<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body style="word-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;" class=""> <br class=""> <div><br class=""> <blockquote type="cite" class=""> <div class="">On Apr 16, 2022, at 12:43 AM, Andrey Grodzovsky <<a href="mailto:andrey.grodzovsky@amd.com" class="">andrey.grodzovsky@amd.com</a>> wrote:</div> <br class="Apple-interchange-newline"> <div class=""> <div class=""> <p class=""><br class=""> </p> <div class="moz-cite-prefix">On 2022-04-15 06:12, Shuotao Xu wrote:<br class=""> </div> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> Hi Andrey, <div class=""><br class=""> </div> <div class="">First I really appreciate the discussion! It helped me understand the driver code greatly. Thank you so much:)</div> <div class="">Please see my inline comments. <br class=""> <div class=""><br class=""> <blockquote type="cite" class=""> <div class="">On Apr 14, 2022, at 11:13 PM, Andrey Grodzovsky <<a href="mailto:andrey.grodzovsky@amd.com" class="moz-txt-link-freetext" moz-do-not-send="true">andrey.grodzovsky@amd.com</a>> wrote:</div> <br class="Apple-interchange-newline"> <div class=""> <div class=""> <p class=""><br class=""> </p> <div class="moz-cite-prefix">On 2022-04-14 10:00, Shuotao Xu wrote:<br class=""> </div> <blockquote type="cite" cite="mid:B7970589-ACF6-41F3-8622-1C0F705F3EE0@microsoft.com" class=""> <br class=""> <div class=""><br class=""> <blockquote type="cite" class=""> <div class="">On Apr 14, 2022, at 1:31 AM, Andrey Grodzovsky <<a href="mailto:andrey.grodzovsky@amd.com" class="moz-txt-link-freetext" moz-do-not-send="true">andrey.grodzovsky@amd.com</a>> wrote:</div> <br class="Apple-interchange-newline"> <div class=""> <div class=""> <p class=""><br class=""> </p> <div class="moz-cite-prefix">On 2022-04-13 12:03, Shuotao Xu wrote:<br class=""> </div> <blockquote type="cite" cite="mid:5A64FAEA-CCE8-4EB6-8E7B-852D4F384255@microsoft.com" class=""> <br class=""> <div class=""><br class=""> <blockquote type="cite" class=""> <div class="">On Apr 11, 2022, at 11:52 PM, Andrey Grodzovsky <<a href="mailto:andrey.grodzovsky@amd.com" class="moz-txt-link-freetext" moz-do-not-send="true">andrey.grodzovsky@amd.com</a>> wrote:</div> <br class="Apple-interchange-newline"> <div class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">[Some people who received this message don't often get email from<span class="Apple-converted-space"> </span></span><a href="mailto:andrey.grodzovsky@amd.com" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px;" class="moz-txt-link-freetext" moz-do-not-send="true">andrey.grodzovsky@amd.com</a><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">. Learn why this is important at<span class="Apple-converted-space"> </span></span><a href="http://aka.ms/LearnAboutSenderIdentification" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px;" class="moz-txt-link-freetext" moz-do-not-send="true">http://aka.ms/LearnAboutSenderIdentification</a><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">.]</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">On 2022-04-08 21:28, Shuotao Xu wrote:</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <blockquote type="cite" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br class=""> <blockquote type="cite" class="">On Apr 8, 2022, at 11:28 PM, Andrey Grodzovsky <<a href="mailto:andrey.grodzovsky@amd.com" class="moz-txt-link-freetext" moz-do-not-send="true">andrey.grodzovsky@amd.com</a>> wrote:<br class=""> <br class=""> [Some people who received this message don't often get email from <a href="mailto:andrey.grodzovsky@amd.com" class="moz-txt-link-freetext" moz-do-not-send="true"> andrey.grodzovsky@amd.com</a>. Learn why this is important at <a href="http://aka.ms/LearnAboutSenderIdentification" class="moz-txt-link-freetext" moz-do-not-send="true"> http://aka.ms/LearnAboutSenderIdentification</a>.]<br class=""> <br class=""> On 2022-04-08 04:45, Shuotao Xu wrote:<br class=""> <blockquote type="cite" class="">Adding PCIe Hotplug Support for AMDKFD: the support of hot-plug of GPU<br class=""> devices can open doors for many advanced applications in data center<br class=""> in the next few years, such as for GPU resource<br class=""> disaggregation. Current AMDKFD does not support hotplug out b/o the<br class=""> following reasons:<br class=""> <br class=""> 1. During PCIe removal, decrement KFD lock which was incremented at<br class=""> the beginning of hw fini; otherwise kfd_open later is going to<br class=""> fail.<br class=""> </blockquote> I assumed you read my comment last time, still you do same approach.<br class=""> More in details bellow<br class=""> </blockquote> Aha, I like your fix:) I was not familiar with drm APIs so just only half understood your comment last time.<br class=""> <br class=""> BTW, I tried hot-plugging out a GPU when rocm application is still running.<br class=""> From dmesg, application is still trying to access the removed kfd device, and are met with some errors.<br class=""> </blockquote> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">Application us supposed to keep running, it holds the drm_device</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">reference as long as it has an open</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">FD to the device and final cleanup will come only after the app will die</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">thus releasing the FD and the last</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">drm_device reference.</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <blockquote type="cite" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> Application would hang and not exiting in this case.