From: zhoucm1 Subject: Re: [PATCH] drm/amdgpu: increase timeout of IB test Date: Tue, 26 Jul 2016 15:40:22 +0800 Message-ID: <579713E6.2040805@amd.com> References: <1469512669-13954-1-git-send-email-David1.Zhou@amd.com> <7ab14e56-0a68-3435-f686-d5b20f9b7eae@vodafone.de> <5797112A.2000200@amd.com> <46261ccd-61dd-0fc5-ba9d-eaafb40da038@vodafone.de> Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="===============0396276130==" Return-path: In-Reply-To: <46261ccd-61dd-0fc5-ba9d-eaafb40da038-ANTagKRnAhcb1SvskN2V4Q@public.gmane.org> List-Id: Discussion list for AMD gfx List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: amd-gfx-bounces-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org Sender: "amd-gfx" To: =?UTF-8?B?Q2hyaXN0aWFuIEvDtm5pZw==?= , amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org --===============0396276130== Content-Type: multipart/alternative; boundary="------------010903070007040906040008" --------------010903070007040906040008 Content-Type: text/plain; charset="utf-8"; format=flowed Content-Transfer-Encoding: quoted-printable On 2016=E5=B9=B407=E6=9C=8826=E6=97=A5 15:32, Christian K=C3=B6nig wrote: > Ok, I really wasn't expecting this. How about 100ms? I tried it just now, 100ms isn't enough as well, see the append log. > > I just want to avoid that a reset takes more than 1 or 2 seconds even=20 > when it didn't worked. > > With 1 second timeout for each IB test we easily need 10+ seconds when=20 > the hardware doesn't response at all. This isn't true, when anyone of them timeout happens,=20 amdgpu_ib_ring_tests will return error. [ 59.286927] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx=20 timeout, last signaled seq=3D30397, last emitted seq=3D30399 [ 59.287253] [drm] IP block:1 is hang! [ 59.287262] [drm] IP block:5 is hang! [ 59.288375] [drm] Some block need full reset! [ 59.288385] pp_set_clockgating_state was not implemented. [ 59.343335] amdgpu 0000:03:00.0: GPU pci config reset [ 59.422762] amdgpu 0000:03:00.0: GPU reset succeeded, trying to resume [ 59.423971] [drm] PCIE GART of 4096M enabled (table at=20 0x0000000000040000). [ 59.433374] current thermal is out of range [ 59.436736] [drm] ring test on 0 succeeded in 14 usecs [ 59.437341] [drm] ring test on 1 succeeded in 27 usecs [ 59.437388] [drm] ring test on 2 succeeded in 23 usecs [ 59.437399] [drm] ring test on 3 succeeded in 5 usecs [ 59.437405] [drm] ring test on 4 succeeded in 2 usecs [ 59.437412] [drm] ring test on 5 succeeded in 2 usecs [ 59.437420] [drm] ring test on 6 succeeded in 3 usecs [ 59.437429] [drm] ring test on 7 succeeded in 3 usecs [ 59.437435] [drm] ring test on 8 succeeded in 2 usecs [ 59.437493] [drm] ring test on 9 succeeded in 6 usecs [ 59.437501] [drm] ring test on 10 succeeded in 6 usecs [ 59.464320] [drm] ring test on 11 succeeded in 2 usecs [ 59.464322] [drm] UVD initialized successfully. [ 59.564307] [drm] ring test on 12 succeeded in 14 usecs [ 59.564316] [drm] ring test on 13 succeeded in 3 usecs [ 59.564317] [drm] VCE initialized successfully. [ 59.662835] [drm:gfx_v8_0_ring_test_ib [amdgpu]] *ERROR* amdgpu: IB=20 test timed out. [ 59.663030] [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* amdgpu:=20 failed testing IB on GFX ring (-110). [ 59.663227] amdgpu 0000:03:00.0: ib ring test failed (-110). [ 59.663349] pp_set_clockgating_state was not implemented. [ 59.749358] amdgpu 0000:03:00.0: GPU pci config reset [ 59.830784] amdgpu 0000:03:00.0: GPU reset succeeded, trying to resume [ 59.831982] [drm] PCIE GART of 4096M enabled (table at=20 0x0000000000040000). [ 59.841412] current thermal is out of range [ 59.844763] [drm] ring test on 0 succeeded in 14 usecs [ 59.845341] [drm] ring test on 1 succeeded in 27 usecs [ 59.845385] [drm] ring test on 2 succeeded in 21 usecs [ 59.845397] [drm] ring test on 3 succeeded in 5 usecs [ 59.845406] [drm] ring test on 4 succeeded in 3 usecs [ 59.845412] [drm] ring test on 5 succeeded in 2 usecs [ 59.845421] [drm] ring test on 6 succeeded in 3 usecs [ 59.845429] [drm] ring test on 7 succeeded in 3 usecs [ 59.845435] [drm] ring test on 8 succeeded in 2 usecs [ 59.845493] [drm] ring test on 9 succeeded in 6 usecs [ 59.845500] [drm] ring test on 10 succeeded in 6 usecs [ 59.872362] [drm] ring test on 11 succeeded in 2 usecs [ 59.872363] [drm] UVD initialized successfully. root@zhoucm1-System-Product-Name:~# dmesg -c [ 59.971707] [drm] ring test on 12 succeeded in 14 usecs [ 59.971717] [drm] ring test on 13 succeeded in 3 usecs [ 59.971717] [drm] VCE initialized successfully. [ 60.070820] [drm:gfx_v8_0_ring_test_ib [amdgpu]] *ERROR* amdgpu: IB=20 test timed out. [ 60.071051] [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* amdgpu:=20 failed testing IB on GFX ring (-110). [ 60.071252] amdgpu 0000:03:00.0: ib ring test failed (-110). [ 60.071375] pp_set_clockgating_state was not implemented. [ 60.151397] amdgpu 0000:03:00.0: GPU pci config reset [ 60.230844] amdgpu 0000:03:00.0: GPU reset succeeded, trying to resume [ 60.232044] [drm] PCIE GART of 4096M enabled (table at=20 0x0000000000040000). [ 60.241412] current thermal is out of range [ 60.244746] [drm] ring test on 0 succeeded in 13 usecs [ 60.245322] [drm] ring test on 1 succeeded in 27 usecs [ 60.245368] [drm] ring test on 2 succeeded in 23 usecs [ 60.245380] [drm] ring test on 3 succeeded in 5 usecs [ 60.245386] [drm] ring test on 4 succeeded in 2 usecs [ 60.245393] [drm] ring test on 5 succeeded in 2 usecs [ 60.245402] [drm] ring test on 6 succeeded in 3 usecs > > Regards, > Christian. > > Am 26.07.2016 um 09:28 schrieb zhoucm1: >> CQE has found timeout when they cherry-pick your timeout patch. >> I also found ib_test could be timeout after gpu reset. >> 1s maybe too long for a simple testing command, but I just think that=20 >> doesn't matter to judge timeout. >> >> Regards, >> David >> >> On 2016=E5=B9=B407=E6=9C=8826=E6=97=A5 15:24, Christian K=C3=B6nig wro= te: >>> Am 26.07.2016 um 07:57 schrieb Chunming Zhou: >>>> we should give enough time to IB test. >>>> >>>> Change-Id: I92bfbe9b3aa35083f41baed8663907abfa15c8e6 >>>> Signed-off-by: Chunming Zhou >>> >>> Do we really need more than 10ms for an IB test? A whole second=20 >>> sounds awful long when we need to do 10+ tests currently. >>> >>> Christian. >>> >>>> --- >>>> drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +- >>>> 1 file changed, 1 insertion(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c=20 >>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c >>>> index 050062e..a31d7ef 100644 >>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c >>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c >>>> @@ -33,7 +33,7 @@ >>>> #include "amdgpu.h" >>>> #include "atom.h" >>>> -#define AMDGPU_IB_TEST_TIMEOUT msecs_to_jiffies(10) >>>> +#define AMDGPU_IB_TEST_TIMEOUT msecs_to_jiffies(1000) >>>> /* >>>> * IB >>> >>> >> > --------------010903070007040906040008 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable
On 2016=E5=B9=B407=E6=9C=8826=E6=97=A5= 15:32, Christian K=C3=B6nig wrote:
Ok, I really wasn't expecting this. How about 100ms?
I tried it just now, 100ms isn't enough as well, see the append log.<= br>

