habanalabs: fix possible deadlock in cache invl failure

Currently there is a deadlock in driver in scenarios where MMU
cache invalidation fails. The issue is basically device reset
being performed without releasing the MMU mutex.
The solution is to skip device reset as it is not necessary.
In addition we introduce a slight code refactor that prints the
invalidation error from a single location.

Signed-off-by: Ofir Bitton <obitton@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
This commit is contained in:
Ofir Bitton 2021-11-09 13:12:38 +02:00 committed by Oded Gabbay
parent 6f61e47a68
commit a1b838adb0
6 changed files with 39 additions and 35 deletions

View file

@ -80,14 +80,13 @@ static int cb_map_mem(struct hl_ctx *ctx, struct hl_cb *cb)
offset += va_block->size;
}
hdev->asic_funcs->mmu_invalidate_cache(hdev, false,
MMU_OP_USERPTR | MMU_OP_SKIP_LOW_CACHE_INV);
rc = hl_mmu_invalidate_cache(hdev, false, MMU_OP_USERPTR | MMU_OP_SKIP_LOW_CACHE_INV);
mutex_unlock(&ctx->mmu_lock);
cb->is_mmu_mapped = true;
return 0;
return rc;
err_va_umap:
list_for_each_entry(va_block, &cb->va_block_list, node) {
@ -98,7 +97,7 @@ static int cb_map_mem(struct hl_ctx *ctx, struct hl_cb *cb)
offset -= va_block->size;
}
hdev->asic_funcs->mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
rc = hl_mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
mutex_unlock(&ctx->mmu_lock);
@ -127,7 +126,7 @@ static void cb_unmap_mem(struct hl_ctx *ctx, struct hl_cb *cb)
"Failed to unmap CB's va 0x%llx\n",
va_block->start);
hdev->asic_funcs->mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
hl_mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
mutex_unlock(&ctx->mmu_lock);

View file

@ -2993,6 +2993,9 @@ int hl_mmu_unmap_page(struct hl_ctx *ctx, u64 virt_addr, u32 page_size,
int hl_mmu_map_contiguous(struct hl_ctx *ctx, u64 virt_addr,
u64 phys_addr, u32 size);
int hl_mmu_unmap_contiguous(struct hl_ctx *ctx, u64 virt_addr, u32 size);
int hl_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard, u32 flags);
int hl_mmu_invalidate_cache_range(struct hl_device *hdev, bool is_hard,
u32 flags, u32 asid, u64 va, u64 size);
void hl_mmu_swap_out(struct hl_ctx *ctx);
void hl_mmu_swap_in(struct hl_ctx *ctx);
int hl_mmu_if_set_funcs(struct hl_device *hdev);

View file

@ -1201,18 +1201,13 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
goto map_err;
}
rc = hdev->asic_funcs->mmu_invalidate_cache_range(hdev, false,
*vm_type | MMU_OP_SKIP_LOW_CACHE_INV,
ctx->asid, ret_vaddr, phys_pg_pack->total_size);
rc = hl_mmu_invalidate_cache_range(hdev, false, *vm_type | MMU_OP_SKIP_LOW_CACHE_INV,
ctx->asid, ret_vaddr, phys_pg_pack->total_size);
mutex_unlock(&ctx->mmu_lock);
if (rc) {
dev_err(hdev->dev,
"mapping handle %u failed due to MMU cache invalidation\n",
handle);
if (rc)
goto map_err;
}
ret_vaddr += phys_pg_pack->offset;
@ -1350,9 +1345,8 @@ static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
* at the loop end rather than for each iteration
*/
if (!ctx_free)
rc = hdev->asic_funcs->mmu_invalidate_cache_range(hdev, true,
*vm_type, ctx->asid, vaddr,
phys_pg_pack->total_size);
rc = hl_mmu_invalidate_cache_range(hdev, true, *vm_type, ctx->asid, vaddr,
phys_pg_pack->total_size);
mutex_unlock(&ctx->mmu_lock);
@ -1365,11 +1359,6 @@ static int unmap_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
if (!ctx_free) {
int tmp_rc;
if (rc)
dev_err(hdev->dev,
"unmapping vaddr 0x%llx failed due to MMU cache invalidation\n",
vaddr);
tmp_rc = add_va_block(hdev, va_range, vaddr,
vaddr + phys_pg_pack->total_size - 1);
if (tmp_rc) {
@ -2640,8 +2629,8 @@ void hl_vm_ctx_fini(struct hl_ctx *ctx)
mutex_lock(&ctx->mmu_lock);
/* invalidate the cache once after the unmapping loop */
hdev->asic_funcs->mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
hdev->asic_funcs->mmu_invalidate_cache(hdev, true, MMU_OP_PHYS_PACK);
hl_mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
hl_mmu_invalidate_cache(hdev, true, MMU_OP_PHYS_PACK);
mutex_unlock(&ctx->mmu_lock);

View file

@ -637,3 +637,28 @@ u64 hl_mmu_descramble_addr(struct hl_device *hdev, u64 addr)
{
return addr;
}
int hl_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard, u32 flags)
{
int rc;
rc = hdev->asic_funcs->mmu_invalidate_cache(hdev, is_hard, flags);
if (rc)
dev_err_ratelimited(hdev->dev, "MMU cache invalidation failed\n");
return rc;
}
int hl_mmu_invalidate_cache_range(struct hl_device *hdev, bool is_hard,
u32 flags, u32 asid, u64 va, u64 size)
{
int rc;
rc = hdev->asic_funcs->mmu_invalidate_cache_range(hdev, is_hard, flags,
asid, va, size);
if (rc)
dev_err_ratelimited(hdev->dev, "MMU cache range invalidation failed\n");
return rc;
}

View file

@ -8366,12 +8366,6 @@ static int gaudi_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard,
WREG32(mmSTLB_INV_SET, 0);
if (rc) {
dev_err_ratelimited(hdev->dev,
"MMU cache invalidation timeout\n");
hl_device_reset(hdev, HL_DRV_RESET_HARD);
}
return rc;
}

View file

@ -5258,12 +5258,6 @@ static int goya_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard,
1000,
timeout_usec);
if (rc) {
dev_err_ratelimited(hdev->dev,
"MMU cache invalidation timeout\n");
hl_device_reset(hdev, HL_DRV_RESET_HARD);
}
return rc;
}