[Intel-gfx] [PATCH] drm: Inject a cond_resched() into long drm_clflush_sg()

David Laight David.Laight at ACULAB.COM
Thu Jan 16 14:40:43 UTC 2020


> I'll do some measurements later this afternoon.

This is an Ivy bridge cpu, so clflush (not clflushopt).
With a cond_resched for every page I get:
(Note these calls are every 10 seconds....)

# tracer: function_graph
#
# CPU  DURATION                  FUNCTION CALLS
# |     |   |                     |   |   |   |
 0) # 3155.126 us |  drm_clflush_sg [drm]();
 1) # 3067.382 us |  drm_clflush_sg [drm]();
 2) # 3063.766 us |  drm_clflush_sg [drm]();
 3) # 3092.302 us |  drm_clflush_sg [drm]();
 0) # 3209.486 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 0)   kworker-7    =>  kworker-319
 ------------------------------------------

 0) # 3633.803 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 1)   kworker-7    =>  kworker-319
 ------------------------------------------

 1) # 3090.278 us |  drm_clflush_sg [drm]();
 2) # 3828.108 us |  drm_clflush_sg [drm]();
 3) # 3049.836 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 3)  kworker-319   =>   kworker-7
 ------------------------------------------

 3) # 3295.017 us |  drm_clflush_sg [drm]();
 3) # 3064.077 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 0)  kworker-319   =>   kworker-7
 ------------------------------------------

 0) # 3182.034 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 3)   kworker-7    =>  kworker-319
 ------------------------------------------

 3) # 3065.754 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 2)  kworker-319   =>   kworker-7
 ------------------------------------------

 2) # 3562.513 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 3)  kworker-319   =>   kworker-7
 ------------------------------------------

 3) # 3048.914 us |  drm_clflush_sg [drm]();
 3) # 3062.469 us |  drm_clflush_sg [drm]();
 3) # 3055.727 us |  drm_clflush_sg [drm]();
 ------------------------------------------
 0)   kworker-7    =>  kworker-319
 ------------------------------------------

Without the cond_sched I suspect more of them are 3.0ms.
Not really a significant difference.
I guess the longer times are the scheduler looking for work?
I don't understand the extra traces - I'm guessing they are process switch related.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)


More information about the Intel-gfx mailing list