korg/linux
0
0
Fork 0
mirror of https://mirrors.bfsu.edu.cn/git/linux.git synced 2024-09-21 12:11:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

accel/habanalabs: gradual sleep in polling memory macro

Browse Source 
It’s better to avoid long sleeps right from the beginning of the polling
since the data may be available much sooner than the sleep period.
Because polling host memory is inexpensive, this change gradually
increases the sleep time up to the user-requested period.

Signed-off-by: Didi Freiman <dfreiman@habana.ai>
Reviewed-by: Ofir Bitton <obitton@habana.ai>
Signed-off-by: Ofir Bitton <obitton@habana.ai>
This commit is contained in:
Didi Freiman 2024-05-07 13:47:12 +03:00 committed by Ofir Bitton
parent 0199e6392e
commit 9dec27bb8a
1 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
drivers/accel/habanalabs/common/habanalabs.h
View File

@ -2729,11 +2729,16 @@ void hl_wreg(struct hl_device *hdev, u32 reg, u32 val);
* updated directly by the device. If false, the host memory being polled will
* be updated by host CPU. Required so host knows whether or not the memory
* might need to be byte-swapped before returning value to caller.
*
* On the first 4 polling iterations the macro goes to sleep for short period of
* time that gradually increases and reaches sleep_us on the fifth iteration.
*/
#define hl_poll_timeout_memory(hdev, addr, val, cond, sleep_us, timeout_us, \
mem_written_by_device) \
({ \
u64 __sleep_step_us; \
ktime_t __timeout; \
u8 __step = 8; \
\
__timeout = ktime_add_us(ktime_get(), timeout_us); \
might_sleep_if(sleep_us); \
@ -2751,8 +2756,10 @@ void hl_wreg(struct hl_device *hdev, u32 reg, u32 val);
(val) = le32_to_cpu(*(__le32 *) &(val)); \
break; \
} \
if (sleep_us) \
usleep_range((sleep_us >> 2) + 1, sleep_us); \
__sleep_step_us = sleep_us >> __step; \
if (__sleep_step_us) \
usleep_range((__sleep_step_us >> 2) + 1, __sleep_step_us); \
__step >>= 1; \
} \
(cond) ? 0 : -ETIMEDOUT; \
})