* [PATCH 2/5] habanalabs: do MMU prefetch as deferred work
2022-05-20 14:33 [PATCH 1/5] habanalabs: order memory manager messages Oded Gabbay
@ 2022-05-20 14:33 ` Oded Gabbay
2022-05-20 14:33 ` [PATCH 3/5] habanalabs: remove hdev from hl_ctx_get args Oded Gabbay
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Oded Gabbay @ 2022-05-20 14:33 UTC (permalink / raw)
To: linux-kernel; +Cc: Ohad Sharabi
From: Ohad Sharabi <osharabi@habana.ai>
When user requests to prefetch the MMU translations, the driver will
not block the user until prefetch is done.
Instead, the prefetch work will be delegated to a WQ which will do it
in the background.
This way, the prefetch may progress without blocking the user at all.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
drivers/misc/habanalabs/common/device.c | 15 ++++++-
drivers/misc/habanalabs/common/habanalabs.h | 27 +++++++++--
drivers/misc/habanalabs/common/memory.c | 31 ++++++++-----
drivers/misc/habanalabs/common/mmu/mmu.c | 50 ++++++++++++++++++---
drivers/misc/habanalabs/gaudi/gaudi.c | 8 +---
drivers/misc/habanalabs/goya/goya.c | 8 +---
6 files changed, 103 insertions(+), 36 deletions(-)
diff --git a/drivers/misc/habanalabs/common/device.c b/drivers/misc/habanalabs/common/device.c
index 0908ac301c70..b4f14c6d3970 100644
--- a/drivers/misc/habanalabs/common/device.c
+++ b/drivers/misc/habanalabs/common/device.c
@@ -612,11 +612,18 @@ static int device_early_init(struct hl_device *hdev)
goto free_eq_wq;
}
+ hdev->pf_wq = alloc_workqueue("hl-prefetch", WQ_UNBOUND, 0);
+ if (!hdev->pf_wq) {
+ dev_err(hdev->dev, "Failed to allocate MMU prefetch workqueue\n");
+ rc = -ENOMEM;
+ goto free_ts_free_wq;
+ }
+
hdev->hl_chip_info = kzalloc(sizeof(struct hwmon_chip_info),
GFP_KERNEL);
if (!hdev->hl_chip_info) {
rc = -ENOMEM;
- goto free_ts_free_wq;
+ goto free_pf_wq;
}
rc = hl_mmu_if_set_funcs(hdev);
@@ -655,6 +662,8 @@ static int device_early_init(struct hl_device *hdev)
hl_mem_mgr_fini(&hdev->kernel_mem_mgr);
free_chip_info:
kfree(hdev->hl_chip_info);
+free_pf_wq:
+ destroy_workqueue(hdev->pf_wq);
free_ts_free_wq:
destroy_workqueue(hdev->ts_free_obj_wq);
free_eq_wq:
@@ -695,6 +704,7 @@ static void device_early_fini(struct hl_device *hdev)
kfree(hdev->hl_chip_info);
+ destroy_workqueue(hdev->pf_wq);
destroy_workqueue(hdev->ts_free_obj_wq);
destroy_workqueue(hdev->eq_wq);
destroy_workqueue(hdev->device_reset_work.wq);
@@ -891,6 +901,9 @@ static void cleanup_resources(struct hl_device *hdev, bool hard_reset, bool fw_r
/* Go over all the queues, release all CS and their jobs */
hl_cs_rollback_all(hdev, skip_wq_flush);
+ /* flush the MMU prefetch workqueue */
+ flush_workqueue(hdev->pf_wq);
+
/* Release all pending user interrupts, each pending user interrupt
* holds a reference to user context
*/
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index 8977ec67dba7..632037b29922 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -1402,8 +1402,7 @@ struct hl_asic_funcs {
u32 flags);
int (*mmu_invalidate_cache_range)(struct hl_device *hdev, bool is_hard,
u32 flags, u32 asid, u64 va, u64 size);
- int (*mmu_prefetch_cache_range)(struct hl_device *hdev, u32 flags, u32 asid, u64 va,
- u64 size);
+ int (*mmu_prefetch_cache_range)(struct hl_ctx *ctx, u32 flags, u32 asid, u64 va, u64 size);
int (*send_heartbeat)(struct hl_device *hdev);
int (*debug_coresight)(struct hl_device *hdev, struct hl_ctx *ctx, void *data);
bool (*is_device_idle)(struct hl_device *hdev, u64 *mask_arr,
@@ -2476,6 +2475,24 @@ struct hl_mmu_funcs {
};
/**
+ * struct hl_prefetch_work - prefetch work structure handler
+ * @pf_work: actual work struct.
