anv: Add helpers for setting/checking device lost

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
This commit is contained in:
Jason Ekstrand
2018-03-13 11:50:33 -07:00
parent 319ff6f1ad
commit cd0960b430
4 changed files with 36 additions and 21 deletions

View File

@@ -1711,7 +1711,7 @@ VkResult anv_CreateDevice(
device->instance = physical_device->instance; device->instance = physical_device->instance;
device->chipset_id = physical_device->chipset_id; device->chipset_id = physical_device->chipset_id;
device->no_hw = physical_device->no_hw; device->no_hw = physical_device->no_hw;
device->lost = false; device->_lost = false;
if (pAllocator) if (pAllocator)
device->alloc = *pAllocator; device->alloc = *pAllocator;
@@ -2048,6 +2048,12 @@ void anv_GetDeviceQueue2(
*pQueue = NULL; *pQueue = NULL;
} }
void
anv_device_set_lost(struct anv_device *device, const char *msg, ...)
{
device->_lost = true;
}
VkResult VkResult
anv_device_query_status(struct anv_device *device) anv_device_query_status(struct anv_device *device)
{ {
@@ -2055,24 +2061,24 @@ anv_device_query_status(struct anv_device *device)
* for it. However, it doesn't hurt to check and it potentially lets us * for it. However, it doesn't hurt to check and it potentially lets us
* avoid an ioctl. * avoid an ioctl.
*/ */
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
uint32_t active, pending; uint32_t active, pending;
int ret = anv_gem_gpu_get_reset_stats(device, &active, &pending); int ret = anv_gem_gpu_get_reset_stats(device, &active, &pending);
if (ret == -1) { if (ret == -1) {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "get_reset_stats failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"get_reset_stats failed: %m"); "get_reset_stats failed: %m");
} }
if (active) { if (active) {
device->lost = true; anv_device_set_lost(device, "GPU hung on one of our command buffers");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"GPU hung on one of our command buffers"); "GPU hung on one of our command buffers");
} else if (pending) { } else if (pending) {
device->lost = true; anv_device_set_lost(device, "GPU hung with commands in-flight");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"GPU hung with commands in-flight"); "GPU hung with commands in-flight");
} }
@@ -2092,7 +2098,7 @@ anv_device_bo_busy(struct anv_device *device, struct anv_bo *bo)
return VK_NOT_READY; return VK_NOT_READY;
} else if (ret == -1) { } else if (ret == -1) {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "gem wait failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"gem wait failed: %m"); "gem wait failed: %m");
} }
@@ -2115,7 +2121,7 @@ anv_device_wait(struct anv_device *device, struct anv_bo *bo,
return VK_TIMEOUT; return VK_TIMEOUT;
} else if (ret == -1) { } else if (ret == -1) {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "gem wait failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"gem wait failed: %m"); "gem wait failed: %m");
} }
@@ -2132,7 +2138,7 @@ VkResult anv_DeviceWaitIdle(
VkDevice _device) VkDevice _device)
{ {
ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device, device, _device);
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
struct anv_batch batch; struct anv_batch batch;
@@ -2806,7 +2812,7 @@ VkResult anv_QueueBindSparse(
VkFence fence) VkFence fence)
{ {
ANV_FROM_HANDLE(anv_queue, queue, _queue); ANV_FROM_HANDLE(anv_queue, queue, _queue);
if (unlikely(queue->device->lost)) if (anv_device_is_lost(queue->device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
return vk_error(VK_ERROR_FEATURE_NOT_PRESENT); return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
@@ -2864,7 +2870,7 @@ VkResult anv_GetEventStatus(
ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_event, event, _event); ANV_FROM_HANDLE(anv_event, event, _event);
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
if (!device->info.has_llc) { if (!device->info.has_llc) {
@@ -3094,7 +3100,7 @@ VkResult anv_GetCalibratedTimestampsEXT(
&pTimestamps[d]); &pTimestamps[d]);
if (ret != 0) { if (ret != 0) {
device->lost = TRUE; anv_device_set_lost(device, "Failed to get a timestamp");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"Failed to read the TIMESTAMP register: %m"); "Failed to read the TIMESTAMP register: %m");
} }

