<html><head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body> <p>Hey Christian, Denis, see bellow - <br> </p> <div class="moz-cite-prefix">On 2021-04-06 6:34 a.m., Christian König wrote:<br> </div> <blockquote type="cite" cite="mid:51d7873d-cf35-6be5-79c2-024937c67f6a@amd.com"> Hi Andrey,<br> <br> well good question. My job is to watch over the implementation and design and while I always help I can adjust anybodies schedule.<br> <br> Is the patch to print a warning when the hardware is accessed without holding the locks merged yet? If not then that would probably be a good starting point.<br> </blockquote> <p><br> </p> <p>It's merged into amd-staging-drm-next and since I work on drm-misc-next I will cherry-pick it into there.<br> </p> <p><br> </p> <blockquote type="cite" cite="mid:51d7873d-cf35-6be5-79c2-024937c67f6a@amd.com"> <br> Then we would need to unify this with the SRCU to make sure that we have both the reset lock as well as block the hotplug code from reusing the MMIO space.<br> </blockquote> <br> <p>In my understanding there is a significant difference between handling of GPU reset and unplug - while GPU reset use case requires any HW accessing code to block and wait for the reset to finish and then proceed, hot-unplug<br> is permanent and hence no need to wait and proceed but rather abort at once. This why I think that in any place we already check for device reset we should also add a check for hot-unplug but the handling would be different<br> in that for hot-unplug we would abort instead of keep waiting.</p> <p>Similar to handling device reset for unplug we obviously also need to stop and block any MMIO accesses once device is unplugged and, as Daniel Vetter mentioned - we have to do it before finishing pci_remove (early device fini)<br> and not later (when last device reference is dropped from user space) in order to prevent reuse of MMIO space we still access by other hot plugging devices. As in device reset case we need to cancel all delay works, stop drm schedule, complete all unfinished fences(both HW and scheduler fences). While you stated strong objection to force signalling scheduler fences from GPU reset, quote: <br> </p> <p>"you can't signal the dma_fence waiting. Waiting for a dma_fence also means you wait for the GPU reset to finish. When we would signal the dma_fence during the GPU reset then we would run into memory corruption because the hardware jobs running after the GPU reset would access memory which is already freed."<br> To my understating this is a key difference with hot-unplug, the device is gone, all those concerns are irrelevant and hence we can actually force signal scheduler fences (setting and error to them before) to force completion of any<br> waiting clients such as possibly IOCTLs or async page flips e.t.c.<br> </p> <p>Beyond blocking all delayed works and scheduler threads we also need to guarantee no IOCTL can access MMIO post device unplug OR in flight IOCTLs are done before we finish pci_remove (amdgpu_pci_remove for us).<br> For this I suggest we do something like what we worked on with Takashi Iwai the ALSA maintainer recently when he helped implementing PCI BARs move support for snd_hda_intel. Take a look at<br> <a class="moz-txt-link-freetext" href="https://cgit.freedesktop.org/~agrodzov/linux/commit/?h=yadro/pcie_hotplug/movable_bars_v9.1&id=cbaa324799718e2b828a8c7b5b001dd896748497">https://cgit.freedesktop.org/~agrodzov/linux/commit/?h=yadro/pcie_hotplug/movable_bars_v9.1&id=cbaa324799718e2b828a8c7b5b001dd896748497</a> and<br> <a class="moz-txt-link-freetext" href="https://cgit.freedesktop.org/~agrodzov/linux/commit/?h=yadro/pcie_hotplug/movable_bars_v9.1&id=e36365d9ab5bbc30bdc221ab4b3437de34492440">https://cgit.freedesktop.org/~agrodzov/linux/commit/?h=yadro/pcie_hotplug/movable_bars_v9.1&id=e36365d9ab5bbc30bdc221ab4b3437de34492440</a><br> We also had same issue there, how to prevent MMIO accesses while the BARs are migrating. What was done there is a refcount was added to count all IOCTLs in flight, for any in flight IOCTL the BAR migration handler would <br> block for the refcount to drop to 0 before it would proceed, for any later IOCTL it stops and wait if device is in migration state. We even don't need the wait part, nothing to wait for, we just return with -ENODEV for this case.</p> <p>The above approach should allow us to wait for all the IOCTLs in flight, together with stopping scheduler threads and cancelling and flushing all in flight work items and timers i think It should give as full solution for the hot-unplug case<br> of preventing any MMIO accesses post device pci_remove.<br> <br> Let me know what you think guys.</p> <p>Andrey</p> <p><br> </p> <blockquote type="cite" cite="mid:51d7873d-cf35-6be5-79c2-024937c67f6a@amd.com"> <br> And then testing, testing, testing to see if we have missed something.