+ * @ctx: compute context.
+ * @va: virtual address to pre-fetch.
+ * @size: pre-fetch size.
+ * @flags: operation flags.
+ * @asid: ASID for maintenance operation.
+ */
+struct hl_prefetch_work {
+ struct work_struct pf_work;
+ struct hl_ctx *ctx;
+ u64 va;
+ u64 size;
+ u32 flags;
+ u32 asid;
+};
+
+/*
* number of user contexts allowed to call wait_for_multi_cs ioctl in
* parallel
*/
@@ -2648,6 +2665,7 @@ struct hl_reset_info {
* context.
* @eq_wq: work queue of event queue for executing work in process context.
* @ts_free_obj_wq: work queue for timestamp registration objects release.
+ * @pf_wq: work queue for MMU pre-fetch operations.
* @kernel_ctx: Kernel driver context structure.
* @kernel_queues: array of hl_hw_queue.
* @cs_mirror_list: CS mirror list for TDR.
@@ -2760,6 +2778,7 @@ struct hl_reset_info {
* @supports_wait_for_multi_cs: true if wait for multi CS is supported
* @is_compute_ctx_active: Whether there is an active compute context executing.
* @compute_ctx_in_release: true if the current compute context is being released.
+ * @supports_mmu_prefetch: true if prefetch is supported, otherwise false.
*/
struct hl_device {
struct pci_dev *pdev;
@@ -2781,6 +2800,7 @@ struct hl_device {
struct workqueue_struct **cq_wq;
struct workqueue_struct *eq_wq;
struct workqueue_struct *ts_free_obj_wq;
+ struct workqueue_struct *pf_wq;
struct hl_ctx *kernel_ctx;
struct hl_hw_queue *kernel_queues;
struct list_head cs_mirror_list;
@@ -2882,6 +2902,7 @@ struct hl_device {
u8 stream_master_qid_arr_size;
u8 is_compute_ctx_active;
u8 compute_ctx_in_release;
+ u8 supports_mmu_prefetch;
/* Parameters for bring-up */
u64 nic_ports_mask;
@@ -3163,7 +3184,7 @@ 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);
-int hl_mmu_prefetch_cache_range(struct hl_device *hdev, u32 flags, u32 asid, u64 va, u64 size);
+int hl_mmu_prefetch_cache_range(struct hl_ctx *ctx, u32 flags, u32 asid, u64 va, u64 size);
u64 hl_mmu_get_next_hop_addr(struct hl_ctx *ctx, u64 curr_pte);
u64 hl_mmu_get_hop_pte_phys_addr(struct hl_ctx *ctx, struct hl_mmu_properties *mmu_prop,
u8 hop_idx, u64 hop_addr, u64 virt_addr);
diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c
index ecf3c094242a..087a55654a4d 100644
--- a/drivers/misc/habanalabs/common/memory.c
+++ b/drivers/misc/habanalabs/common/memory.c
@@ -1102,21 +1102,24 @@ static int get_paddr_from_handle(struct hl_ctx *ctx, struct hl_mem_in *args,
* map a device virtual block to this pages and return the start address of
* this block.
*/
-static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
- u64 *device_addr)
+static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args, u64 *device_addr)
{
- struct hl_device *hdev = ctx->hdev;
- struct hl_vm *vm = &hdev->vm;
struct hl_vm_phys_pg_pack *phys_pg_pack;
+ enum hl_va_range_type va_range_type = 0;
+ struct hl_device *hdev = ctx->hdev;
struct hl_userptr *userptr = NULL;
+ u32 handle = 0, va_block_align;
struct hl_vm_hash_node *hnode;
+ struct hl_vm *vm = &hdev->vm;
struct hl_va_range *va_range;
- enum vm_type *vm_type;
+ bool is_userptr, do_prefetch;
u64 ret_vaddr, hint_addr;
- u32 handle = 0, va_block_align;
+ enum vm_type *vm_type;
int rc;
- bool is_userptr = args->flags & HL_MEM_USERPTR;
- enum hl_va_range_type va_range_type = 0;
+
+ /* set map flags */
+ is_userptr = args->flags & HL_MEM_USERPTR;
+ do_prefetch = hdev->supports_mmu_prefetch && (args->flags & HL_MEM_PREFETCH);
/* Assume failure */
*device_addr = 0;
@@ -1250,15 +1253,19 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
if (rc)
goto map_err;
- if (args->flags & HL_MEM_PREFETCH) {
- rc = hl_mmu_prefetch_cache_range(hdev, *vm_type, ctx->asid, ret_vaddr,
+ mutex_unlock(&ctx->mmu_lock);
+
+ /*
+ * prefetch is done upon user's request. it is performed in WQ as and so can
+ * be outside the MMU lock. the operation itself is already protected by the mmu lock
+ */
+ if (do_prefetch) {
+ rc = hl_mmu_prefetch_cache_range(ctx, *vm_type, ctx->asid, ret_vaddr,
phys_pg_pack->total_size);
if (rc)
goto map_err;
}
- mutex_unlock(&ctx->mmu_lock);
-
ret_vaddr += phys_pg_pack->offset;
hnode->ptr = vm_type;
diff --git a/drivers/misc/habanalabs/common/mmu/mmu.c b/drivers/misc/habanalabs/common/mmu/mmu.c
index 04e53af4c67f..ae9b4923c32b 100644
--- a/drivers/misc/habanalabs/common/mmu/mmu.c
+++ b/drivers/misc/habanalabs/common/mmu/mmu.c
@@ -665,15 +665,53 @@ int hl_mmu_invalidate_cache_range(struct hl_device *hdev, bool is_hard,
return rc;
}
-int hl_mmu_prefetch_cache_range(struct hl_device *hdev, u32 flags, u32 asid, u64 va, u64 size)
+static void hl_mmu_prefetch_work_function(struct work_struct *work)
{
- int rc;
+ struct hl_prefetch_work *pfw = container_of(work, struct hl_prefetch_work, pf_work);
+ struct hl_ctx *ctx = pfw->ctx;
- rc = hdev->asic_funcs->mmu_prefetch_cache_range(hdev, flags, asid, va, size);
- if (rc)
- dev_err_ratelimited(hdev->dev, "MMU cache range prefetch failed\n");
+ if (!hl_device_operational(ctx->hdev, NULL))
+ goto put_ctx;
- return rc;
+ mutex_lock(&ctx->mmu_lock);
+
+ ctx->hdev->asic_funcs->mmu_prefetch_cache_range(ctx, pfw->flags, pfw->asid,
+ pfw->va, pfw->size);
+
+ mutex_unlock(&ctx->mmu_lock);
+
+put_ctx:
+ /*
+ * context was taken in the common mmu prefetch function- see comment there about
+ * context handling.
