In the reset_counter, we use two bits to track a GPU hang and reset. The
low bit is a "reset-in-progress" flag that we set to signal when we need
to break waiters in order for the recovery task to grab the mutex. As
soon as the recovery task has the mutex, we can clear that flag (which
we do by incrementing the reset_counter thereby incrementing the gobal
reset epoch). By clearing that flag when the recovery task holds the
struct_mutex, we can forgo a second flag that simply tells GEM to ignore
the "reset-in-progress" flag.
The second flag we store in the reset_counter is whether the
reset failed and we consider the GPU terminally wedged. Whilst this flag
is set, all access to the GPU (at least through GEM rather than direct mmio
access) is verboten.
PS: Fun is in store, as in the future we want to move from a global
reset epoch to a per-engine reset engine with request recovery.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1460565315-7748-6-git-send-email-chris@chris-wilson.co.uk
struct drm_device *dev = data;
struct drm_i915_private *dev_priv = dev->dev_private;
- *val = i915_reset_counter(&dev_priv->gpu_error);
+ *val = i915_terminally_wedged(&dev_priv->gpu_error);
return 0;
}
* while it is writing to 'i915_wedged'
*/
- if (i915_reset_in_progress_or_wedged(&dev_priv->gpu_error))
+ if (i915_reset_in_progress(&dev_priv->gpu_error))
return -EAGAIN;
intel_runtime_pm_get(dev_priv);
int i915_reset(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = dev->dev_private;
- bool simulated;
+ struct i915_gpu_error *error = &dev_priv->gpu_error;
+ unsigned reset_counter;
int ret;
intel_reset_gt_powersave(dev);
mutex_lock(&dev->struct_mutex);
- i915_gem_reset(dev);
+ /* Clear any previous failed attempts at recovery. Time to try again. */
+ atomic_andnot(I915_WEDGED, &error->reset_counter);
- simulated = dev_priv->gpu_error.stop_rings != 0;
+ /* Clear the reset-in-progress flag and increment the reset epoch. */
+ reset_counter = atomic_inc_return(&error->reset_counter);
+ if (WARN_ON(__i915_reset_in_progress(reset_counter))) {
+ ret = -EIO;
+ goto error;
+ }
+
+ i915_gem_reset(dev);
ret = intel_gpu_reset(dev, ALL_ENGINES);
/* Also reset the gpu hangman. */
- if (simulated) {
+ if (error->stop_rings != 0) {
DRM_INFO("Simulated gpu hang, resetting stop_rings\n");
- dev_priv->gpu_error.stop_rings = 0;
+ error->stop_rings = 0;
if (ret == -ENODEV) {
DRM_INFO("Reset not implemented, but ignoring "
"error for simulated gpu hangs\n");
if (ret) {
DRM_ERROR("Failed to reset chip: %i\n", ret);
- mutex_unlock(&dev->struct_mutex);
- return ret;
+ goto error;
}
intel_overlay_reset(dev_priv);
* was running at the time of the reset (i.e. we weren't VT
* switched away).
*/
-
- /* Used to prevent gem_check_wedged returning -EAGAIN during gpu reset */
- dev_priv->gpu_error.reload_in_reset = true;
-
ret = i915_gem_init_hw(dev);
-
- dev_priv->gpu_error.reload_in_reset = false;
-
- mutex_unlock(&dev->struct_mutex);
if (ret) {
DRM_ERROR("Failed hw init on reset %d\n", ret);
- return ret;
+ goto error;
}
+ mutex_unlock(&dev->struct_mutex);
+
/*
* rps/rc6 re-init is necessary to restore state lost after the
* reset and the re-install of gt irqs. Skip for ironlake per
intel_enable_gt_powersave(dev);
return 0;
+
+error:
+ atomic_or(I915_WEDGED, &error->reset_counter);
+ mutex_unlock(&dev->struct_mutex);
+ return ret;
}
static int i915_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
/* For missed irq/seqno simulation. */
unsigned int test_irq_rings;
-
- /* Used to prevent gem_check_wedged returning -EAGAIN during gpu reset */
- bool reload_in_reset;
};
enum modeset_restore {
{
int ret;
-#define EXIT_COND (!i915_reset_in_progress_or_wedged(error) || \
- i915_terminally_wedged(error))
- if (EXIT_COND)
+ if (!i915_reset_in_progress(error))
return 0;
/*
* we should simply try to bail out and fail as gracefully as possible.
*/
ret = wait_event_interruptible_timeout(error->reset_queue,
- EXIT_COND,
+ !i915_reset_in_progress(error),
10*HZ);
if (ret == 0) {
DRM_ERROR("Timed out waiting for the gpu reset to complete\n");
return -EIO;
} else if (ret < 0) {
return ret;
+ } else {
+ return 0;
}
-#undef EXIT_COND
-
- return 0;
}
int i915_mutex_lock_interruptible(struct drm_device *dev)
bool interruptible)
{
if (i915_reset_in_progress_or_wedged(error)) {
+ /* Recovery complete, but the reset failed ... */
+ if (i915_terminally_wedged(error))
+ return -EIO;
+
/* Non-interruptible callers can't handle -EAGAIN, hence return
* -EIO unconditionally for these. */
if (!interruptible)
return -EIO;
- /* Recovery complete, but the reset failed ... */
- if (i915_terminally_wedged(error))
- return -EIO;
-
- /*
- * Check if GPU Reset is in progress - we need intel_ring_begin
- * to work properly to reinit the hw state while the gpu is
- * still marked as reset-in-progress. Handle this with a flag.
- */
- if (!error->reload_in_reset)
- return -EAGAIN;
+ return -EAGAIN;
}
return 0;
static void i915_reset_and_wakeup(struct drm_device *dev)
{
struct drm_i915_private *dev_priv = to_i915(dev);
- struct i915_gpu_error *error = &dev_priv->gpu_error;
char *error_event[] = { I915_ERROR_UEVENT "=1", NULL };
char *reset_event[] = { I915_RESET_UEVENT "=1", NULL };
char *reset_done_event[] = { I915_ERROR_UEVENT "=0", NULL };
* the reset in-progress bit is only ever set by code outside of this
* work we don't need to worry about any other races.
*/
- if (i915_reset_in_progress_or_wedged(error) && !i915_terminally_wedged(error)) {
+ if (i915_reset_in_progress(&dev_priv->gpu_error)) {
DRM_DEBUG_DRIVER("resetting chip\n");
kobject_uevent_env(&dev->primary->kdev->kobj, KOBJ_CHANGE,
reset_event);
intel_runtime_pm_put(dev_priv);
- if (ret == 0) {
- /*
- * After all the gem state is reset, increment the reset
- * counter and wake up everyone waiting for the reset to
- * complete.
- *
- * Since unlock operations are a one-sided barrier only,
- * we need to insert a barrier here to order any seqno
- * updates before
- * the counter increment.
- */
- smp_mb__before_atomic();
- atomic_inc(&dev_priv->gpu_error.reset_counter);
-
+ if (ret == 0)
kobject_uevent_env(&dev->primary->kdev->kobj,
KOBJ_CHANGE, reset_done_event);
- } else {
- atomic_or(I915_WEDGED, &error->reset_counter);
- }
/*
* Note: The wake_up also serves as a memory barrier so that
projects / linux-beck.git / commitdiff