<br> <br> Christian.<br> <br> <div class="moz-cite-prefix">Am 05.04.21 um 19:58 schrieb Andrey Grodzovsky:<br> </div> <blockquote type="cite" cite="mid:1e37bb4d-f54d-1b7e-4632-94cfcf749528@amd.com"> <p>Denis, Christian, are there any updates in the plan on how to move on with this ? As you know I need very similar code for my up-streaming of device hot-unplug. My latest solution (<a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/archives/amd-gfx/2021-January/058606.html" moz-do-not-send="true">https://lists.freedesktop.org/archives/amd-gfx/2021-January/058606.html</a>) was not acceptable because of low level guards on the register accessors level which was hurting performance. Basically I need a way to prevent any MMIO write accesses from kernel driver after device is removed (UMD accesses are taken care of by page faulting dummy page). We are using now hot-unplug code for Freemont program and so up-streaming became more of a priority then before. This MMIO access issue is currently my main blocker from up-streaming. Is there any way I can assist in pushing this on ?</p> <p>Andrey <br> </p> <div class="moz-cite-prefix">On 2021-03-18 5:51 a.m., Christian König wrote:<br> </div> <blockquote type="cite" cite="mid:378fdffb-99b5-2a14-736d-a06f310b040c@amd.com"> Am 18.03.21 um 10:30 schrieb Li, Dennis:<br> <blockquote type="cite" cite="mid:DM5PR12MB253379E8C89D8A20C8A0245AED699@DM5PR12MB2533.namprd12.prod.outlook.com"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)">  <style>@font-face {font-family:"Cambria Math"; panose-1:2 4 5 3 5 4 6 3 2 4;}@font-face {font-family:等线; panose-1:2 1 6 0 3 1 1 1 1 1;}@font-face {font-family:Calibri; panose-1:2 15 5 2 2 2 4 3 2 4;}@font-face {font-family:"\@等线"; panose-1:2 1 6 0 3 1 1 1 1 1;}@font-face {font-family:"Segoe UI"; panose-1:2 11 5 2 4 2 4 2 2 3;}p.MsoNormal, li.MsoNormal, div.MsoNormal {margin:0in; font-size:11.0pt; font-family:"Calibri",sans-serif;}a:link, span.MsoHyperlink {mso-style-priority:99; color:#0563C1; text-decoration:underline;}span.EmailStyle19 {mso-style-type:personal-reply; font-family:"Calibri",sans-serif; color:windowtext;}.MsoChpDefault {mso-style-type:export-only; font-size:10.0pt;}div.WordSection1 {page:WordSection1;}</style> <div class="WordSection1"> <p class="MsoNormal">>>> The GPU reset doesn't complete the fences we wait for. It only completes the hardware fences as part of the reset.<o:p></o:p></p> <p class="MsoNormal">>>> So waiting for a fence while holding the reset lock is illegal and needs to be avoided.<o:p></o:p></p> <p class="MsoNormal">I understood your concern. It is more complex for DRM GFX, therefore I abandon adding lock protection for DRM ioctls now. Maybe we can try to add all kernel dma_fence waiting in a list, and signal all in recovery threads. Do you have same concern for compute cases? </p> </div> </blockquote> <br> Yes, compute (KFD) is even harder to handle.<br> <br> See you can't signal the dma_fence waiting. Waiting for a dma_fence also means you wait for the GPU reset to finish.<br> <br> When we would signal the dma_fence during the GPU reset then we would run into memory corruption because the hardware jobs running after the GPU reset would access memory which is already freed.<br> <br> <blockquote type="cite" cite="mid:DM5PR12MB253379E8C89D8A20C8A0245AED699@DM5PR12MB2533.namprd12.prod.outlook.com"> <div class="WordSection1"> <p class="MsoNormal"><o:p></o:p></p> <p class="MsoNormal"><o:p> </o:p></p> <p class="MsoNormal">>>> Lockdep also complains about this when it is used correctly. The only reason it doesn't complain here is because you use an atomic+wait_event instead of a locking primitive.<o:p></o:p></p> <p class="MsoNormal">Agree. This approach will escape the monitor of lockdep. Its goal is to block other threads when GPU recovery thread start. But I couldn’t find a better method to solve this problem. Do you have some suggestion? </p> </div> </blockquote> <br> Well, completely abandon those change here.<br> <br> What we need to do is to identify where hardware access happens and then insert taking the read side of the GPU reset lock so that we don't wait for a dma_fence or allocate memory, but still protect the hardware from concurrent access and reset.<br> <br> Regards,<br> Christian.