+ */
+ hl_ctx_put(ctx);
+ kfree(pfw);
+}
+
+int hl_mmu_prefetch_cache_range(struct hl_ctx *ctx, u32 flags, u32 asid, u64 va, u64 size)
+{
+ struct hl_prefetch_work *handle_pf_work;
+
+ handle_pf_work = kmalloc(sizeof(*handle_pf_work), GFP_KERNEL);
+ if (!handle_pf_work)
+ return -ENOMEM;
+
+ INIT_WORK(&handle_pf_work->pf_work, hl_mmu_prefetch_work_function);
+ handle_pf_work->ctx = ctx;
+ handle_pf_work->va = va;
+ handle_pf_work->size = size;
+ handle_pf_work->flags = flags;
+ handle_pf_work->asid = asid;
+
+ /*
+ * as actual prefetch is done in a WQ we must get the context (and put it
+ * at the end of the work function)
+ */
+ hl_ctx_get(ctx->hdev, ctx);
+ queue_work(ctx->hdev->pf_wq, &handle_pf_work->pf_work);
+
+ return 0;
}
u64 hl_mmu_get_next_hop_addr(struct hl_ctx *ctx, u64 curr_pte)
diff --git a/drivers/misc/habanalabs/gaudi/gaudi.c b/drivers/misc/habanalabs/gaudi/gaudi.c
index 1c388537de33..96a83317b302 100644
--- a/drivers/misc/habanalabs/gaudi/gaudi.c
+++ b/drivers/misc/habanalabs/gaudi/gaudi.c
@@ -9174,12 +9174,6 @@ static void gaudi_get_valid_dram_page_orders(struct hl_info_dev_memalloc_page_si
info->page_order_bitmask = 0;
}
-static int gaudi_mmu_prefetch_cache_range(struct hl_device *hdev, u32 flags, u32 asid, u64 va,
- u64 size)
-{
- return 0;
-}
-
static ssize_t infineon_ver_show(struct device *dev, struct device_attribute *attr, char *buf)
{
struct hl_device *hdev = dev_get_drvdata(dev);
@@ -9244,7 +9238,7 @@ static const struct hl_asic_funcs gaudi_funcs = {
.write_pte = gaudi_write_pte,
.mmu_invalidate_cache = gaudi_mmu_invalidate_cache,
.mmu_invalidate_cache_range = gaudi_mmu_invalidate_cache_range,
- .mmu_prefetch_cache_range = gaudi_mmu_prefetch_cache_range,
+ .mmu_prefetch_cache_range = NULL,
.send_heartbeat = gaudi_send_heartbeat,
.debug_coresight = gaudi_debug_coresight,
.is_device_idle = gaudi_is_device_idle,
diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c
index f2d4362f6a46..4cde505a7416 100644
--- a/drivers/misc/habanalabs/goya/goya.c
+++ b/drivers/misc/habanalabs/goya/goya.c
@@ -5428,12 +5428,6 @@ static int goya_get_monitor_dump(struct hl_device *hdev, void *data)
return -EOPNOTSUPP;
}
-static int goya_mmu_prefetch_cache_range(struct hl_device *hdev, u32 flags, u32 asid, u64 va,
- u64 size)
-{
- return 0;
-}
-
static int goya_scrub_device_dram(struct hl_device *hdev, u64 val)
{
return -EOPNOTSUPP;
@@ -5480,7 +5474,7 @@ static const struct hl_asic_funcs goya_funcs = {
.write_pte = goya_write_pte,
.mmu_invalidate_cache = goya_mmu_invalidate_cache,
.mmu_invalidate_cache_range = goya_mmu_invalidate_cache_range,
- .mmu_prefetch_cache_range = goya_mmu_prefetch_cache_range,
+ .mmu_prefetch_cache_range = NULL,
.send_heartbeat = goya_send_heartbeat,
.debug_coresight = goya_debug_coresight,
.is_device_idle = goya_is_device_idle,
--
2.25.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH 3/5] habanalabs: remove hdev from hl_ctx_get args
2022-05-20 14:33 [PATCH 1/5] habanalabs: order memory manager messages Oded Gabbay
2022-05-20 14:33 ` [PATCH 2/5] habanalabs: do MMU prefetch as deferred work Oded Gabbay
@ 2022-05-20 14:33 ` Oded Gabbay
2022-05-20 14:33 ` [PATCH 4/5] habanalabs: fix missing handle shift during mmap Oded Gabbay
2022-05-20 14:33 ` [PATCH 5/5] habanalabs: use separate structure info for each error collect data Oded Gabbay
3 siblings, 0 replies; 5+ messages in thread
From: Oded Gabbay @ 2022-05-20 14:33 UTC (permalink / raw)
To: linux-kernel; +Cc: Ohad Sharabi
From: Ohad Sharabi <osharabi@habana.ai>
This argument is unused by the function.
Signed-off-by: Ohad Sharabi <osharabi@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
drivers/misc/habanalabs/common/command_buffer.c | 2 +-
drivers/misc/habanalabs/common/command_submission.c | 12 ++++++------
drivers/misc/habanalabs/common/context.c | 4 ++--
drivers/misc/habanalabs/common/habanalabs.h | 2 +-
drivers/misc/habanalabs/common/memory.c | 4 ++--
drivers/misc/habanalabs/common/mmu/mmu.c | 2 +-
6 files changed, 13 insertions(+), 13 deletions(-)
diff --git a/drivers/misc/habanalabs/common/command_buffer.c b/drivers/misc/habanalabs/common/command_buffer.c
index 1fac72c38c87..e13b2b39c058 100644
--- a/drivers/misc/habanalabs/common/command_buffer.c
+++ b/drivers/misc/habanalabs/common/command_buffer.c
@@ -283,7 +283,7 @@ static int hl_cb_mmap_mem_alloc(struct hl_mmap_mem_buf *buf, gfp_t gfp, void *ar
cb->buf->mappable_size = cb->size;
cb->buf->private = cb;
- hl_ctx_get(cb_args->hdev, cb->ctx);
+ hl_ctx_get(cb->ctx);
if (cb_args->map_cb) {
if (ctx_id == HL_KERNEL_ASID_ID) {
diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c
index 41206c5d6e7b..d8b6a2a4b210 100644
--- a/drivers/misc/habanalabs/common/command_submission.c
+++ b/drivers/misc/habanalabs/common/command_submission.c
@@ -805,7 +805,7 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx,
}
/* increment refcnt for context */
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
cs->ctx = ctx;
cs->submitted = false;
@@ -1830,7 +1830,7 @@ static int cs_ioctl_reserve_signals(struct hl_fpriv *hpriv,
handle->count = count;
- hl_ctx_get(hdev, hpriv->ctx);
+ hl_ctx_get(hpriv->ctx);
handle->ctx = hpriv->ctx;
mgr = &hpriv->ctx->sig_mgr;
@@ -2524,7 +2524,7 @@ static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
if (timestamp)
*timestamp = 0;
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
fence = hl_ctx_get_fence(ctx, seq);
@@ -2715,7 +2715,7 @@ static int hl_multi_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data)
mcs_data.fence_arr = fence_arr;
mcs_data.arr_len = seq_arr_len;
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
/* wait (with timeout) for the first CS to be completed */
mcs_data.timeout_jiffies = hl_usecs64_to_jiffies(args->in.timeout_us);
@@ -2958,7 +2958,7 @@ static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
timeout = hl_usecs64_to_jiffies(timeout_us);
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
cq_cb = hl_cb_get(cb_mmg, cq_counters_handle);
if (!cq_cb) {
@@ -3107,7 +3107,7 @@ static int _hl_interrupt_wait_ioctl_user_addr(struct hl_device *hdev, struct hl_
timeout = hl_usecs64_to_jiffies(timeout_us);
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
pend = kzalloc(sizeof(*pend), GFP_KERNEL);
if (!pend) {
diff --git a/drivers/misc/habanalabs/common/context.c b/drivers/misc/habanalabs/common/context.c
index c6360e33bce8..ed2cfd0c6e99 100644
--- a/drivers/misc/habanalabs/common/context.c
+++ b/drivers/misc/habanalabs/common/context.c
@@ -262,7 +262,7 @@ int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx)
return rc;
}
-void hl_ctx_get(struct hl_device *hdev, struct hl_ctx *ctx)
+void hl_ctx_get(struct hl_ctx *ctx)
{
kref_get(&ctx->refcount);
}
@@ -284,7 +284,7 @@ struct hl_ctx *hl_get_compute_ctx(struct hl_device *hdev)
* immediately once we find him
*/
ctx = hpriv->ctx;
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
break;
}
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index 632037b29922..99cbed5ef0e6 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -3085,7 +3085,7 @@ int hl_ctx_create(struct hl_device *hdev, struct hl_fpriv *hpriv);
void hl_ctx_free(struct hl_device *hdev, struct hl_ctx *ctx);
int hl_ctx_init(struct hl_device *hdev, struct hl_ctx *ctx, bool is_kernel_ctx);
void hl_ctx_do_release(struct kref *ref);
-void hl_ctx_get(struct hl_device *hdev, struct hl_ctx *ctx);
