Skip to contentSkip to conversation

tests/gem_reset_stats : mask off ring_stop bits

PATCH In review [email protected] 3 participants 6 messages Jun 3, 2015 – Jun 15, 2015

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tests/gem_reset_stats : mask off ring_stop bits

From: Tim Gore <[email protected]>

Function check_gpu_ok checks to make sure that any hangs have cleared by testing for (flags == 0). Some tests set the STOP_RINGS_ALLOW_BAN and STOP_RINGS_ALLOW_ERRORS flags but these do not get cleared by an individual ring reset, (a feature added recently to the driver), leading the check_gpu_ok function to think that the gpu is still hung.

So I mask the flags with STOP_RING_ALL, to ignore the mode bits and look only at the bits that stop the rings.

Once gpu_check_ok sees that the gpu is not hung I write 0 to stop_rings in order to clear it completely. This is because igt_set_stop_rings will only write to stop_rings if either a) they are currently 0 or b) we are writing 0. If we leave the mode bits set then subsequent calls to igt_set_stop_rings to create hangs will fail.

---
 tests/gem_reset_stats.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
tests/gem_reset_stats.c +7 -1
index 0e9db10..2bb4291 100644
@@ -1033,7 +1033,7 @@ static void check_gpu_ok(void)
 	igt_debug("checking gpu state\n");
 
 	while (retry_count--) {
-		flags = igt_get_stop_rings();
+		flags = igt_get_stop_rings() & STOP_RING_ALL;
 		if (flags == 0)
 			break;
 
@@ -1043,6 +1043,12 @@ static void check_gpu_ok(void)
 
 	igt_assert(flags == 0);
 
+	/*
+	 * Clear the _ALLOW_ERRORS and _ALLOW_BAN flags;
+	 * these are not cleared by individual ring reset.
+	 */
+	igt_set_stop_rings(0);
+
 	fd = drm_open_any();
 	gem_quiescent_gpu(fd);
 	close(fd);
Signature · 6 lines
2.3.0

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 12 lines · On Wed, Jun 03, 2015 at 09:20:21AM +0100, [email protected] wrote:

From: Tim Gore <[email protected]>

Function check_gpu_ok checks to make sure that any hangs have cleared by testing for (flags == 0). Some tests set the STOP_RINGS_ALLOW_BAN and STOP_RINGS_ALLOW_ERRORS flags but these do not get cleared by an individual ring reset, (a feature added recently to the driver), leading the check_gpu_ok function to think that the gpu is still hung.

So I mask the flags with STOP_RING_ALL, to ignore the mode bits and look only at the bits that stop the rings.

Quoted 6 lines

Once gpu_check_ok sees that the gpu is not hung I write 0 to stop_rings in order to clear it completely. This is because igt_set_stop_rings will only write to stop_rings if either a) they are currently 0 or b) we are writing 0. If we leave the mode bits set then subsequent calls to igt_set_stop_rings to create hangs will fail.

Can we please just deprecate the stop_rings interface? We can do explicit hang injection and GPU resets on gen4+, most of gen3 but not gen2. Even if we mask of testing for gen2/3, that still provides (almost, just a couple of gen2/3 reset functions will be missed) complete coverage of GEM reset handling.

The benefit is we lose this attrocious interface and remove some hideous complications from the kernel.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#2 lore raw
Quoted 35 lines

-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Wednesday, June 03, 2015 9:30 AM
To: Gore, Tim
Cc: [email protected]; Wood, Thomas
Subject: Re: [Intel-gfx] [PATCH i-g-t] tests/gem_reset_stats : mask off ring_stop bits

Quoted 11 lines · On Wed, Jun 03, 2015 at 09:20:21AM +0100, [email protected] wrote:

From: Tim Gore <[email protected]>

Function check_gpu_ok checks to make sure that any hangs have cleared by testing for (flags == 0). Some tests set the STOP_RINGS_ALLOW_BAN and STOP_RINGS_ALLOW_ERRORS flags but these do not get cleared by an individual ring reset, (a feature added recently to the driver), leading the check_gpu_ok function to think that the gpu is still hung.

So I mask the flags with STOP_RING_ALL, to ignore the mode bits and look only at the bits that stop the rings.

Quoted 6 lines

Once gpu_check_ok sees that the gpu is not hung I write 0 to stop_rings in order to clear it completely. This is because igt_set_stop_rings will only write to stop_rings if either a) they are currently 0 or b) we are writing 0.
If we leave the mode bits set then subsequent calls to igt_set_stop_rings to create hangs will fail.

Can we please just deprecate the stop_rings interface? We can do explicit hang injection and GPU resets on gen4+, most of gen3 but not gen2. Even if we mask of testing for gen2/3, that still provides (almost, just a couple of gen2/3 reset functions will be missed) complete coverage of GEM reset handling.

Quoted 6 lines

The benefit is we lose this attrocious interface and remove some hideous complications from the kernel.
-Chris

--
Chris Wilson, Intel Open Source Technology Centre

Sounds good to me. I think most people here (Egham) see the stop rings interface as Being a rather nasty hack, and it certainly makes pain for the TDR mechanism.

Tim
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

author
#3 lore raw
Quoted 53 lines

-----Original Message-----
From: Intel-gfx [mailto:[email protected]] On Behalf Of Gore, Tim
Sent: Wednesday, June 03, 2015 9:43 AM
To: Chris Wilson
Cc: [email protected]; Wood, Thomas
Subject: Re: [Intel-gfx] [PATCH i-g-t] tests/gem_reset_stats : mask off ring_stop bits

Quoted 8 lines

-----Original Message-----
From: Chris Wilson [mailto:[email protected]]
Sent: Wednesday, June 03, 2015 9:30 AM
To: Gore, Tim
Cc: [email protected]; Wood, Thomas
Subject: Re: [Intel-gfx] [PATCH i-g-t] tests/gem_reset_stats : mask off ring_stop bits

Quoted 6 lines
Quoted 5 lines · On Wed, Jun 03, 2015 at 09:20:21AM +0100, [email protected] wrote:

From: Tim Gore <[email protected]>

Function check_gpu_ok checks to make sure that any hangs have cleared by testing for (flags == 0). Some tests set the STOP_RINGS_ALLOW_BAN and STOP_RINGS_ALLOW_ERRORS flags but

these do

Quoted 2 lines
Quoted 2 lines

not get cleared by an individual ring reset, (a feature added recently to the driver), leading the check_gpu_ok function to think that

the gpu is still hung.

Quoted 22 lines
Quoted 10 lines

So I mask the flags with STOP_RING_ALL, to ignore the mode bits and look only at the bits that stop the rings.

Once gpu_check_ok sees that the gpu is not hung I write 0 to stop_rings in order to clear it completely. This is because igt_set_stop_rings will only write to stop_rings if either a) they are currently 0 or b) we are writing 0.
If we leave the mode bits set then subsequent calls to igt_set_stop_rings to create hangs will fail.

Can we please just deprecate the stop_rings interface? We can do explicit hang injection and GPU resets on gen4+, most of gen3 but not gen2. Even if we mask of testing for gen2/3, that still provides (almost, just a couple of
gen2/3 reset functions will be missed) complete coverage of GEM reset handling.

The benefit is we lose this attrocious interface and remove some hideous complications from the kernel.
-Chris

Quoted 8 lines
Quoted 2 lines

--
Chris Wilson, Intel Open Source Technology Centre

Sounds good to me. I think most people here (Egham) see the stop rings interface as Being a rather nasty hack, and it certainly makes pain for the TDR mechanism.

Tim

However it would still be good to get this fix in for now. I'll start looking At removing the use of stop_rings from IGT when I've finished my current Crop of HSD's, so if someone could push this patch it would be appreciated. One less test failure to be reported :-)

Tim

Quoted 4 lines

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw
Quoted 4 lines · Tim

However it would still be good to get this fix in for now. I'll start looking At removing the use of stop_rings from IGT when I've finished my current Crop of HSD's, so if someone could push this patch it would be appreciated. One less test failure to be reported :-)

Sorry, got sidetracked by my loathing of stop_rings. Patch lgtm indeed.

Pushed, thanks.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#5 lore raw
In reply to Chris Wilson
Quoted 20 lines · Chris Wilson
Quoted 12 lines · On Wed, Jun 03, 2015 at 09:20:21AM +0100, [email protected] wrote:

From: Tim Gore <[email protected]>

Function check_gpu_ok checks to make sure that any hangs have cleared by testing for (flags == 0). Some tests set the STOP_RINGS_ALLOW_BAN and STOP_RINGS_ALLOW_ERRORS flags but these do not get cleared by an individual ring reset, (a feature added recently to the driver), leading the check_gpu_ok function to think that the gpu is still hung.

So I mask the flags with STOP_RING_ALL, to ignore the mode bits and look only at the bits that stop the rings.

Quoted 6 lines

Once gpu_check_ok sees that the gpu is not hung I write 0 to stop_rings in order to clear it completely. This is because igt_set_stop_rings will only write to stop_rings if either a) they are currently 0 or b) we are writing 0. If we leave the mode bits set then subsequent calls to igt_set_stop_rings to create hangs will fail.

Quoted 5 lines

Can we please just deprecate the stop_rings interface? We can do explicit hang injection and GPU resets on gen4+, most of gen3 but not gen2. Even if we mask of testing for gen2/3, that still provides (almost, just a couple of gen2/3 reset functions will be missed) complete coverage of GEM reset handling.

Yeah I guess it's ok to nuke stop_rings. The only issue I see is that stop_rings provided a neat way for the kernel to support gpu resets on platforms where it's not working. I guess we'd need to move that knowledge to igt with an require_gpu_reset macro and skip the tests correctly. Otherwise I don't see an issue here really.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx