_INT(_kern_smp, OID_AUTO, topology, CTLFLAG_RDTUN, &smp_topology, 0,
"Topology override setting; 0 is default provided by hardware.");
#ifdef SMP
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* Enable forwarding of a signal to a process running on a different CPU */
static int forward_signal_enabled = 1;
SYSCTL_INT(_kern_smp, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
&forward_signal_enabled, 0,
"Forwarding of a signal to a process on a different CPU");
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* Variables needed for SMP rendezvous. */
static volatile int smp_rv_ncpus;
static void (*volatile smp_rv_setup_func)(void *arg);
static void (*volatile smp_rv_action_func)(void *arg);
2008-01-02 17:09:15 +00:00
static void (*volatile smp_rv_teardown_func)(void *arg);
2009-01-26 15:32:39 +00:00
static void *volatile smp_rv_func_arg;
static volatile int smp_rv_waiters[4];
/*
* Shared mutex to restrict busywaits between smp_rendezvous() and
* smp(_targeted)_tlb_shootdown(). A deadlock occurs if both of these
* functions trigger at once and cause multiple CPUs to busywait with
* interrupts disabled.
*/
struct mtx smp_ipi_mtx;
/*
* Let the MD SMP code initialize mp_maxid very early if it can.
*/
static void
mp_setmaxid(void *dummy)
{
cpu_mp_setmaxid();
KASSERT(mp_ncpus >= 1, ("%s: CPU count < 1", __func__));
KASSERT(mp_ncpus > 1 || mp_maxid == 0,
("%s: one CPU but mp_maxid is not zero", __func__));
KASSERT(mp_maxid >= mp_ncpus - 1,
("%s: counters out of sync: max %d, count %d", __func__,
mp_maxid, mp_ncpus));
}
SYSINIT(cpu_mp_setmaxid, SI_SUB_TUNABLES, SI_ORDER_FIRST, mp_setmaxid, NULL);
/*
* Call the MD SMP initialization code.
*/
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
static void
mp_start(void *dummy)
{
mtx_init(&smp_ipi_mtx, "smp rendezvous", NULL, MTX_SPIN);
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* Probe for MP hardware. */
if (smp_disabled != 0 || cpu_mp_probe() == 0) {
mp_ncpus = 1;
CPU_SETOF(PCPU_GET(cpuid), &all_cpus);
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
return;
}
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
cpu_mp_start();
printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n",
mp_ncpus);
cpu_mp_announce();
}
SYSINIT(cpu_mp, SI_SUB_CPU, SI_ORDER_THIRD, mp_start, NULL);
void
forward_signal(struct thread *td)
{
int id;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/*
* signotify() has already set TDF_ASTPENDING and TDF_NEEDSIGCHECK on
* this thread, so all we need to do is poke it if it is currently
* executing so that it executes ast().
*/
THREAD_LOCK_ASSERT(td, MA_OWNED);
KASSERT(TD_IS_RUNNING(td),
("forward_signal: thread is not TDS_RUNNING"));
CTR1(KTR_SMP, "forward_signal(%p)", td->td_proc);
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
if (!smp_started || cold || panicstr)
return;
if (!forward_signal_enabled)
return;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* No need to IPI ourself. */
if (td == curthread)
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
return;
id = td->td_oncpu;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
if (id == NOCPU)
return;
ipi_cpu(id, IPI_AST);
}
/*
* When called the executing CPU will send an IPI to all other CPUs
* requesting that they halt execution.
*
* Usually (but not necessarily) called with 'other_cpus' as its arg.
*
* - Signals all CPUs in map to stop.
* - Waits for each to stop.
*
* Returns:
* -1: error
* 0: NA
* 1: ok
*
*/
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if defined(__amd64__) || defined(__i386__)
#define X86 1
#else
#define X86 0
#endif
2009-08-13 17:09:45 +00:00
static int
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
generic_stop_cpus(cpuset_t map, u_int type)
{
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
#ifdef KTR
char cpusetbuf[CPUSETBUFSIZ];
#endif
static volatile u_int stopping_cpu = NOCPU;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
int i;
volatile cpuset_t *cpus;
KASSERT(
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
type == IPI_STOP || type == IPI_STOP_HARD
#if X86
|| type == IPI_SUSPEND
#endif
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
, ("%s: invalid stop type", __func__));
2009-08-13 17:09:45 +00:00
if (!smp_started)
return (0);
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
CTR2(KTR_SMP, "stop_cpus(%s) with %u type",
cpusetobj_strprint(cpusetbuf, &map), type);
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
/*
* When suspending, ensure there are are no IPIs in progress.
* IPIs that have been issued, but not yet delivered (e.g.
* not pending on a vCPU when running under virtualization)
* will be lost, violating FreeBSD's assumption of reliable
* IPI delivery.
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
*/
if (type == IPI_SUSPEND)
mtx_lock_spin(&smp_ipi_mtx);
#endif
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
if (!nmi_is_broadcast || nmi_kdb_lock == 0) {
#endif
if (stopping_cpu != PCPU_GET(cpuid))
while (atomic_cmpset_int(&stopping_cpu, NOCPU,
PCPU_GET(cpuid)) == 0)
while (stopping_cpu != NOCPU)
cpu_spinwait(); /* spin */
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* send the stop IPI to all CPUs in map */
2009-08-13 17:09:45 +00:00
ipi_selected(map, type);
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
}
#endif
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
if (type == IPI_SUSPEND)
cpus = &suspended_cpus;
else
#endif
cpus = &stopped_cpus;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
i = 0;
while (!CPU_SUBSET(cpus, &map)) {
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* spin */
cpu_spinwait();
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
i++;
if (i == 100000000) {
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
printf("timeout stopping cpus\n");
break;
}
}
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
if (type == IPI_SUSPEND)
mtx_unlock_spin(&smp_ipi_mtx);
#endif
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
stopping_cpu = NOCPU;
return (1);
}
2009-08-13 17:09:45 +00:00
int
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
stop_cpus(cpuset_t map)
2009-08-13 17:09:45 +00:00
{
return (generic_stop_cpus(map, IPI_STOP));
}
int
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
stop_cpus_hard(cpuset_t map)
2009-08-13 17:09:45 +00:00
{
return (generic_stop_cpus(map, IPI_STOP_HARD));
}
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
int
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
suspend_cpus(cpuset_t map)
{
return (generic_stop_cpus(map, IPI_SUSPEND));
}
#endif
/*
* Called by a CPU to restart stopped CPUs.
*
* Usually (but not necessarily) called with 'stopped_cpus' as its arg.
*
* - Signals all CPUs in map to restart.
* - Waits for each to restart.
*
* Returns:
* -1: error
* 0: NA
* 1: ok
*/
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
static int
generic_restart_cpus(cpuset_t map, u_int type)
{
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
#ifdef KTR
char cpusetbuf[CPUSETBUFSIZ];
#endif
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
volatile cpuset_t *cpus;
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
KASSERT(type == IPI_STOP || type == IPI_STOP_HARD
#if X86
|| type == IPI_SUSPEND
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
#endif
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
, ("%s: invalid stop type", __func__));
if (!smp_started)
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
return (0);
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
CTR1(KTR_SMP, "restart_cpus(%s)", cpusetobj_strprint(cpusetbuf, &map));
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
if (type == IPI_SUSPEND)
Use resume_cpus() instead of restart_cpus() to resume from ACPI suspension. restart_cpus() worked well enough by accident. Before this set of fixes, resume_cpus() used the same cpuset (started_cpus, meaning CPUs directed to restart) as restart_cpus(). resume_cpus() waited for the wrong cpuset (stopped_cpus) to become empty, but since mixtures of stopped and suspended CPUs are not close to working, stopped_cpus must be empty when resuming so the wait is null -- restart_cpus just allows the other CPUs to restart and returns without waiting. Fix resume_cpus() to wait on a non-wrong cpuset for the ACPI case, and add further kludges to try to keep it working for the XEN case. It was only used for XEN. It waited on suspended_cpus. This works for XEN. However, for ACPI, resuming is a 2-step process. ACPI has already woken up the other CPUs and removed them from suspended_cpus. This fix records the move by putting them in a new cpuset resuming_cpus. Waiting on suspended_cpus would give the same null wait as waiting on stopped_cpus. Wait on resuming_cpus instead. Add a cpuset toresume_cpus to map the CPUs being told to resume to keep this separate from the cpuset started_cpus for mapping the CPUs being told to restart. Mixtures of stopped and suspended/resuming CPUs are still far from working. Describe new and some old cpusets in comments. Add further kludges to cpususpend_handler() to try to avoid breaking it for XEN. XEN doesn't use resumectx(), so it doesn't use the second return path for savectx(), and it goes from the suspended state directly to the restarted state, while ACPI resume goes through the resuming state. Enter the resuming state early for all cases so that resume_cpus can test for being in this state and not have to worry about the intermediate !suspended state for ACPI only. Reviewed by: kib
2017-12-21 09:17:48 +00:00
cpus = &resuming_cpus;
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
else
#endif
cpus = &stopped_cpus;
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* signal other cpus to restart */