+void hl_ctx_get(struct hl_ctx *ctx);
int hl_ctx_put(struct hl_ctx *ctx);
struct hl_ctx *hl_get_compute_ctx(struct hl_device *hdev);
struct hl_fence *hl_ctx_get_fence(struct hl_ctx *ctx, u64 seq);
diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c
index 087a55654a4d..663dd7e589d4 100644
--- a/drivers/misc/habanalabs/common/memory.c
+++ b/drivers/misc/habanalabs/common/memory.c
@@ -1522,7 +1522,7 @@ int hl_hw_block_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma)
vma->vm_ops = &hw_block_vm_ops;
vma->vm_private_data = lnode;
- hl_ctx_get(hdev, ctx);
+ hl_ctx_get(ctx);
rc = hdev->asic_funcs->hw_block_mmap(hdev, vma, block_id, block_size);
if (rc) {
@@ -1832,7 +1832,7 @@ static int export_dmabuf_common(struct hl_ctx *ctx,
}
hl_dmabuf->ctx = ctx;
- hl_ctx_get(hdev, hl_dmabuf->ctx);
+ hl_ctx_get(hl_dmabuf->ctx);
*dmabuf_fd = fd;
diff --git a/drivers/misc/habanalabs/common/mmu/mmu.c b/drivers/misc/habanalabs/common/mmu/mmu.c
index ae9b4923c32b..f3734718d94f 100644
--- a/drivers/misc/habanalabs/common/mmu/mmu.c
+++ b/drivers/misc/habanalabs/common/mmu/mmu.c
@@ -708,7 +708,7 @@ int hl_mmu_prefetch_cache_range(struct hl_ctx *ctx, u32 flags, u32 asid, u64 va,
* as actual prefetch is done in a WQ we must get the context (and put it
* at the end of the work function)
*/
- hl_ctx_get(ctx->hdev, ctx);
+ hl_ctx_get(ctx);
queue_work(ctx->hdev->pf_wq, &handle_pf_work->pf_work);
return 0;
--
2.25.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH 5/5] habanalabs: use separate structure info for each error collect data
2022-05-20 14:33 [PATCH 1/5] habanalabs: order memory manager messages Oded Gabbay
` (2 preceding siblings ...)
2022-05-20 14:33 ` [PATCH 4/5] habanalabs: fix missing handle shift during mmap Oded Gabbay
@ 2022-05-20 14:33 ` Oded Gabbay
3 siblings, 0 replies; 5+ messages in thread
From: Oded Gabbay @ 2022-05-20 14:33 UTC (permalink / raw)
To: linux-kernel; +Cc: Tal Cohen
From: Tal Cohen <talcohen@habana.ai>
Create separate info structure for each error type.
The structures shall be used inside the large structure that contains
the last session error.
This is more scalable for adding more errors in the future.
Signed-off-by: Tal Cohen <talcohen@habana.ai>
Reviewed-by: Oded Gabbay <ogabbay@kernel.org>
Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
---
.../habanalabs/common/command_submission.c | 7 +-
drivers/misc/habanalabs/common/habanalabs.h | 71 +++++++++++--------
.../misc/habanalabs/common/habanalabs_drv.c | 4 +-
.../misc/habanalabs/common/habanalabs_ioctl.c | 18 ++---
drivers/misc/habanalabs/gaudi/gaudi.c | 15 ++--
5 files changed, 63 insertions(+), 52 deletions(-)
diff --git a/drivers/misc/habanalabs/common/command_submission.c b/drivers/misc/habanalabs/common/command_submission.c
index d8b6a2a4b210..fb30b7de4aab 100644
--- a/drivers/misc/habanalabs/common/command_submission.c
+++ b/drivers/misc/habanalabs/common/command_submission.c
@@ -735,11 +735,10 @@ static void cs_timedout(struct work_struct *work)
hdev = cs->ctx->hdev;
/* Save only the first CS timeout parameters */
- rc = atomic_cmpxchg(&hdev->last_error.cs_write_disable, 0, 1);
+ rc = atomic_cmpxchg(&hdev->last_error.cs_timeout.write_disable, 0, 1);
if (!rc) {
- hdev->last_error.open_dev_timestamp = hdev->last_successful_open_ktime;
- hdev->last_error.cs_timeout_timestamp = ktime_get();
- hdev->last_error.cs_timeout_seq = cs->sequence;
+ hdev->last_error.cs_timeout.timestamp = ktime_get();
+ hdev->last_error.cs_timeout.seq = cs->sequence;
}
switch (cs->type) {
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index 99cbed5ef0e6..b0b0f3f89865 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -2568,37 +2568,50 @@ struct hl_clk_throttle {
};
/**
- * struct last_error_session_info - info about last session in which CS timeout or
- * razwi error occurred.
