freebsd-dev/sys/kern/kern_kcov.c
Mark Johnston fee2a2fa39 Change synchonization rules for vm_page reference counting.
There are several mechanisms by which a vm_page reference is held,
preventing the page from being freed back to the page allocator.  In
particular, holding the page's object lock is sufficient to prevent the
page from being freed; holding the busy lock or a wiring is sufficent as
well.  These references are protected by the page lock, which must
therefore be acquired for many per-page operations.  This results in
false sharing since the page locks are external to the vm_page
structures themselves and each lock protects multiple structures.

Transition to using an atomically updated per-page reference counter.
The object's reference is counted using a flag bit in the counter.  A
second flag bit is used to atomically block new references via
pmap_extract_and_hold() while removing managed mappings of a page.
Thus, the reference count of a page is guaranteed not to increase if the
page is unbusied, unmapped, and the object's write lock is held.  As
a consequence of this, the page lock no longer protects a page's
identity; operations which move pages between objects are now
synchronized solely by the objects' locks.

The vm_page_wire() and vm_page_unwire() KPIs are changed.  The former
requires that either the object lock or the busy lock is held.  The
latter no longer has a return value and may free the page if it releases
the last reference to that page.  vm_page_unwire_noq() behaves the same
as before; the caller is responsible for checking its return value and
freeing or enqueuing the page as appropriate.  vm_page_wire_mapped() is
introduced for use in pmap_extract_and_hold().  It fails if the page is
concurrently being unmapped, typically triggering a fallback to the
fault handler.  vm_page_wire() no longer requires the page lock and
vm_page_unwire() now internally acquires the page lock when releasing
the last wiring of a page (since the page lock still protects a page's
queue state).  In particular, synchronization details are no longer
leaked into the caller.

The change excises the page lock from several frequently executed code
paths.  In particular, vm_object_terminate() no longer bounces between
page locks as it releases an object's pages, and direct I/O and
sendfile(SF_NOCACHE) completions no longer require the page lock.  In
these latter cases we now get linear scalability in the common scenario
where different threads are operating on different files.

__FreeBSD_version is bumped.  The DRM ports have been updated to
accomodate the KPI changes.

Reviewed by:	jeff (earlier version)
Tested by:	gallatin (earlier version), pho
Sponsored by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D20486
2019-09-09 21:32:42 +00:00