Use resume_cpus() instead of restart_cpus() to resume from ACPI suspension. restart_cpus() worked well enough by accident. Before this set of fixes, resume_cpus() used the same cpuset (started_cpus, meaning CPUs directed to restart) as restart_cpus(). resume_cpus() waited for the wrong cpuset (stopped_cpus) to become empty, but since mixtures of stopped and suspended CPUs are not close to working, stopped_cpus must be empty when resuming so the wait is null -- restart_cpus just allows the other CPUs to restart and returns without waiting. Fix resume_cpus() to wait on a non-wrong cpuset for the ACPI case, and add further kludges to try to keep it working for the XEN case. It was only used for XEN. It waited on suspended_cpus. This works for XEN. However, for ACPI, resuming is a 2-step process. ACPI has already woken up the other CPUs and removed them from suspended_cpus. This fix records the move by putting them in a new cpuset resuming_cpus. Waiting on suspended_cpus would give the same null wait as waiting on stopped_cpus. Wait on resuming_cpus instead. Add a cpuset toresume_cpus to map the CPUs being told to resume to keep this separate from the cpuset started_cpus for mapping the CPUs being told to restart. Mixtures of stopped and suspended/resuming CPUs are still far from working. Describe new and some old cpusets in comments. Add further kludges to cpususpend_handler() to try to avoid breaking it for XEN. XEN doesn't use resumectx(), so it doesn't use the second return path for savectx(), and it goes from the suspended state directly to the restarted state, while ACPI resume goes through the resuming state. Enter the resuming state early for all cases so that resume_cpus can test for being in this state and not have to worry about the intermediate !suspended state for ACPI only. Reviewed by: kib
2017-12-21 09:17:48 +00:00
#if X86
if (type == IPI_SUSPEND)
CPU_COPY_STORE_REL(&map, &toresume_cpus);
else
#endif
CPU_COPY_STORE_REL(&map, &started_cpus);
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
if (!nmi_is_broadcast || nmi_kdb_lock == 0) {
#endif
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
/* wait for each to clear its bit */
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
while (CPU_OVERLAP(cpus, &map))
cpu_spinwait();
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
}
#endif
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
return (1);
}
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
int
restart_cpus(cpuset_t map)
{
return (generic_restart_cpus(map, IPI_STOP));
}
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#if X86
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
int
resume_cpus(cpuset_t map)
{
return (generic_restart_cpus(map, IPI_SUSPEND));
}
#endif
Handle broadcast NMIs. On several Intel chipsets, diagnostic NMIs sent from BMC or NMIs reporting hardware errors are broadcasted to all CPUs. When kernel is configured to enter kdb on NMI, the outcome is problematic, because each CPU tries to enter kdb. All CPUs are executing NMI handlers, which set the latches disabling the nested NMI delivery; this means that stop_cpus_hard(), used by kdb_enter() to stop other cpus by broadcasting IPI_STOP_HARD NMI, cannot work. One indication of this is the harmless but annoying diagnostic "timeout stopping cpus". Much more harming behaviour is that because all CPUs try to enter kdb, and if ddb is used as debugger, all CPUs issue prompt on console and race for the input, not to mention the simultaneous use of the ddb shared state. Try to fix this by introducing a pseudo-lock for simultaneous attempts to handle NMIs. If one core happens to enter NMI trap handler, other cores see it and simulate reception of the IPI_STOP_HARD. More, generic_stop_cpus() avoids sending IPI_STOP_HARD and avoids waiting for the acknowledgement, relying on the nmi handler on other cores suspending and then restarting the CPU. Since it is impossible to detect at runtime whether some stray NMI is broadcast or unicast, add a knob for administrator (really developer) to configure debugging NMI handling mode. The updated patch was debugged with the help from Andrey Gapon (avg) and discussed with him. Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8249
2016-10-24 16:40:27 +00:00
#undef X86
Add support for suspend/resume/migration operations when running as a Xen PVHVM guest. Submitted by: Roger Pau Monné Sponsored by: Citrix Systems R&D Reviewed by: gibbs Approved by: re (blanket Xen) MFC after: 2 weeks sys/amd64/amd64/mp_machdep.c: sys/i386/i386/mp_machdep.c: - Make sure that are no MMU related IPIs pending on migration. - Reset pending IPI_BITMAP on resume. - Init vcpu_info on resume. sys/amd64/include/intr_machdep.h: sys/i386/include/intr_machdep.h: sys/x86/acpica/acpi_wakeup.c: sys/x86/x86/intr_machdep.c: sys/x86/isa/atpic.c: sys/x86/x86/io_apic.c: sys/x86/x86/local_apic.c: - Add a "suspend_cancelled" parameter to pic_resume(). For the Xen PIC, restoration of interrupt services differs between the aborted suspend and normal resume cases, so we must provide this information. sys/dev/acpica/acpi_timer.c: sys/dev/xen/timer/timer.c: sys/timetc.h: - Don't swap out "suspend safe" timers across a suspend/resume cycle. This includes the Xen PV and ACPI timers. sys/dev/xen/control/control.c: - Perform proper suspend/resume process for PVHVM: - Suspend all APs before going into suspension, this allows us to reset the vcpu_info on resume for each AP. - Reset shared info page and callback on resume. sys/dev/xen/timer/timer.c: - Implement suspend/resume support for the PV timer. Since FreeBSD doesn't perform a per-cpu resume of the timer, we need to call smp_rendezvous in order to correctly resume the timer on each CPU. sys/dev/xen/xenpci/xenpci.c: - Don't reset the PCI interrupt on each suspend/resume. sys/kern/subr_smp.c: - When suspending a PVHVM domain make sure there are no MMU IPIs in-flight, or we will get a lockup on resume due to the fact that pending event channels are not carried over on migration. - Implement a generic version of restart_cpus that can be used by suspended and stopped cpus. sys/x86/xen/hvm.c: - Implement resume support for the hypercall page and shared info. - Clear vcpu_info so it can be reset by APs when resuming from suspension. sys/dev/xen/xenpci/xenpci.c: sys/x86/xen/hvm.c: sys/x86/xen/xen_intr.c: - Support UP kernel configurations. sys/x86/xen/xen_intr.c: - Properly rebind per-cpus VIRQs and IPIs on resume.
