android_kernel_xiaomi_sm8350/drivers/char/drm/drm_lock.c

397 lines
11 KiB
C
Raw Normal View History

/**
* \file drm_lock.c
* IOCTLs for locking
*
* \author Rickard E. (Rik) Faith <faith@valinux.com>
* \author Gareth Hughes <gareth@valinux.com>
*/
/*
* Created: Tue Feb 2 08:37:54 1999 by faith@valinux.com
*
* Copyright 1999 Precision Insight, Inc., Cedar Park, Texas.
* Copyright 2000 VA Linux Systems, Inc., Sunnyvale, California.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#include "drmP.h"
static int drm_notifier(void *priv);
/**
* Lock ioctl.
*
* \param inode device inode.
* \param file_priv DRM file private.
* \param cmd command.
* \param arg user argument, pointing to a drm_lock structure.
* \return zero on success or negative number on failure.
*
* Add the current task to the lock wait queue, and attempt to take to lock.
*/
int drm_lock(struct drm_device *dev, void *data, struct drm_file *file_priv)
{
DECLARE_WAITQUEUE(entry, current);
struct drm_lock *lock = data;
int ret = 0;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
unsigned long irqflags;
++file_priv->lock_count;
if (lock->context == DRM_KERNEL_CONTEXT) {
DRM_ERROR("Process %d using kernel context %d\n",
task_pid_nr(current), lock->context);
return -EINVAL;
}
DRM_DEBUG("%d (pid %d) requests lock (0x%08x), flags = 0x%08x\n",
lock->context, task_pid_nr(current),
dev->lock.hw_lock->lock, lock->flags);
if (drm_core_check_feature(dev, DRIVER_DMA_QUEUE))
if (lock->context < 0)
return -EINVAL;
add_wait_queue(&dev->lock.lock_queue, &entry);
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&dev->lock.spinlock, irqflags);
dev->lock.user_waiters++;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&dev->lock.spinlock, irqflags);
for (;;) {
__set_current_state(TASK_INTERRUPTIBLE);
if (!dev->lock.hw_lock) {
/* Device has been unregistered */
ret = -EINTR;
break;
}
if (drm_lock_take(&dev->lock, lock->context)) {
dev->lock.file_priv = file_priv;
dev->lock.lock_time = jiffies;
atomic_inc(&dev->counts[_DRM_STAT_LOCKS]);
break; /* Got lock */
}
/* Contention */
schedule();
if (signal_pending(current)) {
ret = -ERESTARTSYS;
break;
}
}
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&dev->lock.spinlock, irqflags);
dev->lock.user_waiters--;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&dev->lock.spinlock, irqflags);
__set_current_state(TASK_RUNNING);
remove_wait_queue(&dev->lock.lock_queue, &entry);
DRM_DEBUG("%d %s\n", lock->context,
ret ? "interrupted" : "has lock");
if (ret) return ret;
sigemptyset(&dev->sigmask);
sigaddset(&dev->sigmask, SIGSTOP);
sigaddset(&dev->sigmask, SIGTSTP);
sigaddset(&dev->sigmask, SIGTTIN);
sigaddset(&dev->sigmask, SIGTTOU);
dev->sigdata.context = lock->context;
dev->sigdata.lock = dev->lock.hw_lock;
block_all_signals(drm_notifier, &dev->sigdata, &dev->sigmask);
if (dev->driver->dma_ready && (lock->flags & _DRM_LOCK_READY))
dev->driver->dma_ready(dev);
if (dev->driver->dma_quiescent && (lock->flags & _DRM_LOCK_QUIESCENT))
{
if (dev->driver->dma_quiescent(dev)) {
DRM_DEBUG("%d waiting for DMA quiescent\n",
lock->context);
return -EBUSY;
}
}
if (dev->driver->kernel_context_switch &&
dev->last_context != lock->context) {
dev->driver->kernel_context_switch(dev, dev->last_context,
lock->context);
}
return 0;
}
/**
* Unlock ioctl.
*
* \param inode device inode.
* \param file_priv DRM file private.
* \param cmd command.
* \param arg user argument, pointing to a drm_lock structure.
* \return zero on success or negative number on failure.
*
* Transfer and free the lock.
