diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 338 |
1 files changed, 174 insertions, 164 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c index df9b173c3d0b..9cab36322c16 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c @@ -24,7 +24,7 @@ #include <linux/idr.h> #include <linux/dma-fence-array.h> -#include <drm/drmP.h> + #include "amdgpu.h" #include "amdgpu_trace.h" @@ -43,7 +43,7 @@ static DEFINE_IDA(amdgpu_pasid_ida); /* Helper to free pasid from a fence callback */ struct amdgpu_pasid_cb { struct dma_fence_cb cb; - unsigned int pasid; + u32 pasid; }; /** @@ -62,9 +62,8 @@ int amdgpu_pasid_alloc(unsigned int bits) int pasid = -EINVAL; for (bits = min(bits, 31U); bits > 0; bits--) { - pasid = ida_simple_get(&amdgpu_pasid_ida, - 1U << (bits - 1), 1U << bits, - GFP_KERNEL); + pasid = ida_alloc_range(&amdgpu_pasid_ida, 1U << (bits - 1), + (1U << bits) - 1, GFP_KERNEL); if (pasid != -ENOSPC) break; } @@ -79,10 +78,10 @@ int amdgpu_pasid_alloc(unsigned int bits) * amdgpu_pasid_free - Free a PASID * @pasid: PASID to free */ -void amdgpu_pasid_free(unsigned int pasid) +void amdgpu_pasid_free(u32 pasid) { trace_amdgpu_pasid_freed(pasid); - ida_simple_remove(&amdgpu_pasid_ida, pasid); + ida_free(&amdgpu_pasid_ida, pasid); } static void amdgpu_pasid_free_cb(struct dma_fence *fence, @@ -104,39 +103,22 @@ static void amdgpu_pasid_free_cb(struct dma_fence *fence, * * Free the pasid only after all the fences in resv are signaled. */ -void amdgpu_pasid_free_delayed(struct reservation_object *resv, - unsigned int pasid) +void amdgpu_pasid_free_delayed(struct dma_resv *resv, + u32 pasid) { - struct dma_fence *fence, **fences; struct amdgpu_pasid_cb *cb; - unsigned count; + struct dma_fence *fence; int r; - r = reservation_object_get_fences_rcu(resv, NULL, &count, &fences); + r = dma_resv_get_singleton(resv, DMA_RESV_USAGE_BOOKKEEP, &fence); if (r) goto fallback; - if (count == 0) { + if (!fence) { amdgpu_pasid_free(pasid); return; } - if (count == 1) { - fence = fences[0]; - kfree(fences); - } else { - uint64_t context = dma_fence_context_alloc(1); - struct dma_fence_array *array; - - array = dma_fence_array_create(count, fences, context, - 1, false); - if (!array) { - kfree(fences); - goto fallback; - } - fence = &array->base; - } - cb = kmalloc(sizeof(*cb), GFP_KERNEL); if (!cb) { /* Last resort when we are OOM */ @@ -156,8 +138,8 @@ fallback: /* Not enough memory for the delayed delete, as last resort * block for all the fences to complete. */ - reservation_object_wait_timeout_rcu(resv, true, false, - MAX_SCHEDULE_TIMEOUT); + dma_resv_wait_timeout(resv, DMA_RESV_USAGE_BOOKKEEP, + false, MAX_SCHEDULE_TIMEOUT); amdgpu_pasid_free(pasid); } @@ -182,146 +164,145 @@ bool amdgpu_vmid_had_gpu_reset(struct amdgpu_device *adev, atomic_read(&adev->gpu_reset_counter); } +/* Check if we need to switch to another set of resources */ +static bool amdgpu_vmid_gds_switch_needed(struct amdgpu_vmid *id, + struct amdgpu_job *job) +{ + return id->gds_base != job->gds_base || + id->gds_size != job->gds_size || + id->gws_base != job->gws_base || + id->gws_size != job->gws_size || + id->oa_base != job->oa_base || + id->oa_size != job->oa_size; +} + +/* Check if the id is compatible with the job */ +static bool amdgpu_vmid_compatible(struct amdgpu_vmid *id, + struct amdgpu_job *job) +{ + return id->pd_gpu_addr == job->vm_pd_addr && + !amdgpu_vmid_gds_switch_needed(id, job); +} + /** - * amdgpu_vm_grab_idle - grab idle VMID + * amdgpu_vmid_grab_idle - grab idle VMID * - * @vm: vm to allocate