2013-09-20 05:06:03 +00:00
/*
* All-CPU rendezvous. CPUs are signalled, all execute the setup function
* (if specified), rendezvous, execute the action function (if specified),
* rendezvous again, execute the teardown function (if specified), and then
* resume.
*
* Note that the supplied external functions _must_ be reentrant and aware
* that they are running in parallel and in an unknown lock context.
*/
void
smp_rendezvous_action(void)
{
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
struct thread *td;
void *local_func_arg;
void (*local_setup_func)(void*);
void (*local_action_func)(void*);
void (*local_teardown_func)(void*);
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
#ifdef INVARIANTS
int owepreempt;
#endif
2008-01-02 17:09:15 +00:00
/* Ensure we have up-to-date values. */
atomic_add_acq_int(&smp_rv_waiters[0], 1);
while (smp_rv_waiters[0] < smp_rv_ncpus)
cpu_spinwait();
/* Fetch rendezvous parameters after acquire barrier. */
local_func_arg = smp_rv_func_arg;
local_setup_func = smp_rv_setup_func;
local_action_func = smp_rv_action_func;
local_teardown_func = smp_rv_teardown_func;
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
/*
* Use a nested critical section to prevent any preemptions
* from occurring during a rendezvous action routine.
* Specifically, if a rendezvous handler is invoked via an IPI
* and the interrupted thread was in the critical_exit()
* function after setting td_critnest to 0 but before
* performing a deferred preemption, this routine can be
* invoked with td_critnest set to 0 and td_owepreempt true.
* In that case, a critical_exit() during the rendezvous
* action would trigger a preemption which is not permitted in
* a rendezvous action. To fix this, wrap all of the
* rendezvous action handlers in a critical section. We
* cannot use a regular critical section however as having
* critical_exit() preempt from this routine would also be
* problematic (the preemption must not occur before the IPI
* has been acknowledged via an EOI). Instead, we
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
* intentionally ignore td_owepreempt when leaving the
* critical section. This should be harmless because we do
* not permit rendezvous action routines to schedule threads,
* and thus td_owepreempt should never transition from 0 to 1
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
* during this routine.
*/
td = curthread;
td->td_critnest++;
#ifdef INVARIANTS
owepreempt = td->td_owepreempt;
#endif
/*
* If requested, run a setup function before the main action
* function. Ensure all CPUs have completed the setup
* function before moving on to the action function.
*/
if (local_setup_func != smp_no_rendezvous_barrier) {
if (smp_rv_setup_func != NULL)
smp_rv_setup_func(smp_rv_func_arg);
atomic_add_int(&smp_rv_waiters[1], 1);
while (smp_rv_waiters[1] < smp_rv_ncpus)
cpu_spinwait();
}
if (local_action_func != NULL)
local_action_func(local_func_arg);
if (local_teardown_func != smp_no_rendezvous_barrier) {
/*
* Signal that the main action has been completed. If a
* full exit rendezvous is requested, then all CPUs will
* wait here until all CPUs have finished the main action.
*/
atomic_add_int(&smp_rv_waiters[2], 1);
while (smp_rv_waiters[2] < smp_rv_ncpus)
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
cpu_spinwait();
if (local_teardown_func != NULL)
local_teardown_func(local_func_arg);
}
2008-01-02 17:09:15 +00:00
/*
* Signal that the rendezvous is fully completed by this CPU.
* This means that no member of smp_rv_* pseudo-structure will be
* accessed by this target CPU after this point; in particular,
* memory pointed by smp_rv_func_arg.
*
* The release semantic ensures that all accesses performed by
* the current CPU are visible when smp_rendezvous_cpus()
* returns, by synchronizing with the
* atomic_load_acq_int(&smp_rv_waiters[3]).