I just want to avoid that a reset takes more than 1 or 2 seconds even when it didn't worked.

With 1 second timeout for each IB test we easily need 10+ seconds when the hardware doesn't response at all.
This isn't true, when anyone of them timeout happens, amdgpu_ib_ring_tests will return error.

[=C2=A0=C2=A0 59.286927] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* r= ing gfx timeout, last signaled seq=3D30397, last emitted seq=3D30399
[=C2=A0=C2=A0 59.287253] [drm] IP block:1 is hang!
[=C2=A0=C2=A0 59.287262] [drm] IP block:5 is hang!
[=C2=A0=C2=A0 59.288375] [drm] Some block need full reset!
[=C2=A0=C2=A0 59.288385] pp_set_clockgating_state was not implemented= .
[=C2=A0=C2=A0 59.343335] amdgpu 0000:03:00.0: GPU pci config reset [=C2=A0=C2=A0 59.422762] amdgpu 0000:03:00.0: GPU reset succeeded, tr= ying to resume
[=C2=A0=C2=A0 59.423971] [drm] PCIE GART of 4096M enabled (table at 0x0000000000040000).
[=C2=A0=C2=A0 59.433374] current thermal is out of range
[=C2=A0=C2=A0 59.436736] [drm] ring test on 0 succeeded in 14 usecs [=C2=A0=C2=A0 59.437341] [drm] ring test on 1 succeeded in 27 usecs [=C2=A0=C2=A0 59.437388] [drm] ring test on 2 succeeded in 23 usecs [=C2=A0=C2=A0 59.437399] [drm] ring test on 3 succeeded in 5 usecs [=C2=A0=C2=A0 59.437405] [drm] ring test on 4 succeeded in 2 usecs [=C2=A0=C2=A0 59.437412] [drm] ring test on 5 succeeded in 2 usecs [=C2=A0=C2=A0 59.437420] [drm] ring test on 6 succeeded in 3 usecs [=C2=A0=C2=A0 59.437429] [drm] ring test on 7 succeeded in 3 usecs [=C2=A0=C2=A0 59.437435] [drm] ring test on 8 succeeded in 2 usecs [=C2=A0=C2=A0 59.437493] [drm] ring test on 9 succeeded in 6 usecs [=C2=A0=C2=A0 59.437501] [drm] ring test on 10 succeeded in 6 usecs [=C2=A0=C2=A0 59.464320] [drm] ring test on 11 succeeded in 2 usecs [=C2=A0=C2=A0 59.464322] [drm] UVD initialized successfully.
[=C2=A0=C2=A0 59.564307] [drm] ring test on 12 succeeded in 14 usecs<= br> [=C2=A0=C2=A0 59.564316] [drm] ring test on 13 succeeded in 3 usecs [=C2=A0=C2=A0 59.564317] [drm] VCE initialized successfully.
[=C2=A0=C2=A0 59.662835] [drm:gfx_v8_0_ring_t= est_ib [amdgpu]] *ERROR* amdgpu: IB test timed out.
[=C2=A0=C2=A0 59.663030] [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR= * amdgpu: failed testing IB on GFX ring (-110).

[=C2=A0=C2=A0 59.663227] amdgpu 0000:03:00.0: ib ring test failed (-1= 10).
[=C2=A0=C2=A0 59.663349] pp_set_clockgating_state was not implemented= .
[=C2=A0=C2=A0 59.749358] amdgpu 0000:03:00.0: GPU pci config reset [=C2=A0=C2=A0 59.830784] amdgpu 0000:03:00.0: GPU reset succeeded, tr= ying to resume
[=C2=A0=C2=A0 59.831982] [drm] PCIE GART of 4096M enabled (table at 0x0000000000040000).
[=C2=A0=C2=A0 59.841412] current thermal is out of range
[=C2=A0=C2=A0 59.844763] [drm] ring test on 0 succeeded in 14 usecs [=C2=A0=C2=A0 59.845341] [drm] ring test on 1 succeeded in 27 usecs [=C2=A0=C2=A0 59.845385] [drm] ring test on 2 succeeded in 21 usecs [=C2=A0=C2=A0 59.845397] [drm] ring test on 3 succeeded in 5 usecs [=C2=A0=C2=A0 59.845406] [drm] ring test on 4 succeeded in 3 usecs [=C2=A0=C2=A0 59.845412] [drm] ring test on 5 succeeded in 2 usecs [=C2=A0=C2=A0 59.845421] [drm] ring test on 6 succeeded in 3 usecs [=C2=A0=C2=A0 59.845429] [drm] ring test on 7 succeeded in 3 usecs [=C2=A0=C2=A0 59.845435] [drm] ring test on 8 succeeded in 2 usecs [=C2=A0=C2=A0 59.845493] [drm] ring test on 9 succeeded in 6 usecs [=C2=A0=C2=A0 59.845500] [drm] ring test on 10 succeeded in 6 usecs [=C2=A0=C2=A0 59.872362] [drm] ring test on 11 succeeded in 2 usecs [=C2=A0=C2=A0 59.872363] [drm] UVD initialized successfully.
root@zhoucm1-System-Product-Name:~# dmesg -c
[=C2=A0=C2=A0 59.971707] [drm] ring test on 12 succeeded in 14 usecs<= br> [=C2=A0=C2=A0 59.971717] [drm] ring test on 13 succeeded in 3 usecs [=C2=A0=C2=A0 59.971717] [drm] VCE initialized successfully.
[=C2=A0=C2=A0 60.070820] [drm:gfx_v8_0_ring_t= est_ib [amdgpu]] *ERROR* amdgpu: IB test timed out.
[=C2=A0=C2=A0 60.071051] [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR= * amdgpu: failed testing IB on GFX ring (-110).