- * @open_dev_timestamp: device open timestamp.
- * @cs_timeout_timestamp: CS timeout timestamp.
- * @razwi_timestamp: razwi timestamp.
- * @cs_write_disable: if set writing to CS parameters in the structure is disabled so the
- * first (root cause) CS timeout will not be overwritten.
- * @razwi_write_disable: if set writing to razwi parameters in the structure is disabled so the
- * first (root cause) razwi will not be overwritten.
- * @cs_timeout_seq: CS timeout sequence number.
- * @razwi_addr: address that caused razwi.
- * @razwi_engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does
- * not have engine id it will be set to U16_MAX.
- * @razwi_engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible
- * engines which one them caused the razwi. In that case, it will contain the
- * second possible engine id, otherwise it will be set to U16_MAX.
- * @razwi_non_engine_initiator: in case the initiator of the razwi does not have engine id.
- * @razwi_type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.
+ * struct cs_timeout_info - info of last CS timeout occurred.
+ * @timestamp: CS timeout timestamp.
+ * @write_disable: if set writing to CS parameters in the structure is disabled so,
+ * the first (root cause) CS timeout will not be overwritten.
+ * @seq: CS timeout sequence number.
+ */
+struct cs_timeout_info {
+ ktime_t timestamp;
+ atomic_t write_disable;
+ u64 seq;
+};
+
+/**
+ * struct razwi_info - info about last razwi error occurred.
+ * @timestamp: razwi timestamp.
+ * @write_disable: if set writing to razwi parameters in the structure is disabled so the
+ * first (root cause) razwi will not be overwritten.
+ * @addr: address that caused razwi.
+ * @engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does
+ * not have engine id it will be set to U16_MAX.
+ * @engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible
+ * engines which one them caused the razwi. In that case, it will contain the
+ * second possible engine id, otherwise it will be set to U16_MAX.
+ * @non_engine_initiator: in case the initiator of the razwi does not have engine id.
+ * @type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.
+ */
+struct razwi_info {
+ ktime_t timestamp;
+ atomic_t write_disable;
+ u64 addr;
+ u16 engine_id_1;
+ u16 engine_id_2;
+ u8 non_engine_initiator;
+ u8 type;
+};
+
+/**
+ * struct last_error_session_info - info about last session errors occurred.
+ * @cs_timeout: CS timeout error last information.
+ * @razwi: razwi last information.
*/
struct last_error_session_info {
- ktime_t open_dev_timestamp;
- ktime_t cs_timeout_timestamp;
- ktime_t razwi_timestamp;
- atomic_t cs_write_disable;
- atomic_t razwi_write_disable;
- u64 cs_timeout_seq;
- u64 razwi_addr;
- u16 razwi_engine_id_1;
- u16 razwi_engine_id_2;
- u8 razwi_non_engine_initiator;
- u8 razwi_type;
+ struct cs_timeout_info cs_timeout;
+ struct razwi_info razwi;
};
/**
diff --git a/drivers/misc/habanalabs/common/habanalabs_drv.c b/drivers/misc/habanalabs/common/habanalabs_drv.c
index 9ead0927208d..37edb69a7255 100644
--- a/drivers/misc/habanalabs/common/habanalabs_drv.c
+++ b/drivers/misc/habanalabs/common/habanalabs_drv.c
@@ -195,8 +195,8 @@ int hl_device_open(struct inode *inode, struct file *filp)
hl_debugfs_add_file(hpriv);
- atomic_set(&hdev->last_error.cs_write_disable, 0);
- atomic_set(&hdev->last_error.razwi_write_disable, 0);
+ atomic_set(&hdev->last_error.cs_timeout.write_disable, 0);
+ atomic_set(&hdev->last_error.razwi.write_disable, 0);
