[Intel-gfx] [PATCH v2] drm/i915: Use trace_printk to provide a death rattle for GEM

Thu Nov 9 11:53:53 UTC 2017

On Thu, 2017-11-09 at 11:15 +0000, Chris Wilson wrote:
> Trying to enable printk debugging for GEM is fraught with the issue of
> spam; interactions with HW are very frequent and often boring. However,
> one instance where they are not so boring is just before a BUG; here
> ftrace provides a facility to dump its ringbuffer on an oops. So for CI
> let's enable trace_printk() to capture the last exchanges with HW as a
> death rattle.
> 
> For example,
> [   72.120722] ------------[ cut here ]------------
> [   72.120748] kernel BUG at drivers/gpu/drm/i915/intel_lrc.c:905!
> [   72.120756] invalid opcode: 0000 [#1] SMP
> [   72.120762] Dumping ftrace buffer:
> [   72.120768] ---------------------------------
> ...
> [   72.200424] gem_conc-1064    0..s1 71949306us : intel_lrc_irq_handler: bcs0 in:  ctx=4.1, seqno=161
> [   72.200469] gem_conc-1064    0..s1 71949312us : intel_lrc_irq_handler: bcs0 in:  ctx=6.2, seqno=160
> [   72.200512] gem_conc-1066    1..s1 71949325us : intel_lrc_irq_handler: bcs0 csb[4/4]: status=0x00008002
> [   72.200555] gem_conc-1066    1..s1 71949326us : intel_lrc_irq_handler: bcs0 out: ctx=6, seqno=160
> [   72.200601] gem_conc-1066    1..s. 71956923us : intel_lrc_irq_handler: bcs0 csb[5/5]: status=0x00000014
> [   72.200646] gem_conc-1066    1..s. 71956928us : intel_lrc_irq_handler: bcs0 out: ctx=6, seqno=160
> [   72.200690] gem_conc-1066    1..s. 71956939us : intel_lrc_irq_handler: bcs0 in:  ctx=8.1, seqno=164
> [   72.200736] gem_conc-1066    1..s. 71956940us : intel_lrc_irq_handler: bcs0 in:  ctx=4.2, seqno=162
> [   72.200780] gem_conc-1066    1..s. 71956951us : intel_lrc_irq_handler: bcs0 csb[0/0]: status=0x00008002
> [   72.200824] gem_conc-1066    1..s. 71956951us : intel_lrc_irq_handler: bcs0 out: ctx=4, seqno=162
> [   72.200867]   <idle>-0       1..s1 71967999us : intel_lrc_irq_handler: bcs0 csb[1/1]: status=0x00000014
> [   72.200912]   <idle>-0       1..s1 71968001us : intel_lrc_irq_handler: bcs0 out: ctx=4, seqno=162
> [   72.200956]   <idle>-0       1.Ns1 71979385us : intel_lrc_irq_handler: bcs0 csb[2/2]: status=0x00000018
> [   72.201001]   <idle>-0       1.Ns1 71979388us : intel_lrc_irq_handler: bcs0 out: ctx=8, seqno=164
> [   72.201044] gem_conc-1063    3..s1 72086825us : intel_lrc_irq_handler: bcs0 in:  ctx=7.1, seqno=165
> [   72.201088] gem_conc-1066    1..s. 72086918us : intel_lrc_irq_handler: bcs0 csb[3/3]: status=0x00000001
> [   72.201132] gem_conc-1066    1..s. 72086932us : intel_lrc_irq_handler: bcs0 in:  ctx=7.2, seqno=166
> [   72.201176] gem_conc-1066    1..s. 72086941us : intel_lrc_irq_handler: bcs0 csb[4/4]: status=0x00008002
> [   72.201219] gem_conc-1066    1..s. 72086941us : intel_lrc_irq_handler: bcs0 out: ctx=7, seqno=166
> [   72.201263] gem_conc-1066    1..s. 72103855us : intel_lrc_irq_handler: bcs0 csb[5/5]: status=0x00000018
> [   72.201307] gem_conc-1066    1..s. 72103858us : intel_lrc_irq_handler: bcs0 out: ctx=7, seqno=166
> [   72.201351] gem_conc-1064    0..s1 72116924us : intel_lrc_irq_handler: bcs0 in:  ctx=4.1, seqno=167
> [   72.201394] gem_conc-1064    0..s1 72116981us : intel_lrc_irq_handler: bcs0 in:  ctx=4.2, seqno=168
> [   72.201438] gem_conc-1066    1..s1 72117545us : intel_lrc_irq_handler: bcs0 csb[0/2]: status=0x00000012
> [   72.201482] gem_conc-1066    1..s1 72117547us : intel_lrc_irq_handler: bcs0 out: ctx=4, seqno=168
> [   72.201526] gem_conc-1066    1..s1 72117548us : intel_lrc_irq_handler: bcs0 csb[1/2]: status=0x00008002
> [   72.201570] gem_conc-1066    1..s1 72117548us : intel_lrc_irq_handler: bcs0 out: ctx=4, seqno=168
> [   72.201578] ---------------------------------

You could update this example trace for v2, too.

> 
> v2: Tweak the formatting to be more consistent between in/out.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>

<SNIP>

> +++ b/drivers/gpu/drm/i915/i915_gem.h
> @@ -44,6 +44,12 @@
>  #define GEM_DEBUG_BUG_ON(expr)
>  #endif
>  
> +#if IS_ENABLED(CONFIG_DRM_I915_TRACE_GEM)
> +#define GEM_TRACE(...) trace_printk(__VA_ARGS__)
> +#else
> +#define GEM_TRACE(...)

I guess we want one of them "do { } while(false)" tricks here not to
cause mayhem when disabled.

> @@ -860,6 +867,9 @@ static void intel_lrc_irq_handler(unsigned long data)
>  			 */
>  
>  			status = READ_ONCE(buf[2 * head]); /* maybe mmio! */
> +			GEM_TRACE("%s csb[%d/%d]: status=0x%08x:0x%08x\n",

csb[%d..%d] for better readability?

> @@ -887,6 +897,10 @@ static void intel_lrc_irq_handler(unsigned long data)
>  			GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id);
>  
>  			rq = port_unpack(port, &count);
> +			GEM_TRACE("%s out[0]: ctx=%d.%d, seqno=%x\n",
> +				  engine->name,
> +				  rq->ctx->hw_id, count,

Make up your mind, in above GEM_TRACE count is on its own line.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>

Looks like this could be an useful feature. I'm kinda wondering if the
actual tracepoint could be useful too, and a way to turn the other
tracepoints to the log as future steps.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation