hollalinux
/
linux-sg2042
mirror of https://github.com/sophgo/linux-riscv.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

habanalabs: add support for cs with timestamp 

add support for user to request a timestamp upon
cs completion.

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 2020-11-10 17:26:22 +02:00 committed by Oded Gabbay
parent 9d127ad571
commit bd2f477f20
3 changed files with 41 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
drivers/misc/habanalabs/common/command_submission.c
View File

@ -26,7 +26,7 @@ enum hl_cs_wait_status {
static void job_wq_completion(struct work_struct *work); static void job_wq_completion(struct work_struct *work);
static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
u64 timeout_us, u64 seq, u64 timeout_us, u64 seq,
enum hl_cs_wait_status *status); enum hl_cs_wait_status *status, s64 *timestamp);
static void cs_do_release(struct kref *ref); static void cs_do_release(struct kref *ref);
static void hl_sob_reset(struct kref *ref) static void hl_sob_reset(struct kref *ref)
@ -150,6 +150,7 @@ static void hl_fence_init(struct hl_fence *fence)
{ {
kref_init(&fence->refcount); kref_init(&fence->refcount);
fence->error = 0; fence->error = 0;
fence->timestamp = ktime_set(0, 0);
init_completion(&fence->completion); init_completion(&fence->completion);
} }
@ -404,6 +405,8 @@ out:
else if (!cs->submitted) else if (!cs->submitted)
cs->fence->error = -EBUSY; cs->fence->error = -EBUSY;
if (cs->timestamp)
cs->fence->timestamp = ktime_get();
complete_all(&cs->fence->completion); complete_all(&cs->fence->completion);
hl_fence_put(cs->fence); hl_fence_put(cs->fence);
@ -734,7 +737,8 @@ static int hl_cs_sanity_checks(struct hl_fpriv *hpriv, union hl_cs_args *args)
return -EBUSY; return -EBUSY;
} }
cs_type_flags = args->in.cs_flags & ~HL_CS_FLAGS_FORCE_RESTORE; cs_type_flags = args->in.cs_flags &
~(HL_CS_FLAGS_FORCE_RESTORE | HL_CS_FLAGS_TIMESTAMP);
if (unlikely(cs_type_flags && !is_power_of_2(cs_type_flags))) { if (unlikely(cs_type_flags && !is_power_of_2(cs_type_flags))) {
dev_err(hdev->dev, dev_err(hdev->dev,
@ -798,7 +802,7 @@ static int hl_cs_copy_chunk_array(struct hl_device *hdev,
} }
static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
u32 num_chunks, u64 *cs_seq) u32 num_chunks, u64 *cs_seq, bool timestamp)
{ {
bool int_queues_only = true; bool int_queues_only = true;
struct hl_device *hdev = hpriv->hdev; struct hl_device *hdev = hpriv->hdev;
@ -825,6 +829,7 @@ static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks,
goto free_cs_chunk_array; goto free_cs_chunk_array;
} }
cs->timestamp = !!timestamp;
*cs_seq = cs->sequence; *cs_seq = cs->sequence;
hl_debugfs_add_cs(cs); hl_debugfs_add_cs(cs);
@ -995,7 +1000,7 @@ static int hl_cs_ctx_switch(struct hl_fpriv *hpriv, union hl_cs_args *args,
rc = 0; rc = 0;
} else { } else {
rc = cs_ioctl_default(hpriv, chunks, num_chunks, rc = cs_ioctl_default(hpriv, chunks, num_chunks,
cs_seq); cs_seq, false);
} }
mutex_unlock(&hpriv->restore_phase_mutex); mutex_unlock(&hpriv->restore_phase_mutex);
@ -1013,7 +1018,7 @@ static int hl_cs_ctx_switch(struct hl_fpriv *hpriv, union hl_cs_args *args,
wait_again: wait_again:
ret = _hl_cs_wait_ioctl(hdev, ctx, ret = _hl_cs_wait_ioctl(hdev, ctx,
jiffies_to_usecs(hdev->timeout_jiffies), jiffies_to_usecs(hdev->timeout_jiffies),
*cs_seq, &status); *cs_seq, &status, NULL);
if (ret) { if (ret) {
if (ret == -ERESTARTSYS) { if (ret == -ERESTARTSYS) {
usleep_range(100, 200); usleep_range(100, 200);
@ -1154,7 +1159,7 @@ static int cs_ioctl_signal_wait_create_jobs(struct hl_device *hdev,
static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type, static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,
void __user *chunks, u32 num_chunks, void __user *chunks, u32 num_chunks,
u64 *cs_seq) u64 *cs_seq, bool timestamp)
{ {
struct hl_cs_chunk *cs_chunk_array, *chunk; struct hl_cs_chunk *cs_chunk_array, *chunk;
struct hw_queue_properties *hw_queue_prop; struct hw_queue_properties *hw_queue_prop;
@ -1259,6 +1264,8 @@ static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type,
goto free_cs_chunk_array; goto free_cs_chunk_array;
} }
cs->timestamp = !!timestamp;
/* /*
* Save the signal CS fence for later initialization right before * Save the signal CS fence for later initialization right before
* hanging the wait CS on the queue. * hanging the wait CS on the queue.
@ -1334,10 +1341,11 @@ int hl_cs_ioctl(struct hl_fpriv *hpriv, void *data)
case CS_TYPE_WAIT: case CS_TYPE_WAIT:
case CS_TYPE_COLLECTIVE_WAIT: case CS_TYPE_COLLECTIVE_WAIT:
rc = cs_ioctl_signal_wait(hpriv, cs_type, chunks, num_chunks, rc = cs_ioctl_signal_wait(hpriv, cs_type, chunks, num_chunks,
&cs_seq); &cs_seq, args->in.cs_flags & HL_CS_FLAGS_TIMESTAMP);
break; break;
default: default:
rc = cs_ioctl_default(hpriv, chunks, num_chunks, &cs_seq); rc = cs_ioctl_default(hpriv, chunks, num_chunks, &cs_seq,
args->in.cs_flags & HL_CS_FLAGS_TIMESTAMP);
break; break;
} }
@ -1353,13 +1361,16 @@ out:
static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
u64 timeout_us, u64 seq, u64 timeout_us, u64 seq,
enum hl_cs_wait_status *status) enum hl_cs_wait_status *status, s64 *timestamp)
{ {
struct hl_fence *fence; struct hl_fence *fence;
unsigned long timeout; unsigned long timeout;
int rc = 0; int rc = 0;
long completion_rc; long completion_rc;
if (timestamp)
*timestamp = 0;
if (timeout_us == MAX_SCHEDULE_TIMEOUT) if (timeout_us == MAX_SCHEDULE_TIMEOUT)
timeout = timeout_us; timeout = timeout_us;
else else
@ -1382,10 +1393,13 @@ static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx,
wait_for_completion_interruptible_timeout( wait_for_completion_interruptible_timeout(
&fence->completion, timeout); &fence->completion, timeout);
if (completion_rc > 0) if (completion_rc > 0) {
*status = CS_WAIT_STATUS_COMPLETED; *status = CS_WAIT_STATUS_COMPLETED;
else if (timestamp)
*timestamp = ktime_to_ns(fence->timestamp);
} else {
*status = CS_WAIT_STATUS_BUSY; *status = CS_WAIT_STATUS_BUSY;
}
if (fence->error == -ETIMEDOUT) if (fence->error == -ETIMEDOUT)
rc = -ETIMEDOUT; rc = -ETIMEDOUT;
@ -1411,10 +1425,11 @@ int hl_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data)
union hl_wait_cs_args *args = data; union hl_wait_cs_args *args = data;
enum hl_cs_wait_status status; enum hl_cs_wait_status status;
u64 seq = args->in.seq; u64 seq = args->in.seq;
s64 timestamp;
int rc; int rc;
rc = _hl_cs_wait_ioctl(hdev, hpriv->ctx, args->in.timeout_us, seq, rc = _hl_cs_wait_ioctl(hdev, hpriv->ctx, args->in.timeout_us, seq,
&status); &status, &timestamp);
memset(args, 0, sizeof(*args)); memset(args, 0, sizeof(*args));
@ -1439,6 +1454,11 @@ int hl_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data)
return rc; return rc;
} }
if (timestamp) {
args->out.flags |= HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD;
args->out.timestamp_nsec = timestamp;
}
switch (status) { switch (status) {
case CS_WAIT_STATUS_GONE: case CS_WAIT_STATUS_GONE:
args->out.flags |= HL_WAIT_CS_STATUS_FLAG_GONE; args->out.flags |= HL_WAIT_CS_STATUS_FLAG_GONE;

4
drivers/misc/habanalabs/common/habanalabs.h
View File

@ -481,12 +481,14 @@ struct asic_fixed_properties {
* @completion: fence is implemented using completion * @completion: fence is implemented using completion
* @refcount: refcount for this fence * @refcount: refcount for this fence
* @error: mark this fence with error * @error: mark this fence with error
* @timestamp: timestamp upon completion
* *
*/ */
struct hl_fence { struct hl_fence {
struct completion completion; struct completion completion;
struct kref refcount; struct kref refcount;
int error; int error;
ktime_t timestamp;
}; };
/** /**
@ -1127,6 +1129,7 @@ struct hl_userptr {
* @tdr_active: true if TDR was activated for this CS (to prevent * @tdr_active: true if TDR was activated for this CS (to prevent
* double TDR activation). * double TDR activation).
* @aborted: true if CS was aborted due to some device error. * @aborted: true if CS was aborted due to some device error.
* @timestamp: true if a timestmap must be captured upon completion
*/ */
struct hl_cs { struct hl_cs {
u16 *jobs_in_queue_cnt; u16 *jobs_in_queue_cnt;
@ -1147,6 +1150,7 @@ struct hl_cs {
u8 timedout; u8 timedout;
u8 tdr_active; u8 tdr_active;
u8 aborted; u8 aborted;
u8 timestamp;
}; };
/** /**

4
include/uapi/misc/habanalabs.h
View File

@ -592,6 +592,7 @@ struct hl_cs_chunk {
#define HL_CS_FLAGS_SIGNAL 0x2 #define HL_CS_FLAGS_SIGNAL 0x2
#define HL_CS_FLAGS_WAIT 0x4 #define HL_CS_FLAGS_WAIT 0x4
#define HL_CS_FLAGS_COLLECTIVE_WAIT 0x8 #define HL_CS_FLAGS_COLLECTIVE_WAIT 0x8
#define HL_CS_FLAGS_TIMESTAMP 0x20
#define HL_CS_STATUS_SUCCESS 0 #define HL_CS_STATUS_SUCCESS 0
@ -663,12 +664,15 @@ struct hl_wait_cs_in {
#define HL_WAIT_CS_STATUS_INTERRUPTED 4 #define HL_WAIT_CS_STATUS_INTERRUPTED 4
#define HL_WAIT_CS_STATUS_FLAG_GONE 0x1 #define HL_WAIT_CS_STATUS_FLAG_GONE 0x1
#define HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD 0x2
struct hl_wait_cs_out { struct hl_wait_cs_out {
/* HL_WAIT_CS_STATUS_* */ /* HL_WAIT_CS_STATUS_* */
__u32 status; __u32 status;
/* HL_WAIT_CS_STATUS_FLAG* */ /* HL_WAIT_CS_STATUS_FLAG* */
__u32 flags; __u32 flags;
/* valid only if HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD is set */
__s64 timestamp_nsec;
}; };
union hl_wait_cs_args { union hl_wait_cs_args {