<br class=""> </blockquote> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> </div> </blockquote> <div class=""><br class=""> </div> Actually I tried kill -7 $pid, and the process exists. The dmesg has some warning though.</div> <div class=""><br class=""> </div> <div class=""> <div class="">[ 711.769977] WARNING: CPU: 23 PID: 344 at .../amdgpu-rocm5.0.2/src/amd/amdgpu/amdgpu_object.c:1336 amdgpu_bo_release_notify+0x150/0x160 [amdgpu]</div> <div class="">[ 711.770528] Modules linked in: amdgpu(OE) amdttm(OE) amd_sched(OE) amdkcl(OE) iommu_v2 nf_conntrack_netlink nfnetlink xfrm_user xfrm_algo xt_addrtype br_netfilter xt_CHECKSUM iptable_mangle xt_MASQUERADE iptable_nat nf_nat xt_conntrack nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ipt_REJECT nf_reject_ipv4 xt_tcpudp bridge stp llc ebtable_filter ebtables ip6table_filter ip6_tables iptable_filter overlay binfmt_misc intel_rapl_msr i40iw intel_rapl_common skx_edac nfit x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel rpcrdma kvm sunrpc ipmi_ssif ib_umad ib_ipoib rdma_ucm irqbypass rapl joydev acpi_ipmi input_leds intel_cstate ipmi_si ipmi_devintf mei_me mei intel_pch_thermal ipmi_msghandler ioatdma mac_hid lpc_ich dca acpi_power_meter acpi_pad sch_fq_codel ib_iser rdma_cm iw_cm ib_cm iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi pci_stub ip_tables x_tables autofs4 btrfs blake2b_generic zstd_compress raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor</div> <div class="">[ 711.779359] raid6_pq libcrc32c raid1 raid0 multipath linear mlx5_ib ib_uverbs ib_core ast drm_vram_helper i2c_algo_bit drm_ttm_helper ttm drm_kms_helper syscopyarea crct10dif_pclmul crc32_pclmul ghash_clmulni_intel sysfillrect uas hid_generic sysimgblt aesni_intel mlx5_core fb_sys_fops crypto_simd usbhid cryptd drm i40e pci_hyperv_intf usb_storage glue_helper mlxfw hid ahci libahci wmi</div> <div class="">[ 711.779752] CPU: 23 PID: 344 Comm: kworker/23:1 Tainted: G W OE 5.11.0+ #1</div> <div class="">[ 711.779755] Hardware name: Supermicro SYS-4029GP-TRT2/X11DPG-OT-CPU, BIOS 2.1 08/14/2018</div> <div class="">[ 711.779756] Workqueue: kfd_process_wq kfd_process_wq_release [amdgpu]</div> <div class="">[ 711.779955] RIP: 0010:amdgpu_bo_release_notify+0x150/0x160 [amdgpu]</div> <div class="">[ 711.780141] Code: e8 b5 af 34 f4 e9 1f ff ff ff 48 39 c2 74 07 0f 0b e9 69 ff ff ff 4c 89 e7 e8 3c b4 16 00 e9 5c ff ff ff e8 a2 ce fd f3 eb cf <0f> 0b eb cb e8 d7 02 34 f4 0f 1f 80 00 00 00 00 0f 1f 44 00 00 55</div> <div class="">[ 711.780143] RSP: 0018:ffffa8100dd67c30 EFLAGS: 00010282</div> <div class="">[ 711.780145] RAX: 00000000ffffffea RBX: ffff89980e792058 RCX: 0000000000000000</div> <div class="">[ 711.780147] RDX: 0000000000000000 RSI: ffff89a8f9ad8870 RDI: ffff89a8f9ad8870</div> <div class="">[ 711.780148] RBP: ffffa8100dd67c50 R08: 0000000000000000 R09: fffffffffff99b18</div> <div class="">[ 711.780149] R10: ffffa8100dd67bd0 R11: ffffa8100dd67908 R12: ffff89980e792000</div> <div class="">[ 711.780151] R13: ffff89980e792058 R14: ffff89980e7921bc R15: dead000000000100</div> <div class="">[ 711.780152] FS: 0000000000000000(0000) GS:ffff89a8f9ac0000(0000) knlGS:0000000000000000</div> <div class="">[ 711.780154] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033</div> <div class="">[ 711.780156] CR2: 00007ffddac6f71f CR3: 00000030bb80a003 CR4: 00000000007706e0</div> <div class="">[ 711.780157] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000</div> <div class="">[ 711.780159] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400</div> <div class="">[ 711.780160] PKRU: 55555554</div> <div class="">[ 711.780161] Call Trace:</div> <div class="">[ 711.780163] ttm_bo_release+0x2ae/0x320 [amdttm]</div> <div class="">[ 711.780169] amdttm_bo_put+0x30/0x40 [amdttm]</div> <div class="">[ 711.780357] amdgpu_bo_unref+0x1e/0x30 [amdgpu]</div> <div class="">[ 711.780543] amdgpu_gem_object_free+0x8b/0x160 [amdgpu]</div> <div class="">[ 711.781119] drm_gem_object_free+0x1d/0x30 [drm]</div> <div class="">[ 711.781489] amdgpu_amdkfd_gpuvm_free_memory_of_gpu+0x34a/0x380 [amdgpu]</div> <div class="">[ 711.782044] kfd_process_device_free_bos+0xe0/0x130 [amdgpu]</div> <div class="">[ 711.782611] kfd_process_wq_release+0x286/0x380 [amdgpu]</div> <div class="">[ 711.783172] process_one_work+0x236/0x420</div> <div class="">[ 711.783543] worker_thread+0x34/0x400</div> <div class="">[ 711.783911] ? process_one_work+0x420/0x420</div> <div class="">[ 711.784279] kthread+0x126/0x140</div> <div class="">[ 711.784653] ? kthread_park+0x90/0x90</div> <div class="">[ 711.785018] ret_from_fork+0x22/0x30</div> <div class="">[ 711.785387] ---[ end trace d8f50f6594817c84 ]---</div> <div class="">[ 711.798716] [drm] amdgpu: ttm finalized</div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">So it means the process was stuck in some wait_event_killable (maybe here drm_sched_entity_flush) - you can try 'cat/proc/$process_pid/stack' maybe before<br class=""> you kill it to see where it was stuck so we can go from there.<br class=""> </p> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:5A64FAEA-CCE8-4EB6-8E7B-852D4F384255@microsoft.com" class=""> <div class=""> <div class=""><br class=""> </div> </div> <div class=""> <blockquote type="cite" class=""> <div class=""><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">For graphic apps what i usually see is a crash because of sigsev when</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">the app tries to access</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">an unmapped MMIO region on the device. I haven't tested for compute</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">stack and so there might</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">be something I haven't covered. Hang could mean for example waiting on a</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">fence which is not being</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">signaled - please provide full dmesg from this case.