radv: track and report if a logical device is lost

This currently covers two situations where it's obvious that
the GPU hung:

1) when wait-of-idle doesn't finish in a finite time
2) when a CS submission is cancelled by the kernel

There is still probably some other situations that aren't yet handled.

According to the Vulkan spec, some operations should return
VK_ERROR_DEVICE_LOST when the corresponding logical device is
known to be lost.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5878>
This commit is contained in:
Samuel Pitoiset
2020-07-13 13:59:48 +02:00
committed by Marge Bot
parent c9c53dade0
commit d26f62c667
3 changed files with 71 additions and 4 deletions

View File

@@ -2590,6 +2590,25 @@ static void radv_device_finish_border_color(struct radv_device *device)
}
}
VkResult
_radv_device_set_lost(struct radv_device *device,
const char *file, int line,
const char *msg, ...)
{
VkResult err;
va_list ap;
p_atomic_inc(&device->lost);
va_start(ap, msg);
err = __vk_errorv(device->physical_device->instance, device,
VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT,
VK_ERROR_DEVICE_LOST, file, line, msg, ap);
va_end(ap);
return err;
}
VkResult radv_CreateDevice(
VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo* pCreateInfo,
@@ -4503,7 +4522,7 @@ fail:
* VK_ERROR_DEVICE_LOST to ensure the clients do not attempt
* to submit the same job again to this device.
*/
result = VK_ERROR_DEVICE_LOST;
result = radv_device_set_lost(queue->device, "vkQueueSubmit() failed");
}
radv_free_temp_syncobjs(queue->device,
@@ -4724,6 +4743,9 @@ VkResult radv_QueueSubmit(
uint32_t fence_idx = 0;
bool flushed_caches = false;
if (radv_device_is_lost(queue->device))
return VK_ERROR_DEVICE_LOST;
if (fence != VK_NULL_HANDLE) {
for (uint32_t i = 0; i < submitCount; ++i)
if (radv_submit_has_effects(pSubmits + i))
@@ -4793,6 +4815,9 @@ VkResult radv_QueueWaitIdle(
{
RADV_FROM_HANDLE(radv_queue, queue, _queue);
if (radv_device_is_lost(queue->device))
return VK_ERROR_DEVICE_LOST;
pthread_mutex_lock(&queue->pending_mutex);
while (!list_is_empty(&queue->pending_submissions)) {
pthread_cond_wait(&queue->device->timeline_cond, &queue->pending_mutex);
@@ -4802,9 +4827,10 @@ VkResult radv_QueueWaitIdle(
if (!queue->device->ws->ctx_wait_idle(queue->hw_ctx,
radv_queue_family_to_ring(queue->queue_family_index),
queue->queue_idx)) {
return vk_errorf(queue->device->instance, VK_ERROR_DEVICE_LOST,
"Failed to wait for a '%s' queue to be idle. "
"GPU hang ?", radv_get_queue_family_name(queue));
return radv_device_set_lost(queue->device,
"Failed to wait for a '%s' queue "
"to be idle. GPU hang ?",
radv_get_queue_family_name(queue));
}
return VK_SUCCESS;
@@ -5471,6 +5497,9 @@ static bool radv_sparse_bind_has_effects(const VkBindSparseInfo *info)
VkResult result;
uint32_t fence_idx = 0;
if (radv_device_is_lost(queue->device))
return VK_ERROR_DEVICE_LOST;
if (fence != VK_NULL_HANDLE) {
for (uint32_t i = 0; i < bindInfoCount; ++i)
if (radv_sparse_bind_has_effects(pBindInfo + i))
@@ -5653,6 +5682,10 @@ VkResult radv_WaitForFences(
uint64_t timeout)
{
RADV_FROM_HANDLE(radv_device, device, _device);
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
timeout = radv_get_absolute_timeout(timeout);
if (device->always_use_syncobj &&
@@ -5809,6 +5842,9 @@ VkResult radv_GetFenceStatus(VkDevice _device, VkFence _fence)
fence->temporary.kind != RADV_FENCE_NONE ?
&fence->temporary : &fence->permanent;
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
switch (part->kind) {
case RADV_FENCE_NONE:
break;
@@ -6134,6 +6170,9 @@ radv_GetSemaphoreCounterValue(VkDevice _device,
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_semaphore, semaphore, _semaphore);
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
struct radv_semaphore_part *part =
semaphore->temporary.kind != RADV_SEMAPHORE_NONE ? &semaphore->temporary : &semaphore->permanent;
@@ -6191,6 +6230,10 @@ radv_WaitSemaphores(VkDevice _device,
uint64_t timeout)
{
RADV_FROM_HANDLE(radv_device, device, _device);
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
uint64_t abs_timeout = radv_get_absolute_timeout(timeout);
if (radv_semaphore_from_handle(pWaitInfo->pSemaphores[0])->permanent.kind == RADV_SEMAPHORE_TIMELINE)
@@ -6327,8 +6370,12 @@ VkResult radv_GetEventStatus(
VkDevice _device,
VkEvent _event)
{
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_event, event, _event);
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
if (*event->map == 1)
return VK_EVENT_SET;
return VK_EVENT_RESET;

View File

@@ -851,8 +851,25 @@ struct radv_device {
bool overallocation_disallowed;
uint64_t allocated_memory_size[VK_MAX_MEMORY_HEAPS];
mtx_t overallocation_mutex;
/* Track the number of device loss occurs. */
int lost;
};
VkResult _radv_device_set_lost(struct radv_device *device,
const char *file, int line,
const char *msg, ...)
radv_printflike(4, 5);
#define radv_device_set_lost(dev, ...) \
_radv_device_set_lost(dev, __FILE__, __LINE__, __VA_ARGS__)
static inline bool
radv_device_is_lost(const struct radv_device *device)
{
return unlikely(p_atomic_read(&device->lost));
}
struct radv_device_memory {
struct vk_object_base base;
struct radeon_winsys_bo *bo;

View File

@@ -1368,6 +1368,9 @@ VkResult radv_GetQueryPoolResults(
char *data = pData;
VkResult result = VK_SUCCESS;
if (radv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
for(unsigned i = 0; i < queryCount; ++i, data += stride) {
char *dest = data;
unsigned query = firstQuery + i;