579 lines
15 KiB
C

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (C) 2018 The FreeBSD Foundation. All rights reserved.
* Copyright (C) 2018, 2019 Andrew Turner
*
* This software was developed by Mitchell Horne under sponsorship of
* the FreeBSD Foundation.
*
* This software was developed by SRI International and the University of
* Cambridge Computer Laboratory under DARPA/AFRL contract FA8750-10-C-0237
* ("CTSRD"), as part of the DARPA CRASH research programme.
*
* 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.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/conf.h>
#include <sys/eventhandler.h>
#include <sys/kcov.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mman.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/rwlock.h>
#include <sys/sysctl.h>
#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_extern.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pager.h>
#include <vm/vm_param.h>
MALLOC_DEFINE(M_KCOV_INFO, "kcovinfo", "KCOV info type");
#define KCOV_ELEMENT_SIZE sizeof(uint64_t)
/*
* To know what the code can safely perform at any point in time we use a
* state machine. In the normal case the state transitions are:
*
* OPEN -> READY -> RUNNING -> DYING
* | | ^ | ^ ^
* | | +--------+ | |
* | +-------------------+ |
* +-----------------------------+
*
* The states are:
* OPEN: The kcov fd has been opened, but no buffer is available to store
* coverage data.
* READY: The buffer to store coverage data has been allocated. Userspace
* can set this by using ioctl(fd, KIOSETBUFSIZE, entries);. When
* this has been set the buffer can be written to by the kernel,
* and mmaped by userspace.
* RUNNING: The coverage probes are able to store coverage data in the buffer.
* This is entered with ioctl(fd, KIOENABLE, mode);. The READY state
* can be exited by ioctl(fd, KIODISABLE); or exiting the thread to
* return to the READY state to allow tracing to be reused, or by
* closing the kcov fd to enter the DYING state.
* DYING: The fd has been closed. All states can enter into this state when
* userspace closes the kcov fd.
*
* We need to be careful when moving into and out of the RUNNING state. As
* an interrupt may happen while this is happening the ordering of memory
* operations is important so struct kcov_info is valid for the tracing
* functions.
*
* When moving into the RUNNING state prior stores to struct kcov_info need
* to be observed before the state is set. This allows for interrupts that
* may call into one of the coverage functions to fire at any point while
* being enabled and see a consistent struct kcov_info.
*
* When moving out of the RUNNING state any later stores to struct kcov_info
* need to be observed after the state is set. As with entering this is to
* present a consistent struct kcov_info to interrupts.
*/
typedef enum {
KCOV_STATE_INVALID,
KCOV_STATE_OPEN, /* The device is open, but with no buffer */
KCOV_STATE_READY, /* The buffer has been allocated */
KCOV_STATE_RUNNING, /* Recording trace data */
KCOV_STATE_DYING, /* The fd was closed */
} kcov_state_t;
/*
* (l) Set while holding the kcov_lock mutex and not in the RUNNING state.
* (o) Only set once while in the OPEN state. Cleaned up while in the DYING
* state, and with no thread associated with the struct kcov_info.
* (s) Set atomically to enter or exit the RUNNING state, non-atomically
* otherwise. See above for a description of the other constraints while
* moving into or out of the RUNNING state.
*/
struct kcov_info {
struct thread *thread; /* (l) */
vm_object_t bufobj; /* (o) */
vm_offset_t kvaddr; /* (o) */
size_t entries; /* (o) */
size_t bufsize; /* (o) */
kcov_state_t state; /* (s) */
int mode; /* (l) */
};
/* Prototypes */
static d_open_t kcov_open;
static d_close_t kcov_close;
static d_mmap_single_t kcov_mmap_single;
static d_ioctl_t kcov_ioctl;
static int kcov_alloc(struct kcov_info *info, size_t entries);
static void kcov_free(struct kcov_info *info);
static void kcov_init(const void *unused);
static struct cdevsw kcov_cdevsw = {
.d_version = D_VERSION,
.d_open = kcov_open,
.d_close = kcov_close,
.d_mmap_single = kcov_mmap_single,
.d_ioctl = kcov_ioctl,
.d_name = "kcov",
};
SYSCTL_NODE(_kern, OID_AUTO, kcov, CTLFLAG_RW, 0, "Kernel coverage");
static u_int kcov_max_entries = KCOV_MAXENTRIES;
SYSCTL_UINT(_kern_kcov, OID_AUTO, max_entries, CTLFLAG_RW,
&kcov_max_entries, 0,
"Maximum number of entries in the kcov buffer");
static struct mtx kcov_lock;
static int active_count;
static struct kcov_info *
get_kinfo(struct thread *td)
{
struct kcov_info *info;
/* We might have a NULL thread when releasing the secondary CPUs */
if (td == NULL)
return (NULL);
/*
* We are in an interrupt, stop tracing as it is not explicitly
* part of a syscall.
*/
if (td->td_intr_nesting_level > 0 || td->td_intr_frame != NULL)
return (NULL);
/*
* If info is NULL or the state is not running we are not tracing.
*/
info = td->td_kcov_info;
if (info == NULL ||
atomic_load_acq_int(&info->state) != KCOV_STATE_RUNNING)
return (NULL);
return (info);
}
static void
trace_pc(uintptr_t ret)
{
struct thread *td;
struct kcov_info *info;
uint64_t *buf, index;
td = curthread;
info = get_kinfo(td);
if (info == NULL)
return;
/*
* Check we are in the PC-trace mode.
*/
if (info->mode != KCOV_MODE_TRACE_PC)
return;
KASSERT(info->kvaddr != 0,
("__sanitizer_cov_trace_pc: NULL buf while running"));
buf = (uint64_t *)info->kvaddr;
/* The first entry of the buffer holds the index */
index = buf[0];
if (index + 2 > info->entries)
return;
buf[index + 1] = ret;
buf[0] = index + 1;
}
static bool
trace_cmp(uint64_t type, uint64_t arg1, uint64_t arg2, uint64_t ret)
{
struct thread *td;
struct kcov_info *info;
uint64_t *buf, index;
td = curthread;
info = get_kinfo(td);
if (info == NULL)
return (false);
/*
* Check we are in the comparison-trace mode.
*/
if (info->mode != KCOV_MODE_TRACE_CMP)
return (false);
KASSERT(info->kvaddr != 0,
("__sanitizer_cov_trace_pc: NULL buf while running"));
buf = (uint64_t *)info->kvaddr;
/* The first entry of the buffer holds the index */
index = buf[0];
/* Check we have space to store all elements */
if (index * 4 + 4 + 1 > info->entries)
return (false);
while (1) {
buf[index * 4 + 1] = type;
buf[index * 4 + 2] = arg1;
buf[index * 4 + 3] = arg2;
buf[index * 4 + 4] = ret;
if (atomic_cmpset_64(&buf[0], index, index + 1))
break;
buf[0] = index;
}
return (true);
}
/*
* The fd is being closed, cleanup everything we can.
*/
static void
kcov_mmap_cleanup(void *arg)
{
struct kcov_info *info = arg;
struct thread *thread;
mtx_lock_spin(&kcov_lock);
/*
* Move to KCOV_STATE_DYING to stop adding new entries.
*
* If the thread is running we need to wait until thread exit to
* clean up as it may currently be adding a new entry. If this is
* the case being in KCOV_STATE_DYING will signal that the buffer
* needs to be cleaned up.
*/
atomic_store_int(&info->state, KCOV_STATE_DYING);
atomic_thread_fence_seq_cst();
thread = info->thread;
mtx_unlock_spin(&kcov_lock);
if (thread != NULL)
return;
/*
* We can safely clean up the info struct as it is in the
* KCOV_STATE_DYING state with no thread associated.
*
* The KCOV_STATE_DYING stops new threads from using it.
* The lack of a thread means nothing is currently using the buffers.
*/
kcov_free(info);
}
static int
kcov_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
{
struct kcov_info *info;
int error;
info = malloc(sizeof(struct kcov_info), M_KCOV_INFO, M_ZERO | M_WAITOK);
info->state = KCOV_STATE_OPEN;
info->thread = NULL;
info->mode = -1;
if ((error = devfs_set_cdevpriv(info, kcov_mmap_cleanup)) != 0)
kcov_mmap_cleanup(info);
return (error);
}
static int
kcov_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
{
struct kcov_info *info;
int error;
if ((error = devfs_get_cdevpriv((void **)&info)) != 0)
return (error);
KASSERT(info != NULL, ("kcov_close with no kcov_info structure"));
/* Trying to close, but haven't disabled */
if (info->state == KCOV_STATE_RUNNING)
return (EBUSY);
return (0);
}
static int
kcov_mmap_single(struct cdev *dev, vm_ooffset_t *offset, vm_size_t size,
struct vm_object **object, int nprot)
{
struct kcov_info *info;
int error;
if ((nprot & (PROT_EXEC | PROT_READ | PROT_WRITE)) !=
(PROT_READ | PROT_WRITE))
return (EINVAL);
if ((error = devfs_get_cdevpriv((void **)&info)) != 0)
return (error);
if (info->kvaddr == 0 || size / KCOV_ELEMENT_SIZE != info->entries)
return (EINVAL);
vm_object_reference(info->bufobj);
*offset = 0;
*object = info->bufobj;
return (0);
}
static int
kcov_alloc(struct kcov_info *info, size_t entries)
{
size_t n, pages;
vm_page_t m;
KASSERT(info->kvaddr == 0, ("kcov_alloc: Already have a buffer"));
KASSERT(info->state == KCOV_STATE_OPEN,
("kcov_alloc: Not in open state (%x)", info->state));
if (entries < 2 || entries > kcov_max_entries)
return (EINVAL);
/* Align to page size so mmap can't access other kernel memory */
info->bufsize = roundup2(entries * KCOV_ELEMENT_SIZE, PAGE_SIZE);
pages = info->bufsize / PAGE_SIZE;
