Use HMM helper function hmm_vma_fault() to get physical pages backing
userptr and start CPU page table update track of those pages. Then use
hmm_vma_range_done() to check if those pages are updated before
amdgpu_cs_submit for gfx or before user queues are resumed for kfd.
If userptr pages are updated, for gfx, amdgpu_cs_ioctl will restart
from scratch, for kfd, restore worker is rescheduled to retry.
HMM simplify the CPU page table concurrent update check, so remove
guptasklock, mmu_invalidations, last_set_pages fields from
amdgpu_ttm_tt struct.
HMM does not pin the page (increase page ref count), so remove related
operations like release_pages(), put_page(), mark_page_dirty().
Change-Id: I2e8c0c6f0d8c21e5596a32d7fc91762778bc9e67
Signed-off-by: Philip Yang <Philip.Yang@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 1 -
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 95 +++------
drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 2 -
drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h | 3 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 185 +++++++++---------
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 14 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c | 27 ++-
drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 166 ++++++----------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 1 -
11 files changed, 208 insertions(+), 294 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 70429f7aa9a8..717791d4fa45 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -62,7 +62,6 @@ struct kgd_mem {
atomic_t invalid;
struct amdkfd_process_info *process_info;
- struct page **user_pages;
struct amdgpu_sync sync;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index be1ab43473c6..2897793600f7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -582,28 +582,12 @@ static int init_user_pages(struct kgd_mem *mem, struct mm_struct *mm,
goto out;
}
- /* If no restore worker is running concurrently, user_pages
- * should not be allocated
- */
- WARN(mem->user_pages, "Leaking user_pages array");
-
- mem->user_pages = kvmalloc_array(bo->tbo.ttm->num_pages,
- sizeof(struct page *),
- GFP_KERNEL | __GFP_ZERO);
- if (!mem->user_pages) {
- pr_err("%s: Failed to allocate pages array\n", __func__);
- ret = -ENOMEM;
- goto unregister_out;
- }
-
- ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, mem->user_pages);
+ ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, bo->tbo.ttm->pages);
if (ret) {
pr_err("%s: Failed to get user pages: %d\n", __func__, ret);
- goto free_out;
+ goto unregister_out;
}
- amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, mem->user_pages);
-
ret = amdgpu_bo_reserve(bo, true);
if (ret) {
pr_err("%s: Failed to reserve BO\n", __func__);
@@ -616,11 +600,7 @@ static int init_user_pages(struct kgd_mem *mem, struct mm_struct *mm,
amdgpu_bo_unreserve(bo);
release_out:
- if (ret)
- release_pages(mem->user_pages, bo->tbo.ttm->num_pages);
-free_out:
- kvfree(mem->user_pages);
- mem->user_pages = NULL;
+ amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm);
unregister_out:
if (ret)
amdgpu_mn_unregister(bo);
@@ -679,7 +659,6 @@ static int reserve_bo_and_vm(struct kgd_mem *mem,
ctx->kfd_bo.priority = 0;
ctx->kfd_bo.tv.bo = &bo->tbo;
ctx->kfd_bo.tv.num_shared = 1;
- ctx->kfd_bo.user_pages = NULL;
list_add(&ctx->kfd_bo.tv.head, &ctx->list);
amdgpu_vm_get_pd_bo(vm, &ctx->list, &ctx->vm_pd[0]);
@@ -743,7 +722,6 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
ctx->kfd_bo.priority = 0;
ctx->kfd_bo.tv.bo = &bo->tbo;
ctx->kfd_bo.tv.num_shared = 1;
- ctx->kfd_bo.user_pages = NULL;
list_add(&ctx->kfd_bo.tv.head, &ctx->list);
i = 0;
@@ -1371,15 +1349,6 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
list_del(&bo_list_entry->head);
mutex_unlock(&process_info->lock);
- /* Free user pages if necessary */
- if (mem->user_pages) {
- pr_debug("%s: Freeing user_pages array\n", __func__);
- if (mem->user_pages[0])
- release_pages(mem->user_pages,
- mem->bo->tbo.ttm->num_pages);
- kvfree(mem->user_pages);
- }
-
ret = reserve_bo_and_cond_vms(mem, NULL, BO_VM_ALL, &ctx);
if (unlikely(ret))
return ret;
@@ -1855,25 +1824,11 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
bo = mem->bo;
- if (!mem->user_pages) {
- mem->user_pages =
- kvmalloc_array(bo->tbo.ttm->num_pages,
- sizeof(struct page *),
- GFP_KERNEL | __GFP_ZERO);
- if (!mem->user_pages) {
- pr_err("%s: Failed to allocate pages array\n",
- __func__);
- return -ENOMEM;
- }
- } else if (mem->user_pages[0]) {
- release_pages(mem->user_pages, bo->tbo.ttm->num_pages);
- }
-
/* Get updated user pages */
ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm,
- mem->user_pages);
+ bo->tbo.ttm->pages);
if (ret) {
- mem->user_pages[0] = NULL;
+ bo->tbo.ttm->pages[0] = NULL;
pr_info("%s: Failed to get user pages: %d\n",
__func__, ret);
/* Pretend it succeeded. It will fail later
@@ -1882,12 +1837,6 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info,
* stalled user mode queues.
