<html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body>
<div class="moz-cite-prefix">On 2023-09-13 4:07, Christian König
wrote:<br>
</div>
<blockquote type="cite" cite="mid:303c2bbb-865c-d5da-1418-21dc803f61a3@gmail.com">
[+Fleix]<br>
<br>
Well that looks like quite a serious bug.<br>
<br>
If I'm not completely mistaken the KFD work item tries to restore
the process by moving BOs into memory even after the suspend
freeze. Normally work items are frozen together with the user
space processes unless explicitly marked as not freezable.<br>
<br>
That this causes problem during the first eviction phase is just
the tip of the iceberg here. If a BO is moved into invisible
memory during this we wouldn't be able to get it out of that in
the second phase because SDMA and hw is already turned off.<br>
<br>
@Felix any idea how that can happen? Have you guys marked a work
item / work queue as not freezable?</blockquote>
<p>We don't set anything to non-freezable in KFD.</p>
<p><br>
</p>
<p>Regards,<br>
Felix<br>
</p>
<p><br>
</p>
<blockquote type="cite" cite="mid:303c2bbb-865c-d5da-1418-21dc803f61a3@gmail.com"> Or
maybe the display guys?<br>
<br>
@Xinhui please investigate what work item that is and where that
is coming from. Something like "if (adev->in_suspend)
dump_stack();" in the right place should probably do it.<br>
<br>
Thanks,<br>
Christian.<br>
<br>
<div class="moz-cite-prefix">Am 13.09.23 um 07:13 schrieb Pan,
Xinhui:<br>
</div>
<blockquote type="cite" cite="mid:DM4PR12MB516574733B83A0F3F8451A5887F0A@DM4PR12MB5165.namprd12.prod.outlook.com">
<style type="text/css" style="display:none;">P {margin-top:0;margin-bottom:0;}</style>
<p style="font-family:Arial;font-size:10pt;color:#0000FF;margin:5pt;font-style:normal;font-weight:normal;text-decoration:none;" align="Left"> [AMD Official Use Only - General]<br>
</p>
<br>
<div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof"> I notice that only user space
process are frozen on my side. kthread and workqueue keeps
running. Maybe some kernel configs are not enabled.</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof"> I made one module which just prints
something like i++ with mutex lock both in workqueue and
kthread. I paste some logs below.</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted0"> [438619.696196] XH: 14 from workqueue
<div class="ContentPasted0">[438619.700193] XH: 15 from
kthread</div>
<div class="ContentPasted0">[438620.394335] PM: suspend
entry (deep)</div>
<div class="ContentPasted0">[438620.399619] Filesystems
sync: 0.001 seconds</div>
<div class="ContentPasted0">[438620.403887] PM: Preparing
system for sleep (deep)</div>
<div class="ContentPasted0">[438620.409299] Freezing user
space processes</div>
<div class="ContentPasted0">[438620.414862] Freezing user
space processes completed (elapsed 0.001 seconds)</div>
<div class="ContentPasted0">[438620.421881] OOM killer
disabled.</div>
<div class="ContentPasted0">[438620.425197] Freezing
remaining freezable tasks</div>
<div class="ContentPasted0">[438620.430890] Freezing
remaining freezable tasks completed (elapsed 0.001
seconds)</div>
[438620.438348] PM: Suspending system (deep)<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof"> .....</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> [438623.746038] PM: suspend of devices
complete after 3303.137 msecs
<div class="ContentPasted1">[438623.752125] PM: start
suspend of devices complete after 3309.713 msecs</div>
<div class="ContentPasted1">[438623.758722] PM: suspend
debug: Waiting for 5 second(s).</div>
<div class="ContentPasted1">[438623.792166] XH: 22 from
kthread</div>
[438623.824140] XH: 23 from workqueue</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> So BOs definitely can be in use during
suspend.</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> Even if kthread or workqueue can be stopped
with one special kernel config. I think suspend can only
stop the workqueue with its callback finish. <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> otherwise something like below makes things
crazy.</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> LOCK BO</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> do something<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> -> schedule or wait, anycode might
sleep. Stopped by suspend now? no, i think.<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted1"> UNLOCK BO<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof"> <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof"> I do tests with cmds below.<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> echo devices > /sys/power/pm_test
<div class="ContentPasted2">echo 0 > /sys/power/pm_async</div>
<div class="ContentPasted2">echo 1 >
/sys/power/pm_print_times</div>
<div class="ContentPasted2">echo 1 >
/sys/power/pm_debug_messages</div>
echo 1 > /sys/module/amdgpu/parameters/debug_evictions</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2 ContentPasted3"> ./kfd.sh
--gtest_filter=KFDEvictTest.BasicTest</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2 ContentPasted3"> pm-suspend<br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> thanks</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> xinhui</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> <br>
</div>
<div style="font-family: Aptos, Aptos_EmbeddedFont,
Aptos_MSFontService, Calibri, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);" class="elementToProof
ContentPasted2"> <br>
</div>
<hr tabindex="-1" style="display:inline-block; width:98%">
<div id="divRplyFwdMsg" dir="ltr"><font style="font-size:
11pt; color: rgb(0, 0, 0);" face="Calibri, sans-serif"><b>发件人:</b>
Christian König <a class="moz-txt-link-rfc2396E" href="mailto:ckoenig.leichtzumerken@gmail.com" moz-do-not-send="true"><ckoenig.leichtzumerken@gmail.com></a><br>
<b>发送时间:</b> 2023年9月12日 17:01<br>
<b>收件人:</b> Pan, Xinhui <a class="moz-txt-link-rfc2396E" href="mailto:Xinhui.Pan@amd.com" moz-do-not-send="true"><Xinhui.Pan@amd.com></a>;
<a class="moz-txt-link-abbreviated moz-txt-link-freetext" href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a>
<a class="moz-txt-link-rfc2396E" href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true"><amd-gfx@lists.freedesktop.org></a><br>
<b>抄送:</b> Deucher, Alexander <a class="moz-txt-link-rfc2396E" href="mailto:Alexander.Deucher@amd.com" moz-do-not-send="true"><Alexander.Deucher@amd.com></a>;
Koenig, Christian <a class="moz-txt-link-rfc2396E" href="mailto:Christian.Koenig@amd.com" moz-do-not-send="true"><Christian.Koenig@amd.com></a>;
Fan, Shikang <a class="moz-txt-link-rfc2396E" href="mailto:Shikang.Fan@amd.com" moz-do-not-send="true"><Shikang.Fan@amd.com></a><br>
<b>主题:</b> Re: [PATCH] drm/amdgpu: Ignore first evction
failure during suspend</font>
<div> </div>
</div>
<div class="BodyFragment"><font size="2"><span style="font-size:11pt">
<div class="PlainText">When amdgpu_device_suspend() is
called processes should be frozen <br>
already. In other words KFD queues etc... should
already be idle.<br>
<br>
So when the eviction fails here we missed something
previously and that <br>
in turn can cause tons amount of problems.<br>
<br>
So ignoring those errors is most likely not a good
idea at all.<br>
<br>
Regards,<br>
Christian.<br>
<br>
Am 12.09.23 um 02:21 schrieb Pan, Xinhui:<br>
> [AMD Official Use Only - General]<br>
><br>
> Oh yep, Pinned BO is moved to other LRU list, So
eviction fails because of other reason.<br>
> I will change the comments in the patch.<br>
> The problem is eviction fails as many reasons,
say, BO is locked.<br>
> ASAIK, kfd will stop the queues and flush some
evict/restore work in its suspend callback. SO the
first eviction before kfd callback likely fails.<br>
><br>
> -----Original Message-----<br>
> From: Christian König <a class="moz-txt-link-rfc2396E" href="mailto:ckoenig.leichtzumerken@gmail.com" moz-do-not-send="true"><ckoenig.leichtzumerken@gmail.com></a><br>
> Sent: Friday, September 8, 2023 2:49 PM<br>
> To: Pan, Xinhui <a class="moz-txt-link-rfc2396E" href="mailto:Xinhui.Pan@amd.com" moz-do-not-send="true"><Xinhui.Pan@amd.com></a>;
<a class="moz-txt-link-abbreviated
moz-txt-link-freetext" href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a><br>
> Cc: Deucher, Alexander <a class="moz-txt-link-rfc2396E" href="mailto:Alexander.Deucher@amd.com" moz-do-not-send="true"><Alexander.Deucher@amd.com></a>;
Koenig, Christian <a class="moz-txt-link-rfc2396E" href="mailto:Christian.Koenig@amd.com" moz-do-not-send="true"><Christian.Koenig@amd.com></a>;
Fan, Shikang <a class="moz-txt-link-rfc2396E" href="mailto:Shikang.Fan@amd.com" moz-do-not-send="true"><Shikang.Fan@amd.com></a><br>
> Subject: Re: [PATCH] drm/amdgpu: Ignore first
evction failure during suspend<br>
><br>
> Am 08.09.23 um 05:39 schrieb xinhui pan:<br>
>> Some BOs might be pinned. So the first
eviction's failure will abort<br>
>> the suspend sequence. These pinned BOs will
be unpined afterwards<br>
>> during suspend.<br>
> That doesn't make much sense since pinned BOs
don't cause eviction failure here.<br>
><br>
> What exactly is the error code you see?<br>
><br>
> Christian.<br>
><br>
>> Actaully it has evicted most BOs, so that
should stil work fine in<br>
>> sriov full access mode.<br>
>><br>
>> Fixes: 47ea20762bb7 ("drm/amdgpu: Add an
extra evict_resource call<br>
>> during device_suspend.")<br>
>> Signed-off-by: xinhui pan <a class="moz-txt-link-rfc2396E" href="mailto:xinhui.pan@amd.com" moz-do-not-send="true"><xinhui.pan@amd.com></a><br>
>> ---<br>
>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
| 9 +++++----<br>
>> 1 file changed, 5 insertions(+), 4
deletions(-)<br>
>><br>
>> diff --git
a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br>
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br>
>> index 5c0e2b766026..39af526cdbbe 100644<br>
>> ---
a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br>
>> +++
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br>
>> @@ -4148,10 +4148,11 @@ int
amdgpu_device_suspend(struct drm_device<br>
>> *dev, bool fbcon)<br>
>><br>
>> adev->in_suspend = true;<br>
>><br>
>> - /* Evict the majority of BOs before
grabbing the full access */<br>
>> - r =
amdgpu_device_evict_resources(adev);<br>
>> - if (r)<br>
>> - return r;<br>
>> + /* Try to evict the majority of BOs
before grabbing the full access<br>
>> + * Ignore the ret val at first place as
we will unpin some BOs if any<br>
>> + * afterwards.<br>
>> + */<br>
>> +
(void)amdgpu_device_evict_resources(adev);<br>
>><br>
>> if (amdgpu_sriov_vf(adev)) {<br>
>>
amdgpu_virt_fini_data_exchange(adev);<br>
<br>
</div>
</span></font></div>
</div>
</blockquote>
<br>
</blockquote>
</body>
</html>