<html><head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body> <div class="moz-cite-prefix">On 2023-09-11 22:52, Lang Yu wrote: </div> <blockquote type="cite" cite="mid:ZP%2FSdUd9am%2Ff2WJw@lang-desktop"> <pre class="moz-quote-pre" wrap="">On 09/11/ , Harish Kasiviswanathan wrote: </pre> <blockquote type="cite"> <pre class="moz-quote-pre" wrap="">Heavy-weight TLB flush is required after unmap on all GPUs for correctness and security. Signed-off-by: Harish Kasiviswanathan <a class="moz-txt-link-rfc2396E" href="mailto:Harish.Kasiviswanathan@amd.com"><Harish.Kasiviswanathan@amd.com></a> --- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index b315311dfe2a..b9950074aee0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -1466,8 +1466,7 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type); static inline bool kfd_flush_tlb_after_unmap(struct kfd_dev *dev) { - return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 3) || - KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2) || + return KFD_GC_VERSION(dev) > IP_VERSION(9, 4, 2) || (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 1) && dev->sdma_fw_version >= 18) || KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 0); } </pre> </blockquote> <pre class="moz-quote-pre" wrap=""> 1, If TLB_FLUSH_HEAVYWEIGHT is required after unmap on all GPUs as described in commmit message, why we have this whitelist instead of a blacklist?</pre> </blockquote> That was a bug that this patch is fixing. There were specific GPUs and firmware versions where the TLB flush after unmap was causing intermittent problems in specific tests. This should have always been a blacklist. <blockquote type="cite" cite="mid:ZP%2FSdUd9am%2Ff2WJw@lang-desktop"> <pre class="moz-quote-pre" wrap=""> 2, kfd_flush_tlb(pdd, TLB_FLUSH_HEAVYWEIGHT) is also called in svm_range_unmap_from_gpus(). Why not add this whitelist there?</pre> </blockquote> There was a patch that used kfd_flush_tlb_after_unmap in the SVM code. But you reverted that patch, probably because it caused more problems than it solved. SVM really must flush TLBs the way it does because it is so tightly integrated with Linux's virtual memory management and because with XNACK, memory can be unmapped while GPU work is in progress without preemting queues (implicitly flushing TLBs and caches): <pre>commit 515d7cebc2e2d2b4f0a276d26f3b790a83cdfe06 Author: Lang Yu <a class="moz-txt-link-rfc2396E" href="mailto:Lang.Yu@amd.com"><Lang.Yu@amd.com></a> Date: Wed Apr 20 10:24:31 2022 +0800 Revert "drm/amdkfd: only allow heavy-weight TLB flush on some ASICs for SVM too" This reverts commit 36bf93216ecbe399c40c5e0486f0f0e3a4afa69e. It causes SVM regressions on Vega10 with XNACK-ON. Just revert it at the moment. ./kfdtest --gtest_filter=KFDSVMRangeTest.MigratePolicyTest Signed-off-by: Lang Yu <a class="moz-txt-link-rfc2396E" href="mailto:Lang.Yu@amd.com"><Lang.Yu@amd.com></a> Reviewed-by: Philip Yang<a class="moz-txt-link-rfc2396E" href="mailto:Philip.Yang@amd.com"><Philip.Yang@amd.com></a> Signed-off-by: Alex Deucher <a class="moz-txt-link-rfc2396E" href="mailto:alexander.deucher@amd.com"><alexander.deucher@amd.com></a> </pre> Regards, Felix <blockquote type="cite" cite="mid:ZP%2FSdUd9am%2Ff2WJw@lang-desktop"> <pre class="moz-quote-pre" wrap=""> Regards, Lang </pre> <blockquote type="cite"> <pre class="moz-quote-pre" wrap="">-- 2.34.1 </pre> </blockquote> </blockquote> </body> </html>