*/
}
-
- /* Mark the BO as valid unless it was invalidated
- * again concurrently
- */
- if (atomic_cmpxchg(&mem->invalid, invalid, 0) != invalid)
- return -EAGAIN;
}
return 0;
@@ -1917,7 +1866,8 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
GFP_KERNEL);
if (!pd_bo_list_entries) {
pr_err("%s: Failed to allocate PD BO list entries\n", __func__);
- return -ENOMEM;
+ ret = -ENOMEM;
+ goto out_no_mem;
}
INIT_LIST_HEAD(&resv_list);
@@ -1941,7 +1891,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
ret = ttm_eu_reserve_buffers(&ticket, &resv_list, false, &duplicates);
WARN(!list_empty(&duplicates), "Duplicates should be empty");
if (ret)
- goto out;
+ goto out_free;
amdgpu_sync_create(&sync);
@@ -1967,10 +1917,8 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
bo = mem->bo;
- /* Copy pages array and validate the BO if we got user pages */
- if (mem->user_pages[0]) {
- amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm,
- mem->user_pages);
+ /* Validate the BO if we got user pages */
+ if (bo->tbo.ttm->pages[0]) {
amdgpu_bo_placement_from_domain(bo, mem->domain);
ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
if (ret) {
@@ -1979,16 +1927,16 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
}
}
- /* Validate succeeded, now the BO owns the pages, free
- * our copy of the pointer array. Put this BO back on
- * the userptr_valid_list. If we need to revalidate
- * it, we need to start from scratch.
- */
- kvfree(mem->user_pages);
- mem->user_pages = NULL;
list_move_tail(&mem->validate_list.head,
&process_info->userptr_valid_list);
+ /* Stop HMM track the userptr update. We dont check the return
+ * value for concurrent CPU page table update because we will
+ * reschedule the restore worker if process_info->evicted_bos
+ * is updated.
+ */
+ amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm);
+
/* Update mapping. If the BO was not validated
* (because we couldn't get user pages), this will
* clear the page table entries, which will result in
@@ -2022,8 +1970,15 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
ttm_eu_backoff_reservation(&ticket, &resv_list);
amdgpu_sync_wait(&sync, false);
amdgpu_sync_free(&sync);
-out:
+out_free:
kfree(pd_bo_list_entries);
+out_no_mem:
+ list_for_each_entry_safe(mem, tmp_mem,
+ &process_info->userptr_inval_list,
+ validate_list.head) {
+ bo = mem->bo;
+ amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm);
+ }
return ret;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 5c79da8e1150..3842183970cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -200,8 +200,6 @@ void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list,
if (!bo->parent)
list_add_tail(&e->tv.head, &bucket[priority]);
-
- e->user_pages = NULL;
}
/* Connect the sorted buckets in the output list. */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
index 7c5f5d1601e6..4beab2de09b3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h
@@ -35,8 +35,7 @@ struct amdgpu_bo_list_entry {
struct ttm_validate_buffer tv;
struct amdgpu_bo_va *bo_va;
uint32_t priority;
- struct page **user_pages;
- int user_invalidated;
+ bool user_invalidated;
};
struct amdgpu_bo_list {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 1c49b8266d69..e30e39fb7224 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -52,7 +52,6 @@ static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
p->uf_entry.tv.bo = &bo->tbo;
/* One for TTM and one for the CS job */
p->uf_entry.tv.num_shared = 2;
- p->uf_entry.user_pages = NULL;
drm_gem_object_put_unlocked(gobj);
@@ -532,24 +531,19 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
list_for_each_entry(lobj, validated, tv.head) {
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(lobj->tv.bo);
- bool binding_userptr = false;
struct mm_struct *usermm;
usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
if (usermm && usermm != current->mm)
return -EPERM;
- /* Check if we have user pages and nobody bound the BO already */
- if (amdgpu_ttm_tt_userptr_needs_pages(bo->tbo.ttm) &&
- lobj->user_pages) {
+ if (amdgpu_ttm_tt_is_userptr(bo->tbo.ttm) &&
+ lobj->user_invalidated) {
amdgpu_bo_placement_from_domain(bo,
AMDGPU_GEM_DOMAIN_CPU);
r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
if (r)
return r;
- amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm,
- lobj->user_pages);
- binding_userptr = true;
}
if (p->evictable == lobj)
@@ -558,11 +552,6 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
r = amdgpu_cs_validate(p, bo);
if (r)
return r;
-
- if (binding_userptr) {
- kvfree(lobj->user_pages);
- lobj->user_pages = NULL;
- }
}
return 0;
}
@@ -577,7 +566,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
struct amdgpu_bo *gds;
struct amdgpu_bo *gws;
struct amdgpu_bo *oa;
- unsigned tries = 10;
+ struct list_head userpage_list;
int r;
INIT_LIST_HEAD(&p->validated);
@@ -586,7 +575,6 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
if (cs->in.bo_list_handle) {
if (p->bo_list)
return -EINVAL;
-
r = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle,
&p->bo_list);
if (r)
@@ -613,79 +601,72 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
if (p->uf_entry.tv.bo && !ttm_to_amdgpu_bo(p->uf_entry.tv.bo)->parent)
list_add(&p->uf_entry.tv.head, &p->validated);
- while (1) {
- struct list_head need_pages;
+ r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
+ &duplicates);
+ if (unlikely(r != 0)) {
+ if (r != -ERESTARTSYS)
+ DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
+ goto out;
+ }
- r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
- &duplicates);
- if (unlikely(r != 0)) {
- if (r != -ERESTARTSYS)
- DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
- goto error_free_pages;
- }
+ INIT_LIST_HEAD(&userpage_list);
+ amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
+ struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
- INIT_LIST_HEAD(&need_pages);
- amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
- struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ list_del(&e->tv.head);
+ list_add(&e->tv.head, &userpage_list);
+ amdgpu_bo_unreserve(bo);
+ }
- if (amdgpu_ttm_tt_userptr_invalidated(bo->tbo.ttm,
- &e->user_invalidated) && e->user_pages) {
+ /* Get userptr backing pages. If pages are updated after registered
+ * in amdgpu_gem_userptr_ioctl(), amdgpu_cs_list_validate() will do
+ * amdgpu_ttm_backend_bind() to flush and invalidate new pages
+ */
+ if (!list_empty(&userpage_list)) {
+ /* Unreserve everything again, to avoid circular locking case
+ * bo::reserve -> mmap->sem
+ */
+ ttm_eu_backoff_reservation(&p->ticket, &p->validated);
- /* We acquired a page array, but somebody
- * invalidated it. Free it and try again
- */
- release_pages(e->user_pages,
- bo->tbo.ttm->num_pages);
- kvfree(e->user_pages);
- e->user_pages = NULL;
+ list_for_each_entry(e, &userpage_list, tv.head) {
+ struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
+ bool userpage_invalidated = false;
+ struct page **pages;
+ int i;
+
+ pages = kvmalloc_array(bo->tbo.ttm->num_pages,
+ sizeof(*pages),
+ GFP_KERNEL | __GFP_ZERO);
+ if (!pages) {
+ DRM_ERROR("calloc failure\n");
+ return -ENOMEM;
}
- if (amdgpu_ttm_tt_userptr_needs_pages(bo->tbo.ttm) &&
- !e->user_pages) {
- list_del(&e->tv.head);
- list_add(&e->tv.head, &need_pages);
-
- amdgpu_bo_unreserve(bo);
+ r = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, pages);
+ if (r) {
+ kvfree(pages);
+ return r;
}
- }
- if (list_empty(&need_pages))
- break;
-
- /* Unreserve everything again. */
- ttm_eu_backoff_reservation(&p->ticket, &p->validated);
-
- /* We tried too many times, just abort */
- if (!--tries) {
- r = -EDEADLK;
- DRM_ERROR("deadlock in %s\n", __func__);
- goto error_free_pages;
+ for (i = 0; i < bo->tbo.ttm->num_pages; i++) {
+ if (bo->tbo.ttm->pages[i] != pages[i]) {
+ bo->tbo.ttm->pages[i] = pages[i];
+ userpage_invalidated = true;
+ }
+ }
+ e->user_invalidated = userpage_invalidated;
+ kvfree(pages);
}
- /* Fill the page arrays for all userptrs. */
- list_for_each_entry(e, &need_pages, tv.head) {
- struct ttm_tt *ttm = e->tv.bo->ttm;
-
- e->user_pages = kvmalloc_array(ttm->num_pages,
- sizeof(struct page*),
- GFP_KERNEL | __GFP_ZERO);
- if (!e->user_pages) {
- r = -ENOMEM;
- DRM_ERROR("calloc failure in %s\n", __func__);
- goto error_free_pages;
- }
+ list_splice(&userpage_list, &p->validated);
- r = amdgpu_ttm_tt_get_user_pages(ttm, e->user_pages);
- if (r) {
- DRM_ERROR("amdgpu_ttm_tt_get_user_pages failed.\n");
- kvfree(e->user_pages);
- e->user_pages = NULL;
- goto error_free_pages;
- }
+ r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true,
+ &duplicates);
+ if (unlikely(r != 0)) {
+ if (r != -ERESTARTSYS)
+ DRM_ERROR("ttm_eu_reserve_buffers failed.\n");
+ goto out;
}
-
- /* And try again. */
- list_splice(&need_pages, &p->validated);
}
amdgpu_cs_get_threshold_for_moves(p->adev, &p->bytes_moved_threshold,
@@ -754,17 +735,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
error_validate:
if (r)
ttm_eu_backoff_reservation(&p->ticket, &p->validated);
-
-error_free_pages:
-
- amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
- if (!e->user_pages)
- continue;
-
- release_pages(e->user_pages, e->tv.bo->ttm->num_pages);
- kvfree(e->user_pages);
- }
-
+out:
return r;
}
@@ -1213,8 +1184,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
struct amdgpu_bo_list_entry *e;
struct amdgpu_job *job;
uint64_t seq;
-
- int r;
+ int r = 0;
job = p->job;
p->job = NULL;
@@ -1223,15 +1193,21 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
if (r)
goto error_unlock;
- /* No memory allocation is allowed while holding the mn lock */
+ /* No memory allocation is allowed while holding the mn lock.
+ * p->mn is hold until amdgpu_cs_submit is finished and fence is added
+ * to BOs.
+ */
amdgpu_mn_lock(p->mn);
+
+ /* If userptr are updated after amdgpu_cs_parser_bos(), restart cs */
amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
- if (amdgpu_ttm_tt_userptr_needs_pages(bo->tbo.ttm)) {
- r = -ERESTARTSYS;
- goto error_abort;
- }
+ r |= !amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm);
+ }
+ if (r) {
+ r = -ERESTARTSYS;
+ goto error_abort;
}
job->owner = p->filp;
@@ -1278,14 +1254,20 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
{
struct amdgpu_device *adev = dev->dev_private;
- union drm_amdgpu_cs *cs = data;
- struct amdgpu_cs_parser parser = {};
- bool reserved_buffers = false;
+ union drm_amdgpu_cs *cs;
+ struct amdgpu_cs_parser parser;
+ bool reserved_buffers;
+ int tries = 10;
int i, r;
if (!adev->accel_working)
return -EBUSY;
+restart:
+ memset(&parser, 0, sizeof(parser));
+ cs = data;
+ reserved_buffers = false;
+
parser.adev = adev;
parser.filp = filp;
@@ -1327,6 +1309,15 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
out:
amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
+
+ if (r == -ERESTARTSYS) {
+ if (!--tries) {
+ DRM_ERROR("Possible deadlock? Retry too many times\n");
+ return -EDEADLK;
+ }
+ goto restart;
+ }
+
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index f4f00217546e..92025993bfb1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -336,26 +336,24 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
r = amdgpu_bo_reserve(bo, true);
if (r)
- goto free_pages;
+ goto user_pages_done;
amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
amdgpu_bo_unreserve(bo);
if (r)
- goto free_pages;
+ goto user_pages_done;
}
r = drm_gem_handle_create(filp, gobj, &handle);
- /* drop reference from allocate - handle holds it now */
- drm_gem_object_put_unlocked(gobj);
if (r)
- return r;
+ goto user_pages_done;
args->handle = handle;
- return 0;
-free_pages:
- release_pages(bo->tbo.ttm->pages, bo->tbo.ttm->num_pages);
+user_pages_done:
+ if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE)
+ amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm);
release_object:
drm_gem_object_put_unlocked(gobj);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
index 5d518d2bb9be..5c24fb270480 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
@@ -229,8 +229,6 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
true, false, MAX_SCHEDULE_TIMEOUT);
if (r <= 0)
DRM_ERROR("(%ld) failed to wait for user bo\n", r);
-
- amdgpu_ttm_tt_mark_user_pages(bo->tbo.ttm);
}
}
@@ -356,7 +354,7 @@ static struct hmm_mirror_ops amdgpu_hmm_mirror_ops[] = {
* @adev: amdgpu device pointer
* @type: type of MMU notifier context
*
- * Creates a HMM mirror context for current->mm.
+ * Creates a HMM mirror context for mm.
*/
struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev,
enum amdgpu_mn_type type)
@@ -513,3 +511,26 @@ void amdgpu_mn_unregister(struct amdgpu_bo *bo)
mutex_unlock(&adev->mn_lock);
}
+/* flags used by HMM internal, not related to CPU/GPU PTE flags */
+static const uint64_t hmm_range_flags[HMM_PFN_FLAG_MAX] = {
+ (1 << 0), /* HMM_PFN_VALID */
+ (1 << 1), /* HMM_PFN_WRITE */
+ 0 /* HMM_PFN_DEVICE_PRIVATE */
+};
+
+static const uint64_t hmm_range_values[HMM_PFN_VALUE_MAX] = {
+ 0xfffffffffffffffeUL, /* HMM_PFN_ERROR */
+ 0, /* HMM_PFN_NONE */
+ 0xfffffffffffffffcUL /* HMM_PFN_SPECIAL */
+};
+
+void amdgpu_hmm_init_range(struct hmm_range *range)
+{
+ if (range) {
+ range->flags = hmm_range_flags;
+ range->values = hmm_range_values;
+ range->pfn_shift = PAGE_SHIFT;
+ range->pfns = NULL;
+ INIT_LIST_HEAD(&range->list);
+ }
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
index 0a51fd00021c..4803e216e174 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.h
@@ -25,9 +25,10 @@
#define __AMDGPU_MN_H__
/*
- * MMU Notifier
+ * HMM mirror
*/
struct amdgpu_mn;
+struct hmm_range;
enum amdgpu_mn_type {
AMDGPU_MN_TYPE_GFX,
@@ -41,6 +42,7 @@ struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev,
enum amdgpu_mn_type type);
int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr);
void amdgpu_mn_unregister(struct amdgpu_bo *bo);
+void amdgpu_hmm_init_range(struct hmm_range *range);
#else
static inline void amdgpu_mn_lock(struct amdgpu_mn *mn) {}
static inline void amdgpu_mn_unlock(struct amdgpu_mn *mn) {}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c91ec3101d00..60e7ff4a906d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -43,6 +43,7 @@
#include <linux/pagemap.h>
#include <linux/debugfs.h>
#include <linux/iommu.h>
+#include <linux/hmm.h>
#include "amdgpu.h"
#include "amdgpu_object.h"
#include "amdgpu_trace.h"
@@ -705,11 +706,6 @@ static unsigned long amdgpu_ttm_io_mem_pfn(struct ttm_buffer_object *bo,
/*
* TTM backend functions.
*/
-struct amdgpu_ttm_gup_task_list {
- struct list_head list;
- struct task_struct *task;
-};
-
struct amdgpu_ttm_tt {
struct ttm_dma_tt ttm;
u64 offset;
@@ -718,107 +714,97 @@ struct amdgpu_ttm_tt {
uint32_t userflags;
spinlock_t guptasklock;
struct list_head guptasks;
- atomic_t mmu_invalidations;
- uint32_t last_set_pages;
+ struct hmm_range range;
};
/**
- * amdgpu_ttm_tt_get_user_pages - Pin pages of memory pointed to by a USERPTR
- * pointer to memory
+ * amdgpu_ttm_tt_get_user_pages - get device accessible pages that back user
+ * memory and start HMM tracking CPU page table update
*
- * Called by amdgpu_gem_userptr_ioctl() and amdgpu_cs_parser_bos().
- * This provides a wrapper around the get_user_pages() call to provide
- * device accessible pages that back user memory.
+ * Calling function must call amdgpu_ttm_tt_userptr_range_done() once and only
+ * once afterwards to stop HMM tracking
*/
int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
{
struct amdgpu_ttm_tt *gtt = (void *)ttm;
struct mm_struct *mm = gtt->usertask->mm;
- unsigned int flags = 0;
- unsigned pinned = 0;
- int r;
+ unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
+ struct hmm_range *range = >t->range;
+ int r = 0, i;
if (!mm) /* Happens during process shutdown */
return -ESRCH;
- if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
- flags |= FOLL_WRITE;
+ amdgpu_hmm_init_range(range);
down_read(&mm->mmap_sem);
- if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) {
- /*
- * check that we only use anonymous memory to prevent problems
- * with writeback
- */
- unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
- struct vm_area_struct *vma;
+ range->vma = find_vma(mm, gtt->userptr);
+ if (!range_in_vma(range->vma, gtt->userptr, end))
+ r = -EFAULT;
+ else if ((gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) &&
+ range->vma->vm_file)
+ r = -EPERM;
+ if (r)
+ goto out;
- vma = find_vma(mm, gtt->userptr);
- if (!vma || vma->vm_file || vma->vm_end < end) {
- up_read(&mm->mmap_sem);
- return -EPERM;
- }
+ range->pfns = kvmalloc_array(ttm->num_pages, sizeof(uint64_t),
+ GFP_KERNEL);
+ if (range->pfns == NULL) {
+ r = -ENOMEM;
+ goto out;
}
+ range->start = gtt->userptr;
+ range->end = end;
- /* loop enough times using contiguous pages of memory */
- do {
- unsigned num_pages = ttm->num_pages - pinned;
- uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
- struct page **p = pages + pinned;
- struct amdgpu_ttm_gup_task_list guptask;
-
- guptask.task = current;
- spin_lock(>t->guptasklock);
- list_add(&guptask.list, >t->guptasks);
- spin_unlock(>t->guptasklock);
-
- if (mm == current->mm)
- r = get_user_pages(userptr, num_pages, flags, p, NULL);
- else
- r = get_user_pages_remote(gtt->usertask,
- mm, userptr, num_pages,
- flags, p, NULL, NULL);
+ range->pfns[0] = range->flags[HMM_PFN_VALID];
+ range->pfns[0] |= amdgpu_ttm_tt_is_readonly(ttm) ?
+ 0 : range->flags[HMM_PFN_WRITE];
+ for (i = 1; i < ttm->num_pages; i++)
+ range->pfns[i] = range->pfns[0];
- spin_lock(>t->guptasklock);
- list_del(&guptask.list);
- spin_unlock(>t->guptasklock);
-
- if (r < 0)
- goto release_pages;
+ /* This may trigger page table update */
+ r = hmm_vma_fault(range, true);
+ if (r)
+ goto out_free_pfns;
- pinned += r;
+ up_read(&mm->mmap_sem);
- } while (pinned < ttm->num_pages);
+ for (i = 0; i < ttm->num_pages; i++)
+ pages[i] = hmm_pfn_to_page(range, range->pfns[i]);
- up_read(&mm->mmap_sem);
return 0;
-release_pages:
- release_pages(pages, pinned);
+out_free_pfns:
+ kvfree(range->pfns);
+ range->pfns = NULL;
+out:
up_read(&mm->mmap_sem);
return r;
}
/**
- * amdgpu_ttm_tt_set_user_pages - Copy pages in, putting old pages as necessary.
+ * amdgpu_ttm_tt_userptr_range_done - stop HMM track the CPU page table change
+ * Check if the pages backing this ttm range have been invalidated
*
- * Called by amdgpu_cs_list_validate(). This creates the page list
- * that backs user memory and will ultimately be mapped into the device
- * address space.