*/
atomic_add_rel_int(&smp_rv_waiters[3], 1);
Fix an issue with critical sections and SMP rendezvous handlers. Specifically, a critical_exit() call that drops the nesting level to zero has a brief window where the pending preemption flag is set and the nesting level is set to zero. This is done purposefully to avoid races where a preemption scheduled by an interrupt could be lost otherwise (see revision 144777). However, this does mean that if an interrupt fires during this window and enters and exits a critical section, it may preempt from the interrupt context. This is generally fine as the interrupt code is careful to arrange critical sections so that they are not exited until it is safe to preempt (e.g. interrupts EOI'd and masked if necessary). However, the SMP rendezvous IPI handler does not quite follow this rule, and in general a rendezvous can never be preempted. Rendezvous handlers are also not permitted to schedule threads to execute, so they will not typically trigger preemptions. SMP rendezvous handlers may use spinlocks (carefully) such as the rm_cleanIPI() handler used in rmlocks, but using a spinlock also enters and exits a critical section. If the interrupted top-half code is in the brief window of critical_exit() where the nesting level is zero but a preemption is pending, then releasing the spinlock can trigger a preemption. Because we know that SMP rendezvous handlers can never schedule a thread, we know that a critical_exit() in an SMP rendezvous handler will only preempt in this edge case. We also know that the top-half thread will happily handle the deferred preemption once the SMP rendezvous has completed, so the preemption will not be lost. This makes it safe to employ a workaround where we use a nested critical section in the SMP rendezvous code itself around rendezvous action routines to prevent any preemptions during an SMP rendezvous. The workaround intentionally avoids checking for a deferred preemption when leaving the critical section on the assumption that if there is a pending preemption it will be handled by the interrupted top-half code. Submitted by: mlaier (variation specific to rm_cleanIPI()) Obtained from: Isilon MFC after: 1 week
2011-05-24 13:36:41 +00:00
td->td_critnest--;
KASSERT(owepreempt == td->td_owepreempt,
("rendezvous action changed td_owepreempt"));
}
void
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
smp_rendezvous_cpus(cpuset_t map,
void (* setup_func)(void *),
void (* action_func)(void *),
void (* teardown_func)(void *),
void *arg)
{
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
int curcpumap, i, ncpus = 0;
/* Look comments in the !SMP case. */
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
if (!smp_started) {
spinlock_enter();
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
teardown_func(arg);
spinlock_exit();
Overhaul of the SMP code. Several portions of the SMP kernel support have been made machine independent and various other adjustments have been made to support Alpha SMP. - It splits the per-process portions of hardclock() and statclock() off into hardclock_process() and statclock_process() respectively. hardclock() and statclock() call the *_process() functions for the current process so that UP systems will run as before. For SMP systems, it is simply necessary to ensure that all other processors execute the *_process() functions when the main clock functions are triggered on one CPU by an interrupt. For the alpha 4100, clock interrupts are delievered in a staggered broadcast fashion, so we simply call hardclock/statclock on the boot CPU and call the *_process() functions on the secondaries. For x86, we call statclock and hardclock as usual and then call forward_hardclock/statclock in the MD code to send an IPI to cause the AP's to execute forwared_hardclock/statclock which then call the *_process() functions. - forward_signal() and forward_roundrobin() have been reworked to be MI and to involve less hackery. Now the cpu doing the forward sets any flags, etc. and sends a very simple IPI_AST to the other cpu(s). AST IPIs now just basically return so that they can execute ast() and don't bother with setting the astpending or needresched flags themselves. This also removes the loop in forward_signal() as sched_lock closes the race condition that the loop worked around. - need_resched(), resched_wanted() and clear_resched() have been changed to take a process to act on rather than assuming curproc so that they can be used to implement forward_roundrobin() as described above. - Various other SMP variables have been moved to a MI subr_smp.c and a new header sys/smp.h declares MI SMP variables and API's. The IPI API's from machine/ipl.h have moved to machine/smp.h which is included by sys/smp.h. - The globaldata_register() and globaldata_find() functions as well as the SLIST of globaldata structures has become MI and moved into subr_smp.c. Also, the globaldata list is only available if SMP support is compiled in. Reviewed by: jake, peter Looked over by: eivind
2001-04-27 19:28:25 +00:00
return;
}
CPU_FOREACH(i) {
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (CPU_ISSET(i, &map))
ncpus++;
}
if (ncpus == 0)
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
panic("ncpus is 0 with non-zero map");
mtx_lock_spin(&smp_ipi_mtx);
/* Pass rendezvous parameters via global variables. */
smp_rv_ncpus = ncpus;
smp_rv_setup_func = setup_func;
smp_rv_action_func = action_func;
smp_rv_teardown_func = teardown_func;
smp_rv_func_arg = arg;
smp_rv_waiters[1] = 0;
smp_rv_waiters[2] = 0;
smp_rv_waiters[3] = 0;
atomic_store_rel_int(&smp_rv_waiters[0], 0);
/*
* Signal other processors, which will enter the IPI with
* interrupts off.
*/
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
curcpumap = CPU_ISSET(curcpu, &map);
CPU_CLR(curcpu, &map);
ipi_selected(map, IPI_RENDEZVOUS);
/* Check if the current CPU is in the map */
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (curcpumap != 0)
smp_rendezvous_action();
/*
* Ensure that the master CPU waits for all the other
* CPUs to finish the rendezvous, so that smp_rv_*
* pseudo-structure and the arg are guaranteed to not
* be in use.
*
* Load acquire synchronizes with the release add in
* smp_rendezvous_action(), which ensures that our caller sees
* all memory actions done by the called functions on other
* CPUs.
*/
while (atomic_load_acq_int(&smp_rv_waiters[3]) < ncpus)
cpu_spinwait();
2008-01-02 17:09:15 +00:00
mtx_unlock_spin(&smp_ipi_mtx);
}
void
smp_rendezvous(void (* setup_func)(void *),
void (* action_func)(void *),
void (* teardown_func)(void *),
void *arg)
{
smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func, arg);
}
static struct cpu_group group[MAXCPU * MAX_CACHE_LEVELS + 1];
struct cpu_group *
smp_topo(void)
{
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ];
struct cpu_group *top;
/*
* Check for a fake topology request for debugging purposes.
*/
switch (smp_topology) {
case 1:
/* Dual core with no sharing. */
top = smp_topo_1level(CG_SHARE_NONE, 2, 0);
break;
case 2:
/* No topology, all cpus are equal. */
top = smp_topo_none();
break;
case 3:
/* Dual core with shared L2. */
top = smp_topo_1level(CG_SHARE_L2, 2, 0);
break;
case 4:
/* quad core, shared l3 among each package, private l2. */
top = smp_topo_1level(CG_SHARE_L3, 4, 0);
break;
case 5:
/* quad core, 2 dualcore parts on each package share l2. */
top = smp_topo_2level(CG_SHARE_NONE, 2, CG_SHARE_L2, 2, 0);
break;
case 6:
/* Single-core 2xHTT */
top = smp_topo_1level(CG_SHARE_L1, 2, CG_FLAG_HTT);
break;
case 7:
/* quad core with a shared l3, 8 threads sharing L2. */
top = smp_topo_2level(CG_SHARE_L3, 4, CG_SHARE_L2, 8,
CG_FLAG_SMT);
break;
default:
/* Default, ask the system what it wants. */
top = cpu_topo();
break;
}
/*
* Verify the returned topology.
*/
if (top->cg_count != mp_ncpus)
panic("Built bad topology at %p. CPU count %d != %d",
top, top->cg_count, mp_ncpus);
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (CPU_CMP(&top->cg_mask, &all_cpus))
panic("Built bad topology at %p. CPU mask (%s) != (%s)",
top, cpusetobj_strprint(cpusetbuf, &top->cg_mask),
cpusetobj_strprint(cpusetbuf2, &all_cpus));
/*
* Collapse nonsense levels that may be created out of convenience by
* the MD layers. They cause extra work in the search functions.
*/
while (top->cg_children == 1) {
top = &top->cg_child[0];
top->cg_parent = NULL;
}
return (top);
}
struct cpu_group *
smp_topo_alloc(u_int count)
{
static u_int index;
u_int curr;
curr = index;
index += count;
return (&group[curr]);
}
struct cpu_group *
smp_topo_none(void)
{
struct cpu_group *top;
top = &group[0];
top->cg_parent = NULL;
top->cg_child = NULL;
top->cg_mask = all_cpus;
top->cg_count = mp_ncpus;
top->cg_children = 0;
top->cg_level = CG_SHARE_NONE;
top->cg_flags = 0;
return (top);
}
static int
smp_topo_addleaf(struct cpu_group *parent, struct cpu_group *child, int share,
int count, int flags, int start)
{
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ];
cpuset_t mask;
int i;
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
CPU_ZERO(&mask);
for (i = 0; i < count; i++, start++)
CPU_SET(start, &mask);
child->cg_parent = parent;
child->cg_child = NULL;
child->cg_children = 0;
child->cg_level = share;
child->cg_count = count;
child->cg_flags = flags;
child->cg_mask = mask;
parent->cg_children++;
for (; parent != NULL; parent = parent->cg_parent) {
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (CPU_OVERLAP(&parent->cg_mask, &child->cg_mask))
panic("Duplicate children in %p. mask (%s) child (%s)",
parent,
cpusetobj_strprint(cpusetbuf, &parent->cg_mask),
cpusetobj_strprint(cpusetbuf2, &child->cg_mask));
CPU_OR(&parent->cg_mask, &child->cg_mask);
parent->cg_count += child->cg_count;
}
return (start);
}
struct cpu_group *
smp_topo_1level(int share, int count, int flags)
{
struct cpu_group *child;
struct cpu_group *top;
int packages;
int cpu;
int i;
cpu = 0;
top = &group[0];
packages = mp_ncpus / count;
top->cg_child = child = &group[1];
top->cg_level = CG_SHARE_NONE;
for (i = 0; i < packages; i++, child++)
cpu = smp_topo_addleaf(top, child, share, count, flags, cpu);
return (top);
}
struct cpu_group *
smp_topo_2level(int l2share, int l2count, int l1share, int l1count,
int l1flags)
{
struct cpu_group *top;
struct cpu_group *l1g;
struct cpu_group *l2g;
int cpu;
int i;
int j;
cpu = 0;
top = &group[0];
l2g = &group[1];
top->cg_child = l2g;
top->cg_level = CG_SHARE_NONE;
top->cg_children = mp_ncpus / (l2count * l1count);
l1g = l2g + top->cg_children;
for (i = 0; i < top->cg_children; i++, l2g++) {
l2g->cg_parent = top;
l2g->cg_child = l1g;
l2g->cg_level = l2share;
for (j = 0; j < l2count; j++, l1g++)
cpu = smp_topo_addleaf(l2g, l1g, l1share, l1count,
l1flags, cpu);
}
return (top);
}
struct cpu_group *
smp_topo_find(struct cpu_group *top, int cpu)
{
struct cpu_group *cg;
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
cpuset_t mask;
int children;
int i;
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
CPU_SETOF(cpu, &mask);
cg = top;
for (;;) {
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (!CPU_OVERLAP(&cg->cg_mask, &mask))
return (NULL);
if (cg->cg_children == 0)
return (cg);
children = cg->cg_children;
for (i = 0, cg = cg->cg_child; i < children; cg++, i++)
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
if (CPU_OVERLAP(&cg->cg_mask, &mask))
break;
}
return (NULL);
}
#else /* !SMP */
void
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
smp_rendezvous_cpus(cpuset_t map,
void (*setup_func)(void *),
void (*action_func)(void *),
void (*teardown_func)(void *),
void *arg)
{
/*
* In the !SMP case we just need to ensure the same initial conditions
* as the SMP case.
*/
spinlock_enter();
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
teardown_func(arg);
spinlock_exit();
}
void
2008-01-02 17:09:15 +00:00
smp_rendezvous(void (*setup_func)(void *),
void (*action_func)(void *),
void (*teardown_func)(void *),
void *arg)
{
smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func,
arg);
}
/*
* Provide dummy SMP support for UP kernels. Modules that need to use SMP
* APIs will still work using this dummy support.
*/
static void
mp_setvariables_for_up(void *dummy)
{
mp_ncpus = 1;
mp_maxid = PCPU_GET(cpuid);
CPU_SETOF(mp_maxid, &all_cpus);
KASSERT(PCPU_GET(cpuid) == 0, ("UP must have a CPU ID of zero"));
}
SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST,
mp_setvariables_for_up, NULL);
#endif /* SMP */
void
smp_no_rendezvous_barrier(void *dummy)
{
#ifdef SMP
KASSERT((!smp_started),("smp_no_rendezvous called and smp is started"));
#endif
}
/*
* Wait for specified idle threads to switch once. This ensures that even
* preempted threads have cycled through the switch function once,
* exiting their codepaths. This allows us to change global pointers
* with no other synchronization.
*/
int
quiesce_cpus(cpuset_t map, const char *wmesg, int prio)
{
struct pcpu *pcpu;
u_int gen[MAXCPU];
int error;
int cpu;
error = 0;
for (cpu = 0; cpu <= mp_maxid; cpu++) {
if (!CPU_ISSET(cpu, &map) || CPU_ABSENT(cpu))
continue;
pcpu = pcpu_find(cpu);
gen[cpu] = pcpu->pc_idlethread->td_generation;
}
for (cpu = 0; cpu <= mp_maxid; cpu++) {
if (!CPU_ISSET(cpu, &map) || CPU_ABSENT(cpu))
continue;
pcpu = pcpu_find(cpu);
thread_lock(curthread);
sched_bind(curthread, cpu);
thread_unlock(curthread);
while (gen[cpu] == pcpu->pc_idlethread->td_generation) {
error = tsleep(quiesce_cpus, prio, wmesg, 1);
if (error != EWOULDBLOCK)
goto out;
error = 0;
}
}
out:
thread_lock(curthread);
sched_unbind(curthread);
thread_unlock(curthread);
return (error);
}
int
quiesce_all_cpus(const char *wmesg, int prio)
{
return quiesce_cpus(all_cpus, wmesg, prio);
}
/* Extra care is taken with this sysctl because the data type is volatile */
static int
sysctl_kern_smp_active(SYSCTL_HANDLER_ARGS)
{
int error, active;
active = smp_started;
error = SYSCTL_OUT(req, &active, sizeof(active));
return (error);
}
#ifdef SMP
void
topo_init_node(struct topo_node *node)
{
bzero(node, sizeof(*node));
TAILQ_INIT(&node->children);
}
void
topo_init_root(struct topo_node *root)
{
topo_init_node(root);
root->type = TOPO_TYPE_SYSTEM;
}
/*
* Add a child node with the given ID under the given parent.
* Do nothing if there is already a child with that ID.
*/
struct topo_node *
topo_add_node_by_hwid(struct topo_node *parent, int hwid,
topo_node_type type, uintptr_t subtype)
{
struct topo_node *node;
TAILQ_FOREACH_REVERSE(node, &parent->children,
topo_children, siblings) {
if (node->hwid == hwid
&& node->type == type && node->subtype == subtype) {
return (node);
}
}
node = malloc(sizeof(*node), M_TOPO, M_WAITOK);
topo_init_node(node);
node->parent = parent;
node->hwid = hwid;
node->type = type;
node->subtype = subtype;
TAILQ_INSERT_TAIL(&parent->children, node, siblings);
parent->nchildren++;
return (node);
}
/*
* Find a child node with the given ID under the given parent.
*/
struct topo_node *
topo_find_node_by_hwid(struct topo_node *parent, int hwid,
topo_node_type type, uintptr_t subtype)
{
struct topo_node *node;
TAILQ_FOREACH(node, &parent->children, siblings) {
if (node->hwid == hwid
&& node->type == type && node->subtype == subtype) {
return (node);
}
}
return (NULL);
}
/*
* Given a node change the order of its parent's child nodes such
* that the node becomes the firt child while preserving the cyclic
* order of the children. In other words, the given node is promoted
* by rotation.
*/
void
topo_promote_child(struct topo_node *child)
{
struct topo_node *next;
struct topo_node *node;
struct topo_node *parent;
parent = child->parent;
next = TAILQ_NEXT(child, siblings);
TAILQ_REMOVE(&parent->children, child, siblings);
TAILQ_INSERT_HEAD(&parent->children, child, siblings);
while (next != NULL) {
node = next;
next = TAILQ_NEXT(node, siblings);
TAILQ_REMOVE(&parent->children, node, siblings);
TAILQ_INSERT_AFTER(&parent->children, child, node, siblings);
child = node;
}
}
/*
* Iterate to the next node in the depth-first search (traversal) of
* the topology tree.
*/
struct topo_node *
topo_next_node(struct topo_node *top, struct topo_node *node)
{
struct topo_node *next;
if ((next = TAILQ_FIRST(&node->children)) != NULL)
return (next);
if ((next = TAILQ_NEXT(node, siblings)) != NULL)
return (next);
while (node != top && (node = node->parent) != top)
if ((next = TAILQ_NEXT(node, siblings)) != NULL)
return (next);
return (NULL);
}
/*
* Iterate to the next node in the depth-first search of the topology tree,
* but without descending below the current node.
*/
struct topo_node *
topo_next_nonchild_node(struct topo_node *top, struct topo_node *node)
{
struct topo_node *next;
if ((next = TAILQ_NEXT(node, siblings)) != NULL)
return (next);
while (node != top && (node = node->parent) != top)
if ((next = TAILQ_NEXT(node, siblings)) != NULL)
return (next);
return (NULL);
}
/*
* Assign the given ID to the given topology node that represents a logical
* processor.
*/
void
topo_set_pu_id(struct topo_node *node, cpuid_t id)
{
KASSERT(node->type == TOPO_TYPE_PU,
("topo_set_pu_id: wrong node type: %u", node->type));
KASSERT(CPU_EMPTY(&node->cpuset) && node->cpu_count == 0,
("topo_set_pu_id: cpuset already not empty"));
node->id = id;
CPU_SET(id, &node->cpuset);
node->cpu_count = 1;
node->subtype = 1;
while ((node = node->parent) != NULL) {
KASSERT(!CPU_ISSET(id, &node->cpuset),
("logical ID %u is already set in node %p", id, node));
CPU_SET(id, &node->cpuset);
node->cpu_count++;
}
}
static struct topology_spec {
topo_node_type type;
bool match_subtype;
uintptr_t subtype;
} topology_level_table[TOPO_LEVEL_COUNT] = {
[TOPO_LEVEL_PKG] = { .type = TOPO_TYPE_PKG, },
[TOPO_LEVEL_GROUP] = { .type = TOPO_TYPE_GROUP, },
[TOPO_LEVEL_CACHEGROUP] = {
.type = TOPO_TYPE_CACHE,
.match_subtype = true,
.subtype = CG_SHARE_L3,
},
[TOPO_LEVEL_CORE] = { .type = TOPO_TYPE_CORE, },
[TOPO_LEVEL_THREAD] = { .type = TOPO_TYPE_PU, },
};
static bool
topo_analyze_table(struct topo_node *root, int all, enum topo_level level,
struct topo_analysis *results)
{
struct topology_spec *spec;
struct topo_node *node;
int count;
if (level >= TOPO_LEVEL_COUNT)
return (true);
spec = &topology_level_table[level];
count = 0;
node = topo_next_node(root, root);
while (node != NULL) {
if (node->type != spec->type ||
(spec->match_subtype && node->subtype != spec->subtype)) {
node = topo_next_node(root, node);
continue;
}
if (!all && CPU_EMPTY(&node->cpuset)) {
node = topo_next_nonchild_node(root, node);
continue;
}
count++;
if (!topo_analyze_table(node, all, level + 1, results))
return (false);
node = topo_next_nonchild_node(root, node);
}
/* No explicit subgroups is essentially one subgroup. */
if (count == 0) {
count = 1;
if (!topo_analyze_table(root, all, level + 1, results))
return (false);
}
if (results->entities[level] == -1)
results->entities[level] = count;
else if (results->entities[level] != count)
return (false);
return (true);
}
/*
* Check if the topology is uniform, that is, each package has the same number
* of cores in it and each core has the same number of threads (logical
* processors) in it. If so, calculate the number of packages, the number of
* groups per package, the number of cachegroups per group, and the number of
* logical processors per cachegroup. 'all' parameter tells whether to include
* administratively disabled logical processors into the analysis.
*/
int
topo_analyze(struct topo_node *topo_root, int all,
struct topo_analysis *results)
{
results->entities[TOPO_LEVEL_PKG] = -1;
results->entities[TOPO_LEVEL_CORE] = -1;
results->entities[TOPO_LEVEL_THREAD] = -1;
results->entities[TOPO_LEVEL_GROUP] = -1;
results->entities[TOPO_LEVEL_CACHEGROUP] = -1;
if (!topo_analyze_table(topo_root, all, TOPO_LEVEL_PKG, results))
return (0);
KASSERT(results->entities[TOPO_LEVEL_PKG] > 0,
("bug in topology or analysis"));
return (1);
}
#endif /* SMP */