id for * @ring: ring we want to submit job to - * @sync: sync object where we add dependencies * @idle: resulting idle VMID + * @fence: fence to wait for if no id could be grabbed * * Try to find an idle VMID, if none is idle add a fence to wait to the sync * object. Returns -ENOMEM when we are out of memory. */ -static int amdgpu_vmid_grab_idle(struct amdgpu_vm *vm, - struct amdgpu_ring *ring, - struct amdgpu_sync *sync, - struct amdgpu_vmid **idle) +static int amdgpu_vmid_grab_idle(struct amdgpu_ring *ring, + struct amdgpu_vmid **idle, + struct dma_fence **fence) { struct amdgpu_device *adev = ring->adev; - unsigned vmhub = ring->funcs->vmhub; + unsigned vmhub = ring->vm_hub; struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; - struct dma_fence **fences; - unsigned i; - int r; - - if (ring->vmid_wait && !dma_fence_is_signaled(ring->vmid_wait)) - return amdgpu_sync_fence(adev, sync, ring->vmid_wait, false); - fences = kmalloc_array(sizeof(void *), id_mgr->num_ids, GFP_KERNEL); - if (!fences) - return -ENOMEM; + /* If anybody is waiting for a VMID let everybody wait for fairness */ + if (!dma_fence_is_signaled(ring->vmid_wait)) { + *fence = dma_fence_get(ring->vmid_wait); + return 0; + } /* Check if we have an idle VMID */ - i = 0; - list_for_each_entry((*idle), &id_mgr->ids_lru, list) { - fences[i] = amdgpu_sync_peek_fence(&(*idle)->active, ring); - if (!fences[i]) - break; - ++i; + list_for_each_entry_reverse((*idle), &id_mgr->ids_lru, list) { + /* Don't use per engine and per process VMID at the same time */ + struct amdgpu_ring *r = adev->vm_manager.concurrent_flush ? + NULL : ring; + + *fence = amdgpu_sync_peek_fence(&(*idle)->active, r); + if (!(*fence)) + return 0; } - /* If we can't find a idle VMID to use, wait till one becomes available */ - if (&(*idle)->list == &id_mgr->ids_lru) { - u64 fence_context = adev->vm_manager.fence_context + ring->idx; - unsigned seqno = ++adev->vm_manager.seqno[ring->idx]; - struct dma_fence_array *array; - unsigned j; - - *idle = NULL; - for (j = 0; j < i; ++j) - dma_fence_get(fences[j]); - - array = dma_fence_array_create(i, fences, fence_context, - seqno, true); - if (!array) { - for (j = 0; j < i; ++j) - dma_fence_put(fences[j]); - kfree(fences); - return -ENOMEM; - } - - r = amdgpu_sync_fence(adev, sync, &array->base, false); - dma_fence_put(ring->vmid_wait); - ring->vmid_wait = &array->base; - return r; - } - kfree(fences); + /* + * If we can't find a idle VMID to use, wait on a fence from the least + * recently used in the hope that it will be available soon. + */ + *idle = NULL; + dma_fence_put(ring->vmid_wait); + ring->vmid_wait = dma_fence_get(*fence); + /* This is the reference we return */ + dma_fence_get(*fence); return 0; } /** - * amdgpu_vm_grab_reserved - try to assign reserved VMID + * amdgpu_vmid_grab_reserved - try to assign reserved VMID * * @vm: vm to allocate id for * @ring: ring we want to submit job to - * @sync: sync object where we add dependencies - * @fence: fence protecting ID from reuse * @job: job who wants to use the VMID + * @id: resulting VMID + * @fence: fence to wait for if no id could be grabbed * * Try to assign a reserved VMID. */ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync, - struct dma_fence *fence, struct amdgpu_job *job, - struct amdgpu_vmid **id) + struct amdgpu_vmid **id, + struct dma_fence **fence) { struct amdgpu_device *adev = ring->adev; - unsigned vmhub = ring->funcs->vmhub; + unsigned vmhub = ring->vm_hub; uint64_t fence_context = adev->fence_context + ring->idx; - struct dma_fence *updates = sync->last_vm_update; bool needs_flush = vm->use_cpu_for_update; - int r = 0; + uint64_t updates = amdgpu_vm_tlb_seq(vm); + int r; *id = vm->reserved_vmid[vmhub]; - if (updates && (*id)->flushed_updates && - updates->context == (*id)->flushed_updates->context && - !dma_fence_is_later(updates, (*id)->flushed_updates)) - updates = NULL; - - if ((*id)->owner != vm->entity.fence_context || - job->vm_pd_addr != (*id)->pd_gpu_addr || - updates || !