static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
{
struct fence *fence = NULL;
struct amdgpu_job *job;
int r;
if (!sched_job) {
DRM_ERROR("job is null\n");
return NULL;
}
job = to_amdgpu_job(sched_job);
BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
trace_amdgpu_sched_run_job(job);
r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
job->sync.last_vm_update, job, &fence);
if (r) {
DRM_ERROR("Error scheduling IBs (%d)\n", r);
goto err;
}
err:
/* if gpu reset, hw fence will be replaced here */
fence_put(job->fence);
job->fence = fence;
return fence;
}
/**
* si_dma_ring_test_ib - test an IB on the DMA engine
*
* @ring: amdgpu_ring structure holding ring information
*
* Test a simple IB in the DMA ring (VI).
* Returns 0 on success, error on failure.
*/
static int si_dma_ring_test_ib(struct amdgpu_ring *ring, long timeout)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
struct dma_fence *f = NULL;
unsigned index;
u32 tmp = 0;
u64 gpu_addr;
long r;
r = amdgpu_wb_get(adev, &index);
if (r) {
dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r);
return r;
}
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
memset(&ib, 0, sizeof(ib));
r = amdgpu_ib_get(adev, NULL, 256, &ib);
if (r) {
DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
goto err0;
}
ib.ptr[0] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, 1);
ib.ptr[1] = lower_32_bits(gpu_addr);
ib.ptr[2] = upper_32_bits(gpu_addr) & 0xff;
ib.ptr[3] = 0xDEADBEEF;
ib.length_dw = 4;
r = amdgpu_ib_schedule(ring, 1, &ib, NULL, NULL, &f);
if (r)
goto err1;
r = dma_fence_wait_timeout(f, false, timeout);
if (r == 0) {
DRM_ERROR("amdgpu: IB test timed out\n");
r = -ETIMEDOUT;
goto err1;
} else if (r < 0) {
DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
goto err1;
}
tmp = le32_to_cpu(adev->wb.wb[index]);
if (tmp == 0xDEADBEEF) {
DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
r = 0;
} else {
DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
err1:
amdgpu_ib_free(adev, &ib, NULL);
dma_fence_put(f);
err0:
amdgpu_wb_free(adev, index);
return r;
}
static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job)
{
struct dma_fence *fence = NULL;
struct amdgpu_job *job;
int r;
if (!sched_job) {
DRM_ERROR("job is null\n");
return NULL;
}
job = to_amdgpu_job(sched_job);
BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL));
trace_amdgpu_sched_run_job(job);
r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence);
if (r)
DRM_ERROR("Error scheduling IBs (%d)\n", r);
/* if gpu reset, hw fence will be replaced here */
dma_fence_put(job->fence);
job->fence = dma_fence_get(fence);
amdgpu_job_free_resources(job);
return fence;
}
static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
{
struct amdgpu_fence *fence = NULL;
struct amdgpu_job *job;
int r;
if (!sched_job) {
DRM_ERROR("job is null\n");
return NULL;
}
job = to_amdgpu_job(sched_job);
mutex_lock(&job->job_lock);
r = amdgpu_ib_schedule(job->adev,
job->num_ibs,
job->ibs,
job->base.owner);
if (r) {
DRM_ERROR("Error scheduling IBs (%d)\n", r);
goto err;
}
fence = job->ibs[job->num_ibs - 1].fence;
fence_get(&fence->base);
err:
if (job->free_job)
job->free_job(job);
mutex_unlock(&job->job_lock);
fence_put(&job->base.s_fence->base);
kfree(job);
return fence ? &fence->base : NULL;
}
static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job)
{
struct fence *fence = NULL;
struct amdgpu_job *job;
int r;
if (!sched_job) {
DRM_ERROR("job is null\n");
return NULL;
}
job = to_amdgpu_job(sched_job);
r = amdgpu_sync_wait(&job->sync);
if (r) {
DRM_ERROR("failed to sync wait (%d)\n", r);
return NULL;
}
trace_amdgpu_sched_run_job(job);
r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs,