hdev->open_counter++;
hdev->last_successful_open_jif = jiffies;
diff --git a/drivers/misc/habanalabs/common/habanalabs_ioctl.c b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
index 8fd2b427863f..c7864d6bb0a1 100644
--- a/drivers/misc/habanalabs/common/habanalabs_ioctl.c
+++ b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
@@ -569,7 +569,7 @@ static int last_err_open_dev_info(struct hl_fpriv *hpriv, struct hl_info_args *a
if ((!max_size) || (!out))
return -EINVAL;
- info.timestamp = ktime_to_ns(hdev->last_error.open_dev_timestamp);
+ info.timestamp = ktime_to_ns(hdev->last_successful_open_ktime);
return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;
}
@@ -584,8 +584,8 @@ static int cs_timeout_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
if ((!max_size) || (!out))
return -EINVAL;
- info.seq = hdev->last_error.cs_timeout_seq;
- info.timestamp = ktime_to_ns(hdev->last_error.cs_timeout_timestamp);
+ info.seq = hdev->last_error.cs_timeout.seq;
+ info.timestamp = ktime_to_ns(hdev->last_error.cs_timeout.timestamp);
return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;
}
@@ -600,12 +600,12 @@ static int razwi_info(struct hl_fpriv *hpriv, struct hl_info_args *args)
if ((!max_size) || (!out))
return -EINVAL;
- info.timestamp = ktime_to_ns(hdev->last_error.razwi_timestamp);
- info.addr = hdev->last_error.razwi_addr;
- info.engine_id_1 = hdev->last_error.razwi_engine_id_1;
- info.engine_id_2 = hdev->last_error.razwi_engine_id_2;
- info.no_engine_id = hdev->last_error.razwi_non_engine_initiator;
- info.error_type = hdev->last_error.razwi_type;
+ info.timestamp = ktime_to_ns(hdev->last_error.razwi.timestamp);
+ info.addr = hdev->last_error.razwi.addr;
+ info.engine_id_1 = hdev->last_error.razwi.engine_id_1;
+ info.engine_id_2 = hdev->last_error.razwi.engine_id_2;
+ info.no_engine_id = hdev->last_error.razwi.non_engine_initiator;
+ info.error_type = hdev->last_error.razwi.type;
return copy_to_user(out, &info, min_t(size_t, max_size, sizeof(info))) ? -EFAULT : 0;
}
diff --git a/drivers/misc/habanalabs/gaudi/gaudi.c b/drivers/misc/habanalabs/gaudi/gaudi.c
index 96a83317b302..fba322241096 100644
--- a/drivers/misc/habanalabs/gaudi/gaudi.c
+++ b/drivers/misc/habanalabs/gaudi/gaudi.c
@@ -7404,19 +7404,18 @@ static void gaudi_print_irq_info(struct hl_device *hdev, u16 event_type,
gaudi_print_and_get_mmu_error_info(hdev, &razwi_addr, &razwi_type);
/* In case it's the first razwi, save its parameters*/
- rc = atomic_cmpxchg(&hdev->last_error.razwi_write_disable, 0, 1);
+ rc = atomic_cmpxchg(&hdev->last_error.razwi.write_disable, 0, 1);
if (!rc) {
- hdev->last_error.open_dev_timestamp = hdev->last_successful_open_ktime;
- hdev->last_error.razwi_timestamp = ktime_get();
- hdev->last_error.razwi_addr = razwi_addr;
- hdev->last_error.razwi_engine_id_1 = engine_id_1;
- hdev->last_error.razwi_engine_id_2 = engine_id_2;
+ hdev->last_error.razwi.timestamp = ktime_get();
+ hdev->last_error.razwi.addr = razwi_addr;
+ hdev->last_error.razwi.engine_id_1 = engine_id_1;
+ hdev->last_error.razwi.engine_id_2 = engine_id_2;
/*
* If first engine id holds non valid value the razwi initiator
* does not have engine id
*/
- hdev->last_error.razwi_non_engine_initiator = (engine_id_1 == U16_MAX);
- hdev->last_error.razwi_type = razwi_type;
+ hdev->last_error.razwi.non_engine_initiator = (engine_id_1 == U16_MAX);
+ hdev->last_error.razwi.type = razwi_type;
}
}
--
2.25.1
^ permalink raw reply related [flat|nested] 5+ messages in thread