</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <blockquote type="cite" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br class=""> Do you have any good suggestions on how to fix it down the line? (HIP runtime/libhsakmt or driver)<br class=""> <br class=""> [64036.631333] amdgpu: amdgpu_vm_bo_update failed<br class=""> [64036.631702] amdgpu: validate_invalid_user_pages: update PTE failed<br class=""> [64036.640754] amdgpu: amdgpu_vm_bo_update failed<br class=""> [64036.641120] amdgpu: validate_invalid_user_pages: update PTE failed<br class=""> [64036.650394] amdgpu: amdgpu_vm_bo_update failed<br class=""> [64036.650765] amdgpu: validate_invalid_user_pages: update PTE failed<br class=""> </blockquote> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> </div> </blockquote> <div class=""><br class=""> </div> The full dmesg will just the repetition of those two messages,</div> <div class=""> <div class="">[186885.764079] amdgpu 0000:43:00.0: amdgpu: amdgpu: finishing device.</div> <div class="">[186885.766916] [drm] free PSP TMR buffer</div> <div class="">[186893.868173] amdgpu: amdgpu_vm_bo_update failed</div> <div class="">[186893.868235] amdgpu: validate_invalid_user_pages: update PTE failed</div> <div class="">[186893.876154] amdgpu: amdgpu_vm_bo_update failed</div> <div class="">[186893.876190] amdgpu: validate_invalid_user_pages: update PTE failed</div> <div class="">[186893.884150] amdgpu: amdgpu_vm_bo_update failed</div> <div class="">[186893.884185] amdgpu: validate_invalid_user_pages: update PTE failed</div> </div> <div class=""><br class=""> </div> <div class=""> <blockquote type="cite" class=""> <div class=""><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">This just probably means trying to update PTEs after the physical device</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">is gone - we usually avoid this by</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">first trying to do all HW shutdowns early before PCI remove completion</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">but when it's really tricky by</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">protecting HW access sections with drm_dev_enter/exit scope.</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">For this particular error it would be the best to flush</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">info->restore_userptr_work before the end of</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">amdgpu_pci_remove (rejecting new process creation and calling</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">cancel_delayed_work_sync(&process_info->restore_userptr_work) for all</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">running processes)</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">somewhere in amdgpu_pci_remove.</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br class=""> </div> </blockquote> I tried something like *kfd_process_ref_release* which I think did what you described, but it did not work.</div> </blockquote> <p class=""><br class=""> </p> <p class="">I don't see how kfd_process_ref_release is the same as I mentioned above, what i meant is calling the code above within kgd2kfd_suspend (where you tried to call kfd_kill_all_user_processes bellow) <br class=""> </p> </div> </div> </blockquote> Yes, you are right. It was not called by it. <br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:5A64FAEA-CCE8-4EB6-8E7B-852D4F384255@microsoft.com" class=""> <div class=""><br class=""> </div> <div class="">Instead I tried to kill the process from the kernel, but the amdgpu could **only** be hot-plugged in back successfully only if there was no rocm kernel running when it was plugged out. If not, amdgpu_probe will just hang later. (Maybe because amdgpu was plugged out while running state, it leaves a bad HW state which causes probe to hang).</div> </blockquote> <p class=""><br class=""> </p> <p class="">We usually do asic_reset during probe to reset all HW state (checlk if amdgpu_device_init->amdgpu_asic_reset is running when you plug back). <br class=""> </p> </div> </div> </blockquote> OK<br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""> <br class=""> </p> <blockquote type="cite" cite="mid:5A64FAEA-CCE8-4EB6-8E7B-852D4F384255@microsoft.com" class=""> <div class=""><br class=""> </div> <div class="">I don’t know if this is a viable solution worth pursuing, but I attached the diff anyway.</div> <div class=""><br class=""> </div> <div class="">Another solution could be let compute stack user mode detect a topology change via <span class="">generation_count change, and abort gracefully there.</span></div> <div class=""><br class=""> </div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">index 4e7d9cb09a69..79b4c9b84cd0 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">@@ -697,12 +697,15 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force)</div> <div class=""> return;</div> <div class=""><br class=""> </div> <div class=""> /* for runtime suspend, skip locking kfd */</div> <div class="">- if (!run_pm) {</div> <div class="">+ if (!run_pm && !drm_dev_is_unplugged(kfd->ddev)) {</div> <div class=""> /* For first KFD device suspend all the KFD processes */</div> <div class=""> if (atomic_inc_return(&kfd_locked) == 1)</div> <div class=""> kfd_suspend_all_processes(force);</div> <div class=""> }</div> <div class=""><br class=""> </div> <div class="">+ if (drm_dev_is_unplugged(kfd->ddev))</div> <div class="">+ kfd_kill_all_user_processes();</div> <div class="">+</div> <div class=""> kfd->dqm->ops.stop(kfd->dqm);</div> <div class=""> kfd_iommu_suspend(kfd);</div> <div class=""> }</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">index 55c9e1922714..84cbcd857856 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">@@ -1065,6 +1065,7 @@ void kfd_unref_process(struct kfd_process *p);</div> <div class=""> int kfd_process_evict_queues(struct kfd_process *p, bool force);</div> <div class=""> int kfd_process_restore_queues(struct kfd_process *p);</div> <div class=""> void kfd_suspend_all_processes(bool force);</div> <div class="">+void kfd_kill_all_user_processes(void);</div> <div class=""> /*</div> <div class=""> * kfd_resume_all_processes:</div> <div class=""> * bool sync: If kfd_resume_all_processes() should wait for the</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">index 6cdc855abb6d..fb0c753b682c 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">@@ -2206,6 +2206,24 @@ void kfd_suspend_all_processes(bool force)</div> <div class=""> srcu_read_unlock(&kfd_processes_srcu, idx);</div> <div class=""> }</div> <div class=""><br class=""> </div> <div class="">+</div> <div class="">+void kfd_kill_all_user_processes(void)</div> <div class="">+{</div> <div class="">+ struct kfd_process *p;</div> <div class="">+ struct amdkfd_process_info *p_info;</div> <div class="">+ unsigned int temp;</div> <div class="">+ int idx = srcu_read_lock(&kfd_processes_srcu);</div> <div class="">+</div> <div class="">+ pr_info("Killing all processes\n");</div> <div class="">+ hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {</div> <div class="">+ p_info = p->kgd_process_info;</div> <div class="">+ pr_info("Killing processes, pid = %d", pid_nr(p_info->pid));</div> <div class="">+ kill_pid(p_info->pid, SIGBUS, 1);</div> </blockquote> <p class=""><br class=""> </p> <p class="">From looking into kill_pid I see it only sends a signal but doesn't wait for completion, it would make sense to wait for completion here. In any case I would actually try to put here<span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><br class=""> </span></p> </div> </div> </blockquote> I have made a version which does that with some atomic counters. Please read later in the diff.<br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""></span><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4"><br class=""> </font></span></p> <p class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4">hash_for_each_rcu(</font></span><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4">p_info) <br class=""> cancel_delayed_work_sync(&</font></span><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4"><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""></span><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4">p_info</font></span>->restore_userptr_work) <br class=""> </font></span></p> <p class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4">instead at least that what i meant in the previous mail. </font></span></p> </div> </div> </blockquote> <div class="">I actually tried that earlier, and it did not work. Application still keeps running, and you have to send a kill to the user process.</div> <div class=""><br class=""> </div> <div class="">I have made the following version. It waits for processes to terminate synchronously after sending SIGBUS. After that it does the real work of amdgpu_pci_remove.</div> <div class="">However, it hangs at amdgpu_device_ip_fini_early when it is trying to deinit ip_block 6 <sdma_v4_0> (<a href="https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgitlab.freedesktop.org%2Fagd5f%2Flinux%2F-%2Fblob%2Famd-staging-drm-next%2Fdrivers%2Fgpu%2Fdrm%2Famd%2Famdgpu%2Famdgpu_device.c%23L2818&data=05%7C01%7Cshuotaoxu%40microsoft.com%7C59070142c40742c64d4b08da1eff2599%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637856378522526534%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=2WI5MRDqJeUWgu%2FWg6UvK2tK7wGbJ6zhxHkj3FVwImY%3D&reserved=0" originalsrc="https://gitlab.freedesktop.org/agd5f/linux/-/blob/amd-staging-drm-next/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c#L2818" shash="TgsVOoos0rAwPsEO/eOq4pWqZ9JOfMlmg0QbLG+E2CZrp/+mmY98QqdcrIuj5phJd8+z+AWaZwOzyNR2MMx4U9fYRt+iWi8xi/k2JwSioH3/795eYkWJphjBvlPgx2qvg/nEYZinDxtnXPdRqvNN4fmoWF08KKl3bg+ZtuOnOEI=" class="" moz-do-not-send="true">https://gitlab.freedesktop.org/agd5f/linux/-/blob/amd-staging-drm-next/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c#L2818</a>). I assume that there are still some inflight dma, therefore fini of this ip block thus hangs? </div> <div class=""><br class=""> </div> <div class="">The following is an excerpt of the dmesg: please excuse for putting my own pr_info, but I hope you get my point of where it hangs.</div> <div class=""><br class=""> </div> <div class=""> <div class="">[ 392.344735] amdgpu: all processes has been fully released</div> <div class="">[ 392.346557] amdgpu: amdgpu_acpi_fini done</div> <div class="">[ 392.346568] amdgpu 0000:b3:00.0: amdgpu: amdgpu: finishing device.</div> <div class="">[ 392.349238] amdgpu: amdgpu_device_ip_fini_early enter ip_blocks = 9</div> <div class="">[ 392.349248] amdgpu: Free mem_obj = 000000007bf54275, range_start = 14, range_end = 14</div> <div class="">[ 392.350299] amdgpu: Free mem_obj = 00000000a85bc878, range_start = 12, range_end = 12</div> <div class="">[ 392.350304] amdgpu: Free mem_obj = 00000000b8019e32, range_start = 13, range_end = 13</div> <div class="">[ 392.350308] amdgpu: Free mem_obj = 000000002d296168, range_start = 4, range_end = 11</div> <div class="">[ 392.350313] amdgpu: Free mem_obj = 000000001fc4f934, range_start = 0, range_end = 3</div> <div class="">[ 392.350322] amdgpu: amdgpu_amdkfd_suspend(adev, false) done</div> <div class="">[ 392.350672] amdgpu: hw_fini of IP block[8] <jpeg_v2_5> done 0</div> <div class="">[ 392.350679] amdgpu: hw_fini of IP block[7] <vcn_v2_5> done 0</div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">I just remembered that the idea to actively kill and wait for running user processes during unplug was rejected<br class=""> as a bad idea in the first iteration of unplug work I did (don't remember why now, need to look) so this is a no go.