<br> <br> <blockquote type="cite" cite="mid:DM5PR12MB253379E8C89D8A20C8A0245AED699@DM5PR12MB2533.namprd12.prod.outlook.com"> <div class="WordSection1"> <p class="MsoNormal"><o:p></o:p></p> <p class="MsoNormal"><o:p> </o:p></p> <p class="MsoNormal">Best Regards<o:p></o:p></p> <p class="MsoNormal">Dennis Li<o:p></o:p></p> <p class="MsoNormal"><o:p> </o:p></p> <div> <div style="border:none;border-top:solid #E1E1E1 1.0pt;padding:3.0pt 0in 0in 0in"> <p class="MsoNormal"><b>From:</b> Koenig, Christian <a class="moz-txt-link-rfc2396E" href="mailto:Christian.Koenig@amd.com" moz-do-not-send="true"><Christian.Koenig@amd.com></a> <br> <b>Sent:</b> Thursday, March 18, 2021 4:59 PM<br> <b>To:</b> Li, Dennis <a class="moz-txt-link-rfc2396E" href="mailto:Dennis.Li@amd.com" moz-do-not-send="true"><Dennis.Li@amd.com></a>; <a class="moz-txt-link-abbreviated" href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a>; Deucher, Alexander <a class="moz-txt-link-rfc2396E" href="mailto:Alexander.Deucher@amd.com" moz-do-not-send="true"><Alexander.Deucher@amd.com></a>; Kuehling, Felix <a class="moz-txt-link-rfc2396E" href="mailto:Felix.Kuehling@amd.com" moz-do-not-send="true"><Felix.Kuehling@amd.com></a>; Zhang, Hawking <a class="moz-txt-link-rfc2396E" href="mailto:Hawking.Zhang@amd.com" moz-do-not-send="true"><Hawking.Zhang@amd.com></a><br> <b>Subject:</b> AW: [PATCH 0/4] Refine GPU recovery sequence to enhance its stability<o:p></o:p></p> </div> </div> <p class="MsoNormal"><o:p> </o:p></p> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">Exactly that's what you don't seem to understand.<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black"><o:p> </o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">The GPU reset doesn't complete the fences we wait for. It only completes the hardware fences as part of the reset.<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black"><o:p> </o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">So waiting for a fence while holding the reset lock is illegal and needs to be avoided.<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black"><o:p> </o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">Lockdep also complains about this when it is used correctly. The only reason it doesn't complain here is because you use an atomic+wait_event instead of a locking primitive.<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black"><o:p> </o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">Regards,<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><span style="font-size:12.0pt;font-family:"Segoe UI",sans-serif;color:black">Christian.<o:p></o:p></span></p> </div> <div> <p class="MsoNormal"><o:p> </o:p></p> </div> <div class="MsoNormal" style="text-align:center" align="center"> <hr width="98%" size="2" align="center"> </div> <div id="divRplyFwdMsg"> <p class="MsoNormal"><b><span style="color:black">Von:</span></b><span style="color:black"> Li, Dennis <<a href="mailto:Dennis.Li@amd.com" moz-do-not-send="true">Dennis.Li@amd.com</a>><br> <b>Gesendet:</b> Donnerstag, 18. März 2021 09:28<br> <b>An:</b> Koenig, Christian <<a href="mailto:Christian.Koenig@amd.com" moz-do-not-send="true">Christian.Koenig@amd.com</a>>; <a href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a> <<a href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a>>; Deucher, Alexander <<a href="mailto:Alexander.Deucher@amd.com" moz-do-not-send="true">Alexander.Deucher@amd.com</a>>; Kuehling, Felix <<a href="mailto:Felix.Kuehling@amd.com" moz-do-not-send="true">Felix.Kuehling@amd.com</a>>; Zhang, Hawking <<a href="mailto:Hawking.Zhang@amd.com" moz-do-not-send="true">Hawking.Zhang@amd.com</a>><br> <b>Betreff:</b> RE: [PATCH 0/4] Refine GPU recovery sequence to enhance its stability</span> <o:p></o:p></p> <div> <p class="MsoNormal"> <o:p></o:p></p> </div> </div> <div> <div> <p class="MsoNormal">>>> Those two steps need to be exchanged or otherwise it is possible that new delayed work items etc are started before the lock is taken.<br> What about adding check for adev->in_gpu_reset in work item? If exchange the two steps, it maybe introduce the deadlock. For example, the user thread hold the read lock and waiting for the fence, if recovery thread try to hold write lock and then complete fences, in this case, recovery thread will always be blocked.<o:p></o:p></p> </div> <div> <p class="MsoNormal" style="margin-bottom:12.0pt"><br> Best Regards<br> Dennis Li<br> -----Original Message-----<br> From: Koenig, Christian <<a href="mailto:Christian.Koenig@amd.com" moz-do-not-send="true">Christian.Koenig@amd.com</a>> <br> Sent: Thursday, March 18, 2021 3:54 PM<br> To: Li, Dennis <<a href="mailto:Dennis.Li@amd.com" moz-do-not-send="true">Dennis.Li@amd.com</a>>; <a href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true"> amd-gfx@lists.freedesktop.org</a>; Deucher, Alexander <<a href="mailto:Alexander.Deucher@amd.com" moz-do-not-send="true">Alexander.Deucher@amd.com</a>>; Kuehling, Felix <<a href="mailto:Felix.Kuehling@amd.com" moz-do-not-send="true">Felix.Kuehling@amd.com</a>>; Zhang, Hawking <<a href="mailto:Hawking.Zhang@amd.com" moz-do-not-send="true">Hawking.Zhang@amd.com</a>><br> Subject: Re: [PATCH 0/4] Refine GPU recovery sequence to enhance its stability<br> <br> Am 18.03.21 um 08:23 schrieb Dennis Li:<br> > We have defined two variables in_gpu_reset and reset_sem in adev object. The atomic type variable in_gpu_reset is used to avoid recovery thread reenter and make lower functions return more earlier when recovery start, but couldn't block recovery thread when it access hardware. The r/w semaphore reset_sem is used to solve these synchronization issues between recovery thread and other threads.