job->sync.last_vm_update, job, &fence);
if (r) {
DRM_ERROR("Error scheduling IBs (%d)\n", r);
goto err;
}
err:
job->fence = fence;
amdgpu_job_free(job);
return fence;
}
int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
uint32_t vmid, uint64_t gpu_addr,
uint32_t *ib_cmd, uint32_t ib_len)
{
struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
struct amdgpu_job *job;
struct amdgpu_ib *ib;
struct amdgpu_ring *ring;
struct dma_fence *f = NULL;
int ret;
switch (engine) {
case KGD_ENGINE_MEC1:
ring = &adev->gfx.compute_ring[0];
break;
case KGD_ENGINE_SDMA1:
ring = &adev->sdma.instance[0].ring;
break;
case KGD_ENGINE_SDMA2:
ring = &adev->sdma.instance[1].ring;
break;
default:
pr_err("Invalid engine in IB submission: %d\n", engine);
ret = -EINVAL;
goto err;
}
ret = amdgpu_job_alloc(adev, 1, &job, NULL);
if (ret)
goto err;
ib = &job->ibs[0];
memset(ib, 0, sizeof(struct amdgpu_ib));
ib->gpu_addr = gpu_addr;
ib->ptr = ib_cmd;
ib->length_dw = ib_len;
/* This works for NO_HWS. TODO: need to handle without knowing VMID */
job->vmid = vmid;
ret = amdgpu_ib_schedule(ring, 1, ib, job, &f);
if (ret) {
DRM_ERROR("amdgpu: failed to schedule IB.\n");
goto err_ib_sched;
}
ret = dma_fence_wait(f, false);
err_ib_sched:
dma_fence_put(f);
amdgpu_job_free(job);
err:
return ret;
}
int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
struct amdgpu_ring *ring,
struct amdgpu_ib *ibs,
unsigned num_ibs,
int (*free_job)(struct amdgpu_job *),
void *owner,
struct fence **f)
{
int r = 0;
if (amdgpu_enable_scheduler) {
struct amdgpu_job *job =
kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
if (!job)
return -ENOMEM;
job->base.sched = &ring->sched;
job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
job->adev = adev;
job->ibs = ibs;
job->num_ibs = num_ibs;
job->base.owner = owner;
mutex_init(&job->job_lock);
job->free_job = free_job;
mutex_lock(&job->job_lock);
r = amd_sched_entity_push_job(&job->base);
if (r) {
mutex_unlock(&job->job_lock);
kfree(job);
return r;
}
*f = fence_get(&job->base.s_fence->base);
mutex_unlock(&job->job_lock);
} else {
r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
if (r)
return r;
*f = fence_get(&ibs[num_ibs - 1].fence->base);
}
return 0;
}
/**
* cik_sdma_ring_test_ib - test an IB on the DMA engine
*
* @ring: amdgpu_ring structure holding ring information
*
* Test a simple IB in the DMA ring (CIK).
* Returns 0 on success, error on failure.
*/
static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
struct amdgpu_ib ib;
unsigned i;
unsigned index;
int r;
u32 tmp = 0;
u64 gpu_addr;
r = amdgpu_wb_get(adev, &index);
if (r) {
dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r);
return r;
}
gpu_addr = adev->wb.gpu_addr + (index * 4);
tmp = 0xCAFEDEAD;
adev->wb.wb[index] = cpu_to_le32(tmp);
r = amdgpu_ib_get(ring, NULL, 256, &ib);
if (r) {
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
return r;
}
ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0);
ib.ptr[1] = lower_32_bits(gpu_addr);
ib.ptr[2] = upper_32_bits(gpu_addr);
ib.ptr[3] = 1;
ib.ptr[4] = 0xDEADBEEF;
ib.length_dw = 5;
r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
return r;
}
r = amdgpu_fence_wait(ib.fence, false);
if (r) {
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
return r;
}
for (i = 0; i < adev->usec_timeout; i++) {
tmp = le32_to_cpu(adev->wb.wb[index]);
if (tmp == 0xDEADBEEF)
break;
DRM_UDELAY(1);
}
if (i < adev->usec_timeout) {
DRM_INFO("ib test on ring %d succeeded in %u usecs\n",
ib.fence->ring->idx, i);
} else {
DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
r = -EINVAL;
}
amdgpu_ib_free(adev, &ib);
amdgpu_wb_free(adev, index);
return r;
}