[=C2=A0=C2=A0 60.071252] amdgpu 0000:03:00.0: ib ring test failed (-1= 10).
[=C2=A0=C2=A0 60.071375] pp_set_clockgating_state was not implemented= .
[=C2=A0=C2=A0 60.151397] amdgpu 0000:03:00.0: GPU pci config reset [=C2=A0=C2=A0 60.230844] amdgpu 0000:03:00.0: GPU reset succeeded, tr= ying to resume
[=C2=A0=C2=A0 60.232044] [drm] PCIE GART of 4096M enabled (table at 0x0000000000040000).
[=C2=A0=C2=A0 60.241412] current thermal is out of range
[=C2=A0=C2=A0 60.244746] [drm] ring test on 0 succeeded in 13 usecs [=C2=A0=C2=A0 60.245322] [drm] ring test on 1 succeeded in 27 usecs [=C2=A0=C2=A0 60.245368] [drm] ring test on 2 succeeded in 23 usecs [=C2=A0=C2=A0 60.245380] [drm] ring test on 3 succeeded in 5 usecs [=C2=A0=C2=A0 60.245386] [drm] ring test on 4 succeeded in 2 usecs [=C2=A0=C2=A0 60.245393] [drm] ring test on 5 succeeded in 2 usecs [=C2=A0=C2=A0 60.245402] [drm] ring test on 6 succeeded in 3 usecs

Regards,
Christian.

Am 26.07.2016 um 09:28 schrieb zhoucm1:
CQE has found timeout when they cherry-pick your timeout patch.
I also found ib_test could be timeout after gpu reset.
1s maybe too long for a simple testing command, but I just think that doesn't matter to judge timeout.

Regards,
David

On 2016=E5=B9=B407=E6=9C=8826=E6=97=A5 15:24, Christian K=C3=B6ni= g wrote:
Am 26.07.2016 um 07:57 schrieb Chunming Zhou:
we should give enough time to IB test= .

Change-Id: I92bfbe9b3aa35083f41baed8663907abfa15c8e6
Signed-off-by: Chunming Zhou <David1.Zhou-5C7GfCeVMHo@public.gmane.org>

Do we really need more than 10ms for an IB test? A whole second sounds awful long when we need to do 10+ tests currently.

Christian.

---
=C2=A0 drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +-
=C2=A0 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index 050062e..a31d7ef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -33,7 +33,7 @@
=C2=A0 #include "amdgpu.h"
=C2=A0 #include "atom.h"
=C2=A0 -#define AMDGPU_IB_TEST_TIMEOUT=C2=A0=C2=A0=C2=A0 msec= s_to_jiffies(10)
+#define AMDGPU_IB_TEST_TIMEOUT=C2=A0=C2=A0=C2=A0 msecs_to_ji= ffies(1000)
=C2=A0=C2=A0=C2=A0 /*
=C2=A0=C2=A0 * IB





--------------010903070007040906040008-- --===============0396276130== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: base64 Content-Disposition: inline X19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX19fX18KYW1kLWdmeCBt YWlsaW5nIGxpc3QKYW1kLWdmeEBsaXN0cy5mcmVlZGVza3RvcC5vcmcKaHR0cHM6Ly9saXN0cy5m cmVlZGVza3RvcC5vcmcvbWFpbG1hbi9saXN0aW5mby9hbWQtZ2Z4Cg== --===============0396276130==--