(*id)->last_flush || + if ((*id)->owner != vm->immediate.fence_context || + !amdgpu_vmid_compatible(*id, job) || + (*id)->flushed_updates < updates || + !(*id)->last_flush || ((*id)->last_flush->context != fence_context && - !dma_fence_is_signaled((*id)->last_flush))) { + !dma_fence_is_signaled((*id)->last_flush))) + needs_flush = true; + + if ((*id)->owner != vm->immediate.fence_context || + (!adev->vm_manager.concurrent_flush && needs_flush)) { struct dma_fence *tmp; + /* Don't use per engine and per process VMID at the + * same time + */ + if (adev->vm_manager.concurrent_flush) + ring = NULL; + /* to prevent one context starved by another context */ (*id)->pd_gpu_addr = 0; tmp = amdgpu_sync_peek_fence(&(*id)->active, ring); if (tmp) { *id = NULL; - r = amdgpu_sync_fence(adev, sync, tmp, false); - return r; + *fence = dma_fence_get(tmp); + return 0; } - needs_flush = true; } /* Good we can use this VMID. Remember this submission as * user of the VMID. */ - r = amdgpu_sync_fence(ring->adev, &(*id)->active, fence, false); + r = amdgpu_sync_fence(&(*id)->active, &job->base.s_fence->finished, + GFP_ATOMIC); if (r) return r; - if (updates) { - dma_fence_put((*id)->flushed_updates); - (*id)->flushed_updates = dma_fence_get(updates); - } job->vm_needs_flush = needs_flush; + job->spm_update_needed = true; return 0; } /** - * amdgpu_vm_grab_used - try to reuse a VMID + * amdgpu_vmid_grab_used - try to reuse a VMID * * @vm: vm to allocate id for * @ring: ring we want to submit job to - * @sync: sync object where we add dependencies - * @fence: fence protecting ID from reuse * @job: job who wants to use the VMID * @id: resulting VMID * @@ -329,16 +310,14 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm, */ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync, - struct dma_fence *fence, struct amdgpu_job *job, struct amdgpu_vmid **id) { struct amdgpu_device *adev = ring->adev; - unsigned vmhub = ring->funcs->vmhub; + unsigned vmhub = ring->vm_hub; struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; uint64_t fence_context = adev->fence_context + ring->idx; - struct dma_fence *updates = sync->last_vm_update; + uint64_t updates = amdgpu_vm_tlb_seq(vm); int r; job->vm_needs_flush = vm->use_cpu_for_update; @@ -346,13 +325,12 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, /* Check if we can use a VMID already assigned to this VM */ list_for_each_entry_reverse((*id), &id_mgr->ids_lru, list) { bool needs_flush = vm->use_cpu_for_update; - struct dma_fence *flushed; /* Check all the prerequisites to using this VMID */ - if ((*id)->owner != vm->entity.fence_context) + if ((*id)->owner != vm->immediate.fence_context) continue; - if ((*id)->pd_gpu_addr != job->vm_pd_addr) + if (!amdgpu_vmid_compatible(*id, job)) continue; if (!(*id)->last_flush || @@ -360,26 +338,21 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, !dma_fence_is_signaled((*id)->last_flush))) needs_flush = true; - flushed = (*id)->flushed_updates; - if (updates && (!flushed || dma_fence_is_later(updates, flushed))) + if ((*id)->flushed_updates < updates) needs_flush = true; - /* Concurrent flushes are only possible starting with Vega10 */ - if (adev->asic_type < CHIP_VEGA10 && needs_flush) + if (needs_flush && !adev->vm_manager.concurrent_flush) continue; /* Good, we can use this VMID. Remember this submission as * user of the VMID. */ - r = amdgpu_sync_fence(ring->adev, &(*id)->active, fence, false); + r = amdgpu_sync_fence(&(*id)->active, + &job->base.s_fence->finished, + GFP_ATOMIC); if (r) return r; - if (updates && (!flushed || dma_fence_is_later(updates, flushed))) { - dma_fence_put((*id)->flushed_updates); - (*id)->flushed_updates = dma_fence_get(updates); - } - job->vm_needs_flush |= needs_flush; return 0; } @@ -389,70 +362,74 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, } /** - * amdgpu_vm_grab_id - allocate the next free VMID + * amdgpu_vmid_grab - allocate the next free VMID * * @vm: vm to allocate id for * @ring: ring we want to submit job to - * @sync: sync object where we add dependencies - * @fence: fence protecting ID from reuse * @job: job who wants to use the VMID + * @fence: fence to wait for if no id could be grabbed * * Allocate an id for the vm, adding fences to the sync obj as necessary. */ int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync, struct dma_fence *fence, - struct amdgpu_job *job) + struct amdgpu_job *job, struct dma_fence **fence) { struct amdgpu_device *adev = ring->adev; - unsigned vmhub = ring->funcs->vmhub; + unsigned vmhub = ring->vm_hub; struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; struct amdgpu_vmid *idle = NULL; struct amdgpu_vmid *id = NULL; int r = 0; mutex_lock(&id_mgr->lock); - r = amdgpu_vmid_grab_idle(vm, ring, sync, &idle); + r = amdgpu_vmid_grab_idle(ring, &idle, fence); if (r || !idle) goto error; - if (vm->reserved_vmid[vmhub]) { - r = amdgpu_vmid_grab_reserved(vm, ring, sync, fence, job, &id); + if (amdgpu_vmid_uses_reserved(vm, vmhub)) { + r = amdgpu_vmid_grab_reserved(vm, ring, job, &id, fence); if (r || !id) goto error; } else { - r = amdgpu_vmid_grab_used(vm, ring, sync, fence, job, &id); + r = amdgpu_vmid_grab_used(vm, ring, job, &id); if (r) goto error; if (!id) { - struct dma_fence *updates = sync->last_vm_update; - /* Still no ID to use? Then use the idle one found earlier */ id = idle; /* Remember this submission as user of the VMID */ - r = amdgpu_sync_fence(ring->adev, &id->active, - fence, false); + r = amdgpu_sync_fence(&id->active, + &job->base.s_fence->finished, + GFP_ATOMIC); if (r) goto error; - dma_fence_put(id->flushed_updates); - id->flushed_updates = dma_fence_get(updates); job->vm_needs_flush = true; } list_move_tail(&id->list, &id_mgr->ids_lru); } - id->pd_gpu_addr = job->vm_pd_addr; - id->owner = vm->entity.fence_context; - + job->gds_switch_needed = amdgpu_vmid_gds_switch_needed(id, job); if (job->vm_needs_flush) { + id->flushed_updates = amdgpu_vm_tlb_seq(vm); dma_fence_put(id->last_flush); id->last_flush = NULL; } job->vmid = id - id_mgr->ids; job->pasid = vm->pasid; + + id->gds_base = job->gds_base; + id->gds_size = job->gds_size; + id->gws_base = job->gws_base; + id->gws_size = job->gws_size; + id->oa_base = job->oa_base; + id->oa_size = job->oa_size; + id->pd_gpu_addr = job->vm_pd_addr; + id->owner = vm->immediate.fence_context; + trace_amdgpu_vm_grab_id(vm, ring, job); error: @@ -460,29 +437,47 @@ error: return r; } -int amdgpu_vmid_alloc_reserved(struct amdgpu_device *adev, - struct amdgpu_vm *vm, +/* + * amdgpu_vmid_uses_reserved - check if a VM will use a reserved VMID + * @vm: the VM to