View File

@@ -1028,7 +1028,7 @@ struct anv_device {
pthread_mutex_t mutex; pthread_mutex_t mutex;
pthread_cond_t queue_submit; pthread_cond_t queue_submit;
bool lost; bool _lost;
}; };
static inline struct anv_state_pool * static inline struct anv_state_pool *
@@ -1075,6 +1075,15 @@ anv_state_flush(struct anv_device *device, struct anv_state state)
void anv_device_init_blorp(struct anv_device *device); void anv_device_init_blorp(struct anv_device *device);
void anv_device_finish_blorp(struct anv_device *device); void anv_device_finish_blorp(struct anv_device *device);
void anv_device_set_lost(struct anv_device *device,
const char *msg, ...);
static inline bool
anv_device_is_lost(struct anv_device *device)
{
return unlikely(device->_lost);
}
VkResult anv_device_execbuf(struct anv_device *device, VkResult anv_device_execbuf(struct anv_device *device,
struct drm_i915_gem_execbuffer2 *execbuf, struct drm_i915_gem_execbuffer2 *execbuf,
struct anv_bo **execbuf_bos); struct anv_bo **execbuf_bos);

View File

@@ -42,7 +42,7 @@ anv_device_execbuf(struct anv_device *device,
int ret = device->no_hw ? 0 : anv_gem_execbuffer(device, execbuf); int ret = device->no_hw ? 0 : anv_gem_execbuffer(device, execbuf);
if (ret != 0) { if (ret != 0) {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "execbuf2 failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"execbuf2 failed: %m"); "execbuf2 failed: %m");
} }
@@ -245,7 +245,7 @@ out:
*/ */
result = vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, result = vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"vkQueueSubmit() failed"); "vkQueueSubmit() failed");
device->lost = true; anv_device_set_lost(device, "vkQueueSubmit() failed");
} }
pthread_mutex_unlock(&device->mutex); pthread_mutex_unlock(&device->mutex);
@@ -398,7 +398,7 @@ VkResult anv_GetFenceStatus(
ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_fence, fence, _fence); ANV_FROM_HANDLE(anv_fence, fence, _fence);
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
struct anv_fence_impl *impl = struct anv_fence_impl *impl =
@@ -438,7 +438,7 @@ VkResult anv_GetFenceStatus(
return VK_NOT_READY; return VK_NOT_READY;
} else { } else {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "drm_syncobj_wait failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"drm_syncobj_wait failed: %m"); "drm_syncobj_wait failed: %m");
} }
@@ -526,7 +526,7 @@ anv_wait_for_syncobj_fences(struct anv_device *device,
return VK_TIMEOUT; return VK_TIMEOUT;
} else { } else {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "drm_syncobj_wait failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"drm_syncobj_wait failed: %m"); "drm_syncobj_wait failed: %m");
} }
@@ -670,7 +670,7 @@ anv_wait_for_bo_fences(struct anv_device *device,
} }
done: done:
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
return result; return result;
@@ -760,7 +760,7 @@ VkResult anv_WaitForFences(
{ {
ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device, device, _device);
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
if (anv_all_fences_syncobj(fenceCount, pFences)) { if (anv_all_fences_syncobj(fenceCount, pFences)) {

View File

@@ -181,7 +181,7 @@ wait_for_available(struct anv_device *device,
continue; continue;
} else if (ret == -1) { } else if (ret == -1) {
/* We don't know the real error. */ /* We don't know the real error. */
device->lost = true; anv_device_set_lost(device, "gem wait failed: %m");
return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST, return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
"gem wait failed: %m"); "gem wait failed: %m");
} else { } else {
@@ -224,7 +224,7 @@ VkResult genX(GetQueryPoolResults)(
pool->type == VK_QUERY_TYPE_PIPELINE_STATISTICS || pool->type == VK_QUERY_TYPE_PIPELINE_STATISTICS ||
pool->type == VK_QUERY_TYPE_TIMESTAMP); pool->type == VK_QUERY_TYPE_TIMESTAMP);
if (unlikely(device->lost)) if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST; return VK_ERROR_DEVICE_LOST;
if (pData == NULL) if (pData == NULL)