*/
int drm_unlock(struct drm_device *dev, void *data, struct drm_file *file_priv)
{
struct drm_lock *lock = data;
unsigned long irqflags;
if (lock->context == DRM_KERNEL_CONTEXT) {
DRM_ERROR("Process %d using kernel context %d\n",
task_pid_nr(current), lock->context);
return -EINVAL;
}
spin_lock_irqsave(&dev->tasklet_lock, irqflags);
if (dev->locked_tasklet_func) {
dev->locked_tasklet_func(dev);
dev->locked_tasklet_func = NULL;
}
spin_unlock_irqrestore(&dev->tasklet_lock, irqflags);
atomic_inc(&dev->counts[_DRM_STAT_UNLOCKS]);
/* kernel_context_switch isn't used by any of the x86 drm
* modules but is required by the Sparc driver.
*/
if (dev->driver->kernel_context_switch_unlock)
dev->driver->kernel_context_switch_unlock(dev);
else {
if (drm_lock_free(&dev->lock,lock->context)) {
/* FIXME: Should really bail out here. */
}
}
unblock_all_signals();
return 0;
}
/**
* Take the heavyweight lock.
*
* \param lock lock pointer.
* \param context locking context.
* \return one if the lock is held, or zero otherwise.
*
* Attempt to mark the lock as held by the given context, via the \p cmpxchg instruction.
*/
int drm_lock_take(struct drm_lock_data *lock_data,
unsigned int context)
{
unsigned int old, new, prev;
volatile unsigned int *lock = &lock_data->hw_lock->lock;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
unsigned long irqflags;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&lock_data->spinlock, irqflags);
do {
old = *lock;
if (old & _DRM_LOCK_HELD)
new = old | _DRM_LOCK_CONT;
else {
new = context | _DRM_LOCK_HELD |
((lock_data->user_waiters + lock_data->kernel_waiters > 1) ?
_DRM_LOCK_CONT : 0);
}
prev = cmpxchg(lock, old, new);
} while (prev != old);
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
if (_DRM_LOCKING_CONTEXT(old) == context) {
if (old & _DRM_LOCK_HELD) {
if (context != DRM_KERNEL_CONTEXT) {
DRM_ERROR("%d holds heavyweight lock\n",
context);
}
return 0;
}
}
if ((_DRM_LOCKING_CONTEXT(new)) == context && (new & _DRM_LOCK_HELD)) {
/* Have lock */
return 1;
}
return 0;
}
/**
* This takes a lock forcibly and hands it to context. Should ONLY be used
* inside *_unlock to give lock to kernel before calling *_dma_schedule.
*
* \param dev DRM device.
* \param lock lock pointer.
* \param context locking context.
* \return always one.
*
* Resets the lock file pointer.
* Marks the lock as held by the given context, via the \p cmpxchg instruction.
*/
static int drm_lock_transfer(struct drm_lock_data *lock_data,
unsigned int context)
{
unsigned int old, new, prev;
volatile unsigned int *lock = &lock_data->hw_lock->lock;
lock_data->file_priv = NULL;
do {
old = *lock;
new = context | _DRM_LOCK_HELD;
prev = cmpxchg(lock, old, new);
} while (prev != old);
return 1;
}
/**
* Free lock.
*
* \param dev DRM device.
* \param lock lock.
* \param context context.
*
* Resets the lock file pointer.
* Marks the lock as not held, via the \p cmpxchg instruction. Wakes any task
* waiting on the lock queue.
*/
int drm_lock_free(struct drm_lock_data *lock_data, unsigned int context)
{
unsigned int old, new, prev;
volatile unsigned int *lock = &lock_data->hw_lock->lock;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
unsigned long irqflags;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&lock_data->spinlock, irqflags);
if (lock_data->kernel_waiters != 0) {
drm_lock_transfer(lock_data, 0);
lock_data->idle_has_lock = 1;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
return 1;
}
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
do {
old = *lock;
new = _DRM_LOCKING_CONTEXT(old);
prev = cmpxchg(lock, old, new);
} while (prev != old);
if (_DRM_LOCK_IS_HELD(old) && _DRM_LOCKING_CONTEXT(old) != context) {
DRM_ERROR("%d freed heavyweight lock held by %d\n",
context, _DRM_LOCKING_CONTEXT(old));
return 1;
}
wake_up_interruptible(&lock_data->lock_queue);
return 0;
}
/**
* If we get here, it means that the process has called DRM_IOCTL_LOCK
* without calling DRM_IOCTL_UNLOCK.
*
* If the lock is not held, then let the signal proceed as usual. If the lock
* is held, then set the contended flag and keep the signal blocked.
*
* \param priv pointer to a drm_sigdata structure.
* \return one if the signal should be delivered normally, or zero if the
* signal should be blocked.
*/
static int drm_notifier(void *priv)
{
struct drm_sigdata *s = (struct drm_sigdata *) priv;
unsigned int old, new, prev;
/* Allow signal delivery if lock isn't held */
if (!s->lock || !_DRM_LOCK_IS_HELD(s->lock->lock)
|| _DRM_LOCKING_CONTEXT(s->lock->lock) != s->context)
return 1;
/* Otherwise, set flag to force call to
drmUnlock */
do {
old = s->lock->lock;
new = old | _DRM_LOCK_CONT;
prev = cmpxchg(&s->lock->lock, old, new);
} while (prev != old);
return 0;
}
/**
* This function returns immediately and takes the hw lock
* with the kernel context if it is free, otherwise it gets the highest priority when and if
* it is eventually released.
*
* This guarantees that the kernel will _eventually_ have the lock _unless_ it is held
* by a blocked process. (In the latter case an explicit wait for the hardware lock would cause
* a deadlock, which is why the "idlelock" was invented).
*
* This should be sufficient to wait for GPU idle without
* having to worry about starvation.
*/
void drm_idlelock_take(struct drm_lock_data *lock_data)
{
int ret = 0;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
unsigned long irqflags;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&lock_data->spinlock, irqflags);
lock_data->kernel_waiters++;
if (!lock_data->idle_has_lock) {
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
ret = drm_lock_take(lock_data, DRM_KERNEL_CONTEXT);
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&lock_data->spinlock, irqflags);
if (ret == 1)
lock_data->idle_has_lock = 1;
}
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
}
EXPORT_SYMBOL(drm_idlelock_take);
void drm_idlelock_release(struct drm_lock_data *lock_data)
{
unsigned int old, prev;
volatile unsigned int *lock = &lock_data->hw_lock->lock;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
unsigned long irqflags;
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_lock_irqsave(&lock_data->spinlock, irqflags);
if (--lock_data->kernel_waiters == 0) {
if (lock_data->idle_has_lock) {
do {
old = *lock;
prev = cmpxchg(lock, old, DRM_KERNEL_CONTEXT);
} while (prev != old);
wake_up_interruptible(&lock_data->lock_queue);
lock_data->idle_has_lock = 0;
}
}
drm: Fix race that can lockup the kernel The i915_vblank_swap() function schedules an automatic buffer swap upon receipt of the vertical sync interrupt. Such an operation is lengthy so it can't be allowed to happen in normal interrupt context, thus the DRM implements this by scheduling the work in a kernel softirq-scheduled tasklet. In order for the buffer swap to work safely, the DRM's central lock must be taken, via a call to drm_lock_take() located in drivers/char/drm/drm_irq.c within the function drm_locked_tasklet_func(). The lock-taking logic uses a non-interrupt-blocking spinlock to implement the manipulations needed to take the lock. This semantic would be safe if all attempts to use the spinlock only happen from process context. However this buffer swap happens from softirq context which is really a form of interrupt context. Thus we have an unsafe situation, in that drm_locked_tasklet_func() can block on a spinlock already taken by a thread in process context which will never get scheduled again because of the blocked softirq tasklet. This wedges the kernel hard. To trigger this bug, run a dual-head cloned mode configuration which uses the i915 drm, then execute an opengl application which synchronizes buffer swaps against the vertical sync interrupt. In my testing, a lockup always results after running anywhere from 5 minutes to an hour and a half. I believe dual-head is needed to really trigger the problem because then the vertical sync interrupt handling is no longer predictable (due to being interrupt-sourced from two different heads running at different speeds). This raises the probability of the tasklet trying to run while the userspace DRI is doing things to the GPU (and manipulating the DRM lock). The fix is to change the relevant spinlock semantics to be the interrupt-blocking form. After this change I am no longer able to trigger the lockup; the longest test run so far was 20 hours (test stopped after that point). Note: I have examined the places where this spinlock is being employed; all are reasonably short bounded sequences and should be suitable for interrupts being blocked without impacting overall kernel interrupt response latency. Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Dave Airlie <airlied@redhat.com>
2008-03-13 16:30:35 -04:00
spin_unlock_irqrestore(&lock_data->spinlock, irqflags);
}
EXPORT_SYMBOL(drm_idlelock_release);
int drm_i_have_hw_lock(struct drm_device *dev, struct drm_file *file_priv)
{
return (file_priv->lock_count && dev->lock.hw_lock &&
_DRM_LOCK_IS_HELD(dev->lock.hw_lock->lock) &&
dev->lock.file_priv == file_priv);
}
EXPORT_SYMBOL(drm_i_have_hw_lock);