check + * @vmhub: the VMHUB which will be used + * + * Returns: True if the VM will use a reserved VMID. + */ +bool amdgpu_vmid_uses_reserved(struct amdgpu_vm *vm, unsigned int vmhub) +{ + return vm->reserved_vmid[vmhub]; +} + +/* + * amdgpu_vmid_alloc_reserved - reserve a specific VMID for this vm + * @adev: amdgpu device structure + * @vm: the VM to reserve an ID for + * @vmhub: the VMHUB which should be used + * + * Mostly used to have a reserved VMID for debugging and SPM. + * + * Returns: 0 for success, -ENOENT if an ID is already reserved. + */ +int amdgpu_vmid_alloc_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned vmhub) { - struct amdgpu_vmid_mgr *id_mgr; - struct amdgpu_vmid *idle; + struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; + struct amdgpu_vmid *id; int r = 0; - id_mgr = &adev->vm_manager.id_mgr[vmhub]; mutex_lock(&id_mgr->lock); if (vm->reserved_vmid[vmhub]) goto unlock; - if (atomic_inc_return(&id_mgr->reserved_vmid_num) > - AMDGPU_VM_MAX_RESERVED_VMID) { - DRM_ERROR("Over limitation of reserved vmid\n"); - atomic_dec(&id_mgr->reserved_vmid_num); - r = -EINVAL; + if (id_mgr->reserved_vmid) { + r = -ENOENT; goto unlock; } - /* Select the first entry VMID */ - idle = list_first_entry(&id_mgr->ids_lru, struct amdgpu_vmid, list); - list_del_init(&idle->list); - vm->reserved_vmid[vmhub] = idle; + /* Remove from normal round robin handling */ + id = list_first_entry(&id_mgr->ids_lru, struct amdgpu_vmid, list); + list_del_init(&id->list); + vm->reserved_vmid[vmhub] = id; + id_mgr->reserved_vmid = true; mutex_unlock(&id_mgr->lock); return 0; @@ -491,8 +486,13 @@ unlock: return r; } -void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, - struct amdgpu_vm *vm, +/* + * amdgpu_vmid_free_reserved - free up a reserved VMID again + * @adev: amdgpu device structure + * @vm: the VM with the reserved ID + * @vmhub: the VMHUB which should be used + */ +void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned vmhub) { struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub]; @@ -502,7 +502,7 @@ void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, list_add(&vm->reserved_vmid[vmhub]->list, &id_mgr->ids_lru); vm->reserved_vmid[vmhub] = NULL; - atomic_dec(&id_mgr->reserved_vmid_num); + id_mgr->reserved_vmid = false; } mutex_unlock(&id_mgr->lock); } @@ -511,6 +511,7 @@ void amdgpu_vmid_free_reserved(struct amdgpu_device *adev, * amdgpu_vmid_reset - reset VMID to zero * * @adev: amdgpu device structure + * @vmhub: vmhub type * @vmid: vmid number to use * * Reset saved GDW, GWS and OA to force switch on next flush. @@ -569,7 +570,17 @@ void amdgpu_vmid_mgr_init(struct amdgpu_device *adev) mutex_init(&id_mgr->lock); INIT_LIST_HEAD(&id_mgr->ids_lru); - atomic_set(&id_mgr->reserved_vmid_num, 0); + + /* for GC <10, SDMA uses MMHUB so use first_kfd_vmid for both GC and MM */ + if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(10, 0, 0)) + /* manage only VMIDs not used by KFD */ + id_mgr->num_ids = adev->vm_manager.first_kfd_vmid; + else if (AMDGPU_IS_MMHUB0(i) || + AMDGPU_IS_MMHUB1(i)) + id_mgr->num_ids = 16; + else + /* manage only VMIDs not used by KFD */ + id_mgr->num_ids = adev->vm_manager.first_kfd_vmid; /* skip over VMID 0, since it is the system VM */ for (j = 1; j < id_mgr->num_ids; ++j) { @@ -600,7 +611,6 @@ void amdgpu_vmid_mgr_fini(struct amdgpu_device *adev) struct amdgpu_vmid *id = &id_mgr->ids[j]; amdgpu_sync_free(&id->active); - dma_fence_put(id->flushed_updates); dma_fence_put(id->last_flush); dma_fence_put(id->pasid_mapping); } |