<br class=""> </p> </div> </div> </blockquote> <div class="">Maybe an application has kfd open, but was not accessing the dev. So kill it at unplug could kill the process unnecessarily.</div> <div class="">However, the latest version I had with the sleep function got rid of the IP block fini hang.</div> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class="">Our policy is to let zombie processes (zombie in a sense that the underlying device is gone) live as long as they want <br class=""> (as long as you able to terminate them - which you do, so that ok)<br class=""> and the system should finish PCI remove gracefully and be able to hot plug back the device. Hence, i suggest dropping<br class=""> this direction of forcing all user processes to be killed, confirm you have graceful shutdown and remove of device<br class=""> from PCI topology and then concentrate on why when you plug back it hangs. </p> </div> </div> </blockquote> So I basically revert back to the original solution which you suggested.</div> <div class=""><br class=""> </div> <div class=""> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">index 4e7d9cb09a69..5504a18b5a45 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">@@ -697,7 +697,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force)</div> <div class=""> return;</div> <div class=""><br class=""> </div> <div class=""> /* for runtime suspend, skip locking kfd */</div> <div class="">- if (!run_pm) {</div> <div class="">+ if (!run_pm && !drm_dev_is_unplugged(kfd->ddev)) {</div> <div class=""> /* For first KFD device suspend all the KFD processes */</div> <div class=""> if (atomic_inc_return(&kfd_locked) == 1)</div> <div class=""> kfd_suspend_all_processes(force);</div> </div> <div class=""><br class=""> </div> <div class=""> <blockquote type="cite" class=""> <div class=""> <p class="">First confirm if ASIC reset happens on<br class=""> next init. </p> </div> </blockquote> <div class=""> <div class="">This patch works great at <b class="">planned</b> plugout, where all the rocm processes are killed before plugout. And device can be added back without a problem.</div> <div class="">However <b class="">unplanned</b> plugout when there is rocm processes are running just don’t work.</div> </div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">Still I am not clear if ASIC reset happens on plug back or no, can you trace this please ?</p> <div class=""><br class=""> </div> </div> </div> </blockquote> <div><br class=""> </div> I tried add pr_info into asic_reset functions, but cannot trace any upon plug-back.<br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""></div> <blockquote type="cite" class=""> <div class=""> <p class="">Second please confirm if the timing you kill manually the user process has impact on whether you have a hang<br class=""> on next plug back (if you kill before </p> </div> </blockquote> <b class="">Scenario 0: Kill before plug back</b></div> <div class=""><br class=""> </div> <div class=""> <div class="">1. echo 1 > /sys/bus/pci/…/remove, would finish. </div> <div class="">But the application won’t exit until there is a kill signal.</div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">Why you think it must exit ? <br class=""> </p> </div> </div> </blockquote> Because rocm will need to release the drm descriptor to get amdgpu_amdkfd_device_fini_sw called, which would eventually call kgd2kfd_device_exit called. This would clean up kfd_topology at least. Otherwise I don’t see how it would be added back without messing up kfd topology to say the least. </div> <div><br class=""> </div> <div>However, those are all based my own observations. Please explain why it does not need exit if you believe so?</div> <div><br class=""> </div> <div> <blockquote type="cite" class=""> <div class=""> <div class=""> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""><br class=""> </div> <div class="">2. kill the the process. The application does several things and seems trigger drm_release in the kernel, which are met with kernel NULL pointer deference related to sysfs_remove. Then the whole fs just freeze.</div> <div class=""><br class=""> </div> <div class=""> <div class="">[ +0.002498] BUG: kernel NULL pointer dereference, address: 0000000000000098</div> <div class="">[ +0.000486] #PF: supervisor read access in kernel mode</div> <div class="">[ +0.000545] #PF: error_code(0x0000) - not-present page</div> <div class="">[ +0.000551] PGD 0 P4D 0</div> <div class="">[ +0.000553] Oops: 0000 [#1] SMP NOPTI</div> <div class="">[ +0.000540] CPU: 75 PID: 4911 Comm: kworker/75:2 Tainted: G W E 5.13.0-kfd #1</div> <div class="">[ +0.000559] Hardware name: INGRASYS TURING /MB , BIOS K71FQ28A 10/05/2021</div> <div class="">[ +0.000567] Workqueue: events delayed_fput</div> <div class="">[ +0.000563] RIP: 0010:kernfs_find_ns+0x1b/0x100</div> <div class="">[ +0.000569] Code: ff ff e8 88 59 9f 00 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 57 8b 05 df f0 7b 01 41 56 41 55 49 89 f5 41 54 55 48 89 fd 53 <44> 0f b7 b7 98 00 00 00 48 89 d3 4c 8b 67 70 66 41 83 e6 20 41 0f</div> <div class="">[ +0.001193] RSP: 0018:ffffb9875db5fc98 EFLAGS: 00010246</div> <div class="">[ +0.000602] RAX: 0000000000000000 RBX: ffffa101f79507d8 RCX: 0000000000000000</div> <div class="">[ +0.000612] RDX: 0000000000000000 RSI: ffffffffc09a9417 RDI: 0000000000000000</div> <div class="">[ +0.000604] RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000</div> <div class="">[ +0.000760] R10: ffffb9875db5fcd0 R11: 0000000000000000 R12: 0000000000000000</div> <div class="">[ +0.000597] R13: ffffffffc09a9417 R14: ffffa08363fb2d18 R15: 0000000000000000</div> <div class="">[ +0.000702] FS: 0000000000000000(0000) GS:ffffa0ffbfcc0000(0000) knlGS:0000000000000000</div> <div