freebsd-skq/sys/kern/subr_smp.c

725 lines
19 KiB
C
Raw Normal View History

/*-
* 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.
* 3. Neither the name of the author nor the names of any co-contributors
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
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* 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/mutex.h>
#include <sys/pcpu.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
2011-05-18 15:50:12 +00:00
volatile cpuset_t stopped_cpus;
volatile cpuset_t started_cpus;
cpuset_t hlt_cpus_mask;
cpuset_t logical_cpus_mask;
2002-03-19 21:25:46 +00:00
void (*cpustop_restartfunc)(void);
#endif
/* 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;
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
SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD, NULL, "Kernel SMP");
SYSCTL_UINT(_kern_smp, OID_AUTO, maxid, CTLFLAG_RD, &mp_maxid, 0,
"Max CPU ID.");
SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD, &mp_maxcpus, 0,
"Max number of CPUs that the system was compiled for.");
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_active = 0; /* are the APs allowed to run? */
SYSCTL_INT(_kern_smp, OID_AUTO, active, CTLFLAG_RW, &smp_active, 0,
"Number of Auxillary Processors (APs) that were successfully started");
int smp_disabled = 0; /* has smp been disabled? */
SYSCTL_INT(_kern_smp, OID_AUTO, disabled, CTLFLAG_RDTUN, &smp_disabled, 0,
"SMP has been disabled from the loader");
TUNABLE_INT("kern.smp.disabled", &smp_disabled);
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, &smp_cpus, 0,
"Number of CPUs online");
int smp_topology = 0; /* Which topology we're using. */
SYSCTL_INT(_kern_smp, OID_AUTO, topology, CTLFLAG_RD, &smp_topology, 0,
"Topology override setting; 0 is default provided by hardware.");
TUNABLE_INT("kern.smp.topology", &smp_topology);
#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[3];
static volatile int smp_rv_generation;
/*
* 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();
}
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;
all_cpus = PCPU_GET(cpumask);
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
*
*/
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;
KASSERT(
#if defined(__amd64__)
type == IPI_STOP || type == IPI_STOP_HARD || type == IPI_SUSPEND,
#else
type == IPI_STOP || type == IPI_STOP_HARD,
#endif
2009-08-13 17:09:45 +00:00
("%s: invalid stop type", __func__));
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
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);
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;
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
while (!CPU_SUBSET(&stopped_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++;
#ifdef DIAGNOSTIC
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 (i == 100000) {
printf("timeout stopping cpus\n");
break;
}
#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
}
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));
}
#if defined(__amd64__)
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
*/
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
restart_cpus(cpuset_t 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
#ifdef KTR
char cpusetbuf[CPUSETBUFSIZ];
#endif
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
CTR1(KTR_SMP, "restart_cpus(%s)", cpusetobj_strprint(cpusetbuf, &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
/* signal other cpus to restart */
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_COPY_STORE_REL(&map, &started_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
/* wait for each to clear its bit */
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
while (CPU_OVERLAP(&stopped_cpus, &map))
cpu_spinwait();
return 1;
}
/*
* 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*);
int generation;
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;
generation = smp_rv_generation;
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_rendevous_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);
/*
* 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.
*
* Note that the write by the last CPU to finish the action
* may become visible to different CPUs at different times.
* As a result, the CPU that initiated the rendezvous may
* exit the rendezvous and drop the lock allowing another
* rendezvous to be initiated on the same CPU or a different
* CPU. In that case the exit sentinel may be cleared before
* all CPUs have noticed causing those CPUs to hang forever.
* Workaround this by using a generation count to notice when
* this race occurs and to exit the rendezvous in that case.
*/
MPASS(generation == smp_rv_generation);
atomic_add_int(&smp_rv_waiters[2], 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
if (local_teardown_func != smp_no_rendevous_barrier) {
while (smp_rv_waiters[2] < smp_rv_ncpus &&
generation == smp_rv_generation)
cpu_spinwait();
if (local_teardown_func != NULL)
local_teardown_func(local_func_arg);
}
2008-01-02 17:09:15 +00:00
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;
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) {
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
teardown_func(arg);
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);
atomic_add_acq_int(&smp_rv_generation, 1);
/* 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;
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();
/*
* If the caller did not request an exit barrier to be enforced
* on each CPU, ensure that this CPU waits for all the other
* CPUs to finish the rendezvous.
*/
2008-01-02 17:09:15 +00:00
if (teardown_func == smp_no_rendevous_barrier)
while (atomic_load_acq_int(&smp_rv_waiters[2]) < 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];
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));
return (top);
}
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)
{
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
teardown_func(arg);
}
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)
{
if (setup_func != NULL)
setup_func(arg);
if (action_func != NULL)
action_func(arg);
if (teardown_func != NULL)
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);
all_cpus = PCPU_GET(cpumask);
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_rendevous_barrier(void *dummy)
{
#ifdef SMP
KASSERT((!smp_started),("smp_no_rendevous called and smp is started"));
#endif
}