+ * Returns: true if pages are still valid
*/
-void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
+bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm)
{
struct amdgpu_ttm_tt *gtt = (void *)ttm;
- unsigned i;
+ bool r;
- gtt->last_set_pages = atomic_read(>t->mmu_invalidations);
- for (i = 0; i < ttm->num_pages; ++i) {
- if (ttm->pages[i])
- put_page(ttm->pages[i]);
+ if (!gtt || !gtt->userptr)
+ return false;
+
+ r = hmm_vma_range_done(>t->range);
- ttm->pages[i] = pages ? pages[i] : NULL;
+ if (gtt->range.pfns) {
+ kvfree(gtt->range.pfns);
+ gtt->range.pfns = NULL;
}
+
+ return r;
}
/**
@@ -903,9 +889,6 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
/* unmap the pages mapped to the device */
dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
- /* mark the pages as dirty */
- amdgpu_ttm_tt_mark_user_pages(ttm);
-
sg_free_table(ttm->sg);
}
@@ -1207,7 +1190,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_tt *ttm)
bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
if (gtt && gtt->userptr) {
- amdgpu_ttm_tt_set_user_pages(ttm, NULL);
+ memset(ttm->pages, 0, ttm->num_pages * sizeof(struct page *));
kfree(ttm->sg);
ttm->page_flags &= ~TTM_PAGE_FLAG_SG;
return;
@@ -1258,8 +1241,6 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
spin_lock_init(>t->guptasklock);
INIT_LIST_HEAD(>t->guptasks);
- atomic_set(>t->mmu_invalidations, 0);
- gtt->last_set_pages = 0;
return 0;
}
@@ -1289,7 +1270,6 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
unsigned long end)
{
struct amdgpu_ttm_tt *gtt = (void *)ttm;
- struct amdgpu_ttm_gup_task_list *entry;
unsigned long size;
if (gtt == NULL || !gtt->userptr)
@@ -1302,48 +1282,20 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
if (gtt->userptr > end || gtt->userptr + size <= start)
return false;
- /* Search the lists of tasks that hold this mapping and see
- * if current is one of them. If it is return false.
- */
- spin_lock(>t->guptasklock);
- list_for_each_entry(entry, >t->guptasks, list) {
- if (entry->task == current) {
- spin_unlock(>t->guptasklock);
- return false;
- }
- }
- spin_unlock(>t->guptasklock);
-
- atomic_inc(>t->mmu_invalidations);
-
return true;
}
/**
- * amdgpu_ttm_tt_userptr_invalidated - Has the ttm_tt object been invalidated?
+ * amdgpu_ttm_tt_is_userptr - Have the pages backing by userptr?
*/
-bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
- int *last_invalidated)
-{
- struct amdgpu_ttm_tt *gtt = (void *)ttm;
- int prev_invalidated = *last_invalidated;
-
- *last_invalidated = atomic_read(>t->mmu_invalidations);
- return prev_invalidated != *last_invalidated;
-}
-
-/**
- * amdgpu_ttm_tt_userptr_needs_pages - Have the pages backing this ttm_tt object
- * been invalidated since the last time they've been set?
- */
-bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm)
+bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm)
{
struct amdgpu_ttm_tt *gtt = (void *)ttm;
if (gtt == NULL || !gtt->userptr)
return false;
- return atomic_read(>t->mmu_invalidations) != gtt->last_set_pages;
+ return true;
}
/**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index b5b2d101f7db..c5a1e78a09b1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -102,7 +102,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages);
-void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages);
+bool amdgpu_ttm_tt_get_user_pages_done(struct ttm_tt *ttm);
void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm);
int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
uint32_t flags);
@@ -112,7 +112,7 @@ bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
unsigned long end);
bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
int *last_invalidated);
-bool amdgpu_ttm_tt_userptr_needs_pages(struct ttm_tt *ttm);
+bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm);
bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm);
uint64_t amdgpu_ttm_tt_pde_flags(struct ttm_tt *ttm, struct ttm_mem_reg *mem);
uint64_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index e73d152659a2..1f8ab1a4fabc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -619,7 +619,6 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
entry->tv.bo = &vm->root.base.bo->tbo;
/* One for the VM updates, one for TTM and one for the CS job */
entry->tv.num_shared = 3;
- entry->user_pages = NULL;
list_add(&entry->tv.head, validated);
}
--
2.17.1
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