class="">[ +0.000666] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033</div> <div class="">[ +0.000658] CR2: 0000000000000098 CR3: 0000005747812005 CR4: 0000000000770ee0</div> <div class="">[ +0.000715] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000</div> <div class="">[ +0.000655] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400</div> <div class="">[ +0.000592] PKRU: 55555554</div> <div class="">[ +0.000580] Call Trace:</div> <div class="">[ +0.000591] kernfs_find_and_get_ns+0x2f/0x50</div> <div class="">[ +0.000584] sysfs_remove_file_from_group+0x20/0x50</div> <div class="">[ +0.000580] amdgpu_ras_sysfs_remove+0x3d/0xd0 [amdgpu]</div> <div class="">[ +0.000737] amdgpu_ras_late_fini+0x1d/0x40 [amdgpu]</div> <div class="">[ +0.000750] amdgpu_sdma_ras_fini+0x96/0xb0 [amdgpu]</div> <div class="">[ +0.000742] ? gfx_v10_0_resume+0x10/0x10 [amdgpu]</div> <div class="">[ +0.000738] sdma_v4_0_sw_fini+0x23/0x90 [amdgpu]</div> <div class="">[ +0.000717] amdgpu_device_fini_sw+0xae/0x260 [amdgpu]</div> <div class="">[ +0.000704] amdgpu_driver_release_kms+0x12/0x30 [amdgpu]</div> <div class="">[ +0.000687] drm_dev_release+0x20/0x40 [drm]</div> <div class="">[ +0.000583] drm_release+0xa8/0xf0 [drm]</div> <div class="">[ +0.000584] __fput+0xa5/0x250</div> <div class="">[ +0.000621] delayed_fput+0x1f/0x30</div> <div class="">[ +0.000726] process_one_work+0x26e/0x580</div> <div class="">[ +0.000581] ? process_one_work+0x580/0x580</div> <div class="">[ +0.000611] worker_thread+0x4d/0x3d0</div> <div class="">[ +0.000611] ? process_one_work+0x580/0x580</div> <div class="">[ +0.000605] kthread+0x117/0x150</div> <div class="">[ +0.000611] ? kthread_park+0x90/0x90</div> <div class="">[ +0.000619] ret_from_fork+0x1f/0x30</div> <div class="">[ +0.000625] Modules linked in: amdgpu(E) xt_conntrack xt_MASQUERADE nfnetlink xt_addrtype iptable_filter iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 br_netfilter x86_pkg_temp_thermal cdc_ether usbnet acpi_pad msr ip_tables x_tables ast drm_vram_helper iommu_v2 drm_ttm_helper gpu_sched ttm drm_kms_helper cfbfillrect syscopyarea cfbimgblt sysfillrect sysimgblt fb_sys_fops cfbcopyarea drm drm_panel_orientati</div> <div class="">on_quirks [last unloaded: amdgpu]</div> </div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">This is a known regression, all SYSFS components must be removed before pci_remove code runs otherwise you get either warnings for single file removals or<br class=""> OOPSEs for sysfs gorup removals like here. Please try to move amdgpu_ras_sysfs_remove from amdgpu_ras_late_fini to the end of amdgpu_ras_pre_fini (which happens before pci remove)</p> <div class=""><br class=""> </div> </div> </div> </blockquote> <div><br class=""> </div> I fixed it in the newer patch, please see it below.</div> <div>I first plugout the device, then kill the rocm user process. Then it has other OOPSES related to ttm_bo_cleanup_refs.</div> <div><br class=""> </div> <div>[ +0.000006] BUG: kernel NULL pointer dereference, address: 0000000000000010<br class=""> [ +0.000349] #PF: supervisor read access in kernel mode<br class=""> [ +0.000340] #PF: error_code(0x0000) - not-present page<br class=""> [ +0.000341] PGD 0 P4D 0<br class=""> [ +0.000336] Oops: 0000 [#1] SMP NOPTI<br class=""> [ +0.000345] CPU: 9 PID: 95 Comm: kworker/9:1 Tainted: G W E 5.13.0-kfd #1<br class=""> [ +0.000367] Hardware name: INGRASYS TURING /MB , BIOS K71FQ28A 10/05/2021<br class=""> [ +0.000376] Workqueue: events delayed_fput<br class=""> [ +0.000422] RIP: 0010:ttm_resource_free+0x24/0x40 [ttm]<br class=""> [ +0.000464] Code: 00 00 0f 1f 40 00 0f 1f 44 00 00 53 48 89 f3 48 8b 36 48 85 f6 74 21 48 8b 87 28 02 00 00 48 63 56 10 48 8b bc d0 b8 00 00 00 <48> 8b 47 10 ff 50 08 48 c7 03 00 00 00 00 5b c3 66 66 2e 0f 1f 84<br class=""> [ +0.001009] RSP: 0018:ffffb21c59413c98 EFLAGS: 00010282<br class=""> [ +0.000515] RAX: ffff8b1aa4285f68 RBX: ffff8b1a823b5ea0 RCX: 00000000002a000c<br class=""> [ +0.000536] RDX: 0000000000000000 RSI: ffff8b1acb84db80 RDI: 0000000000000000<br class=""> [ +0.000539] RBP: 0000000000000001 R08: 0000000000000000 R09: ffffffffc03c3e00<br class=""> [ +0.000543] R10: 0000000000000000 R11: 0000000000000000 R12: ffff8b1a823b5ec8<br class=""> [ +0.000542] R13: 0000000000000000 R14: ffff8b1a823b5d90 R15: ffff8b1a823b5ec8<br class=""> [ +0.000544] FS: 0000000000000000(0000) GS:ffff8b187f440000(0000) knlGS:0000000000000000<br class=""> [ +0.000559] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033<br class=""> [ +0.000575] CR2: 0000000000000010 CR3: 00000076e6812004 CR4: 0000000000770ee0<br class=""> [ +0.000575] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000<br class=""> [ +0.000579] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400<br class=""> [ +0.000575] PKRU: 55555554<br class=""> [ +0.000568] Call Trace:<br class=""> [ +0.000567] ttm_bo_cleanup_refs+0xe4/0x290 [ttm]<br class=""> [ +0.000588] ttm_bo_delayed_delete+0x147/0x250 [ttm]<br class=""> [ +0.000589] ttm_device_fini+0xad/0x1b0 [ttm]<br class=""> [ +0.000590] amdgpu_ttm_fini+0x2a7/0x310 [amdgpu]<br class=""> [ +0.000730] gmc_v9_0_sw_fini+0x3a/0x40 [amdgpu]<br class=""> [ +0.000753] amdgpu_device_fini_sw+0xae/0x260 [amdgpu]<br class=""> [ +0.000734] amdgpu_driver_release_kms+0x12/0x30 [amdgpu]<br class=""> [ +0.000737] drm_dev_release+0x20/0x40 [drm]<br class=""> [ +0.000626] drm_release+0xa8/0xf0 [drm]<br class=""> [ +0.000625] __fput+0xa5/0x250<br class=""> [ +0.000606] delayed_fput+0x1f/0x30<br class=""> [ +0.000607] process_one_work+0x26e/0x580<br class=""> [ +0.000608] ? process_one_work+0x580/0x580<br class=""> [ +0.000616] worker_thread+0x4d/0x3d0<br class=""> [ +0.000614] ? process_one_work+0x580/0x580<br class=""> [ +0.000617] kthread+0x117/0x150<br class=""> [ +0.000615] ? kthread_park+0x90/0x90<br class=""> [ +0.000621] ret_from_fork+0x1f/0x30<br class=""> [ +0.000603] Modules linked in: amdgpu(E) xt_conntrack xt_MASQUERADE nfnetlink xt_addrtype iptable_filter iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 br_netfilter x86_pkg_temp_thermal cdc_ether usbnet acpi_pad msr ip_tables x_tables ast drm_vram_helper drm_ttm_helper iommu_v2 ttm gpu_sched drm_kms_helper cfbfillrect syscopyarea cfbimgblt sysfillrect sysimgblt fb_sys_fops cfbcopyarea drm drm_panel_orientation_quirks [last unloaded: amdgpu]<br class=""> [ +0.002840] CR2: 0000000000000010<br class=""> <div class=""> <div><font color="#000000" class=""><span style="caret-color: rgb(0, 0, 0);" class="">[ +0.000755] ---[ end trace 9737737402551e39 ]--</span></font></div> </div> <br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""> <div class=""><br class=""> </div> <div class="">3. echo 1 > /sys/bus/pci/rescan. This would just hang. I assume the sysfs is broken.</div> <div class=""><br class=""> </div> <div class="">Based on 1 & 2, it seems that 1 won’t let the amdgpu exit gracefully, because 2 will do some cleanup maybe should have happened before 1.</div> <div class=""> <blockquote type="cite" class=""> <div class=""> <p class="">or you kill after plug back does it makes a difference). <br class=""> </p> </div> </blockquote> </div> <div class=""><b class="">Scenario 2: Kill after plug back</b></div> <div class=""><br class=""> </div> <div class="">If I perform rescan before kill, then the driver seemed probed fine. But kill will have the same issue which messed up the sysfs the same way as in Scenario 2.</div> <div class=""><br class=""> </div> <div class=""><br class=""> </div> <div class=""><b class="">Final Comments:</b></div> <div class=""><br class=""> </div> <span class="">0. cancel_delayed_work_sync(&p_info->restore_userptr_work) would make the repletion of </span>amdgpu_vm_bo_update failure go away, but it does not solve the issues in those scenarios.</div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">Still - it's better to do it this way even for those failures to go awaya</p> <div class=""><br class=""> </div> </div> </div> </blockquote> Cancel_delayed_work is insufficient, you will need to make sure the work won’t be processed after plugout. Please see my patch<br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""><span class=""><br class=""> </span> <div class="">1. For planned hotplug, this patch should work as long as you follow some protocol, i.e. kill before plugout. Is this patch an acceptable one since it provides some added feature than before?</div> </div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">Let's try to fix more as I advised above. <br class=""> </p> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""> <div class=""><br class=""> </div> <div class="">2. For unplanned hotplug when there is rocm app running, the patch that kill all processes and wait for 5 sec would work consistently. But it seems that it is an unacceptable solution for official release. I can hold it for our own internal usage. It seems that kill after removal would cause problems, and I don’t know if there is a quick fix by me because of my limited understanding of the amdgpu driver. Maybe AMD could have a quick fix; Or it is really a difficult one. This feature may or may not be a blocking issue in our GPU disaggregation research down the way. Please let us know for either cases, and we would like to learn and help as much as we could!</div> </div> </div> </div> </blockquote> <p class=""><br class=""> </p> <p class="">I am currently not sure why it helps. I will need to setup my own ROCm setup and retest hot plug to check this in more depth but currently i have higher priorities. Please try to confirm ASIC reset always takes place on plug back<br class=""> and fix the sysfs OOPs as I advised above to clear up at least some of the issues. Also please describe to me exactly what you steps to reproduce this scenario so later I might be able to do it myself.</p> <div class=""><br class=""> </div> </div> </div> </blockquote> <div>I can still try to help to fix the bug in my spare time. My setup is as follows</div> <div><br class=""> </div> <div> <ol class=""> <li class="">I have a server with 4 AMD MI100 GPUs. I think 1 GPU would also work.</li><li class="">I used the <a href="https://github.com/RadeonOpenCompute/ROCK-Kernel-Driver/tree/roc-5.0.x" class="">https://github.com/RadeonOpenCompute/ROCK-Kernel-Driver/tree/roc-5.0.x</a> as the starting point, and apply Mukul’s patch and my patch.</li><li class="">Then I run a tensorflow benchmark from a docker.</li><ul class=""> <li class="">docker run -it --device=/dev/kfd --device=/dev/dri --group-add video rocm/tensorflow:rocm4.5.2-tf1.15-dev</li><li class="">And run the following benchmark in the docker: python benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py --num_gpus=4 --batch_size=32 --model=resnet50 --variable_update=parameter_server</li><ul class=""> <li class="">Might to need to adjust num_gpus parameter based on your setup</li></ul> </ul> <li class="">Remove a GPU at random time.</li><li class="">Do whatever is needed to before plugback and reverify the benchmark can still run.</li></ol> </div> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class="">Also, we have hotplug test suite in libdrm (graphic stack), so maybe u can install libdrm and run that test suite to see if it exposes more issues.<br class=""> </p> </div> </div> </blockquote> OK I could try it some time.</div> <div><br class=""> </div> <div>The following is the new diff.</div> <div><br class=""> </div> <div> <div>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h</div> <div>index 182b7eae598a..48c3cd4054de 100644</div> <div>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h</div> <div>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h</div> <div>@@ -1327,7 +1327,7 @@ int emu_soc_asic_init(struct amdgpu_device *adev);</div> <div> * ASICs macro.