if ((info->kvaddr = kva_alloc(info->bufsize)) == 0)
return (ENOMEM);
info->bufobj = vm_pager_allocate(OBJT_PHYS, 0, info->bufsize,
PROT_READ | PROT_WRITE, 0, curthread->td_ucred);
VM_OBJECT_WLOCK(info->bufobj);
for (n = 0; n < pages; n++) {
m = vm_page_grab(info->bufobj, n,
VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_WIRED);
m->valid = VM_PAGE_BITS_ALL;
pmap_qenter(info->kvaddr + n * PAGE_SIZE, &m, 1);
}
VM_OBJECT_WUNLOCK(info->bufobj);
info->entries = entries;
return (0);
}
static void
kcov_free(struct kcov_info *info)
{
vm_page_t m;
size_t i;
if (info->kvaddr != 0) {
pmap_qremove(info->kvaddr, info->bufsize / PAGE_SIZE);
kva_free(info->kvaddr, info->bufsize);
}
if (info->bufobj != NULL) {
VM_OBJECT_WLOCK(info->bufobj);
m = vm_page_lookup(info->bufobj, 0);
for (i = 0; i < info->bufsize / PAGE_SIZE; i++) {
vm_page_unwire_noq(m);
m = vm_page_next(m);
}
VM_OBJECT_WUNLOCK(info->bufobj);
vm_object_deallocate(info->bufobj);
}
free(info, M_KCOV_INFO);
}
static int
kcov_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag __unused,
struct thread *td)
{
struct kcov_info *info;
int mode, error;
if ((error = devfs_get_cdevpriv((void **)&info)) != 0)
return (error);
if (cmd == KIOSETBUFSIZE) {
/*
* Set the size of the coverage buffer. Should be called
* before enabling coverage collection for that thread.
*/
if (info->state != KCOV_STATE_OPEN) {
return (EBUSY);
}
error = kcov_alloc(info, *(u_int *)data);
if (error == 0)
info->state = KCOV_STATE_READY;
return (error);
}
mtx_lock_spin(&kcov_lock);
switch (cmd) {
case KIOENABLE:
if (info->state != KCOV_STATE_READY) {
error = EBUSY;
break;
}
if (td->td_kcov_info != NULL) {
error = EINVAL;
break;
}
mode = *(int *)data;
if (mode != KCOV_MODE_TRACE_PC && mode != KCOV_MODE_TRACE_CMP) {
error = EINVAL;
break;
}
/* Lets hope nobody opens this 2 billion times */
KASSERT(active_count < INT_MAX,
("%s: Open too many times", __func__));
active_count++;
if (active_count == 1) {
cov_register_pc(&trace_pc);
cov_register_cmp(&trace_cmp);
}
KASSERT(info->thread == NULL,
("Enabling kcov when already enabled"));
info->thread = td;
info->mode = mode;
/*
* Ensure the mode has been set before starting coverage
* tracing.
*/
atomic_store_rel_int(&info->state, KCOV_STATE_RUNNING);
td->td_kcov_info = info;
break;
case KIODISABLE:
/* Only the currently enabled thread may disable itself */
if (info->state != KCOV_STATE_RUNNING ||
info != td->td_kcov_info) {
error = EINVAL;
break;
}
KASSERT(active_count > 0, ("%s: Open count is zero", __func__));
active_count--;
if (active_count == 0) {
cov_unregister_pc();
cov_unregister_cmp();
}
td->td_kcov_info = NULL;
atomic_store_int(&info->state, KCOV_STATE_READY);
/*
* Ensure we have exited the READY state before clearing the
* rest of the info struct.
*/
atomic_thread_fence_rel();
info->mode = -1;
info->thread = NULL;
break;
default:
error = EINVAL;
break;
}
mtx_unlock_spin(&kcov_lock);
return (error);
}
static void
kcov_thread_dtor(void *arg __unused, struct thread *td)
{
struct kcov_info *info;
info = td->td_kcov_info;
if (info == NULL)
return;
mtx_lock_spin(&kcov_lock);
KASSERT(active_count > 0, ("%s: Open count is zero", __func__));
active_count--;
if (active_count == 0) {
cov_unregister_pc();
cov_unregister_cmp();
}
td->td_kcov_info = NULL;
if (info->state != KCOV_STATE_DYING) {
/*
* The kcov file is still open. Mark it as unused and
* wait for it to be closed before cleaning up.
*/
atomic_store_int(&info->state, KCOV_STATE_READY);
atomic_thread_fence_seq_cst();
/* This info struct is unused */
info->thread = NULL;
mtx_unlock_spin(&kcov_lock);
return;
}
mtx_unlock_spin(&kcov_lock);
/*
* We can safely clean up the info struct as it is in the
* KCOV_STATE_DYING state where the info struct is associated with
* the current thread that's about to exit.
*
* The KCOV_STATE_DYING stops new threads from using it.
* It also stops the current thread from trying to use the info struct.
*/
kcov_free(info);
}
static void
kcov_init(const void *unused)
{
struct make_dev_args args;
struct cdev *dev;
mtx_init(&kcov_lock, "kcov lock", NULL, MTX_SPIN);
make_dev_args_init(&args);
args.mda_devsw = &kcov_cdevsw;
args.mda_uid = UID_ROOT;
args.mda_gid = GID_WHEEL;
args.mda_mode = 0600;
if (make_dev_s(&args, &dev, "kcov") != 0) {
printf("%s", "Failed to create kcov device");
return;
}
EVENTHANDLER_REGISTER(thread_dtor, kcov_thread_dtor, NULL,
EVENTHANDLER_PRI_ANY);
}
SYSINIT(kcovdev, SI_SUB_LAST, SI_ORDER_ANY, kcov_init, NULL);