<br> ><br> > The original solution locked registers' access in lower functions, which will introduce following issues:<br> ><br> > 1) many lower functions are used in both recovery thread and others. Firstly we must harvest these functions, it is easy to miss someones. Secondly these functions need select which lock (read lock or write lock) will be used, according to the thread it is running in. If the thread context isn't considered, the added lock will easily introduce deadlock. Besides that, in most time, developer easily forget to add locks for new functions.<br> ><br> > 2) performance drop. More lower functions are more frequently called.<br> ><br> > 3) easily introduce false positive lockdep complaint, because write lock has big range in recovery thread, but low level functions will hold read lock may be protected by other locks in other threads.<br> ><br> > Therefore the new solution will try to add lock protection for ioctls of kfd. Its goal is that there are no threads except for recovery thread or its children (for xgmi) to access hardware when doing GPU reset and resume. So refine recovery thread as the following:<br> ><br> > Step 0: atomic_cmpxchg(&adev->in_gpu_reset, 0, 1)<br> > 1). if failed, it means system had a recovery thread running, current thread exit directly;<br> > 2). if success, enter recovery thread;<br> ><br> > Step 1: cancel all delay works, stop drm schedule, complete all unreceived fences and so on. It try to stop or pause other threads.<br> ><br> > Step 2: call down_write(&adev->reset_sem) to hold write lock, which will block recovery thread until other threads release read locks.<br> <br> Those two steps need to be exchanged or otherwise it is possible that new delayed work items etc are started before the lock is taken.<br> <br> Just to make it clear until this is fixed the whole patch set is a NAK.<br> <br> Regards,<br> Christian.<br> <br> ><br> > Step 3: normally, there is only recovery threads running to access hardware, it is safe to do gpu reset now.<br> ><br> > Step 4: do post gpu reset, such as call all ips' resume functions;<br> ><br> > Step 5: atomic set adev->in_gpu_reset as 0, wake up other threads and release write lock. Recovery thread exit normally.<br> ><br> > Other threads call the amdgpu_read_lock to synchronize with recovery thread. If it finds that in_gpu_reset is 1, it should release read lock if it has holden one, and then blocks itself to wait for recovery finished event. If thread successfully hold read lock and in_gpu_reset is 0, it continues. It will exit normally or be stopped by recovery thread in step 1.<br> ><br> > Dennis Li (4):<br> > drm/amdgpu: remove reset lock from low level functions<br> > drm/amdgpu: refine the GPU recovery sequence<br> > drm/amdgpu: instead of using down/up_read directly<br> > drm/amdkfd: add reset lock protection for kfd entry functions<br> ><br> > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 6 +<br> > drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 14 +-<br> > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 173 +++++++++++++-----<br> > .../gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c | 8 -<br> > drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 4 +-<br> > drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 9 +-<br> > drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 5 +-<br> > drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 5 +-<br> > drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 172 ++++++++++++++++-<br> > drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 +-<br> > drivers/gpu/drm/amd/amdkfd/kfd_process.c | 4 +<br> > .../amd/amdkfd/kfd_process_queue_manager.c | 17 ++<br> > 12 files changed, 345 insertions(+), 75 deletions(-)<br> ><o:p></o:p></p> </div> </div> </div> </blockquote> <br> <br> <fieldset class="mimeAttachmentHeader"></fieldset> <pre class="moz-quote-pre" wrap="">_______________________________________________ amd-gfx mailing list <a class="moz-txt-link-abbreviated" href="mailto:amd-gfx@lists.freedesktop.org" moz-do-not-send="true">amd-gfx@lists.freedesktop.org</a> <a class="moz-txt-link-freetext" href="https://lists.freedesktop.org/mailman/listinfo/amd-gfx" moz-do-not-send="true">https://lists.freedesktop.org/mailman/listinfo/amd-gfx</a> </pre> </blockquote> </blockquote> <br> </blockquote> </body> </html>