</div> <div> */</div> <div> #define amdgpu_asic_set_vga_state(adev, state) (adev)->asic_funcs->set_vga_state((adev), (state))</div> <div>-#define amdgpu_asic_reset(adev) (adev)->asic_funcs->reset((adev))</div> <div>+#define amdgpu_asic_reset(adev) ({int r; pr_info("performing amdgpu_asic_reset\n"); r = (adev)->asic_funcs->reset((adev));r;})</div> <div> #define amdgpu_asic_reset_method(adev) (adev)->asic_funcs->reset_method((adev))</div> <div> #define amdgpu_asic_get_xclk(adev) (adev)->asic_funcs->get_xclk((adev))</div> <div> #define amdgpu_asic_set_uvd_clocks(adev, v, d) (adev)->asic_funcs->set_uvd_clocks((adev), (v), (d))</div> <div>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div>index 27c74fcec455..842abd7150a6 100644</div> <div>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div>@@ -134,6 +134,7 @@ struct amdkfd_process_info {</div> <div> </div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>/* MMU-notifier related fields */</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>atomic_t evicted_bos;</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_t invalid;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>struct delayed_work restore_userptr_work;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>struct pid *pid;</div> <div> };</div> <div>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c</div> <div>index 99d2b15bcbf3..2a588eb9f456 100644</div> <div>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c</div> <div>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c</div> <div>@@ -1325,6 +1325,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,</div> <div> </div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>info->pid = get_task_pid(current->group_leader, PIDTYPE_PID);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>atomic_set(&info->evicted_bos, 0);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_set(&info->invalid, 0);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>INIT_DELAYED_WORK(&info->restore_userptr_work,</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span> amdgpu_amdkfd_restore_userptr_worker);</div> <div> </div> <div>@@ -2693,6 +2694,9 @@ static void amdgpu_amdkfd_restore_userptr_worker(struct work_struct *work)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>struct mm_struct *mm;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>int evicted_bos;</div> <div> </div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>if (atomic_read(&process_info->invalid))</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>return;</div> <div>+</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>evicted_bos = atomic_read(&process_info->evicted_bos);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!evicted_bos)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return;</div> <div>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c</div> <div>index ec38517ab33f..e7d85d8d282d 100644</div> <div>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c</div> <div>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c</div> <div>@@ -1054,6 +1054,7 @@ void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,</div> <div> */</div> <div> void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)</div> <div> {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_debug("%s called\n",__func__);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);</div> <div> }</div> <div> </div> <div>@@ -1066,6 +1067,7 @@ void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)</div> <div> */</div> <div> int amdgpu_device_pci_reset(struct amdgpu_device *adev)</div> <div> {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_debug("%s called\n",__func__);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return pci_reset_function(adev->pdev);</div> <div> }</div> <div> </div> <div>@@ -4702,6 +4704,8 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle,</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>bool need_full_reset, skip_hw_reset, vram_lost = false;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>int r = 0;</div> <div> </div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_debug("%s called\n",__func__);</div> <div>+</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>/* Try reset handler method first */</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device,</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span> reset_list);</div> <div>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c</div> <div>index 49bdf9ff7350..b469acb65c1e 100644</div> <div>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c</div> <div>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c</div> <div>@@ -2518,7 +2518,6 @@ void amdgpu_ras_late_fini(struct amdgpu_device *adev,</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!ras_block || !ih_info)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return;</div> <div> </div> <div>-<span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_ras_sysfs_remove(adev, ras_block);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (ih_info->cb)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_ras_interrupt_remove_handler(adev, ih_info);</div> <div> }</div> <div>@@ -2577,6 +2576,7 @@ void amdgpu_ras_suspend(struct amdgpu_device *adev)</div> <div> int amdgpu_ras_pre_fini(struct amdgpu_device *adev)</div> <div> {</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>struct amdgpu_ras *con = amdgpu_ras_get_context(adev);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>struct ras_manager *obj, *tmp;</div> <div> </div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!adev->ras_enabled || !con)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return 0;</div> <div>@@ -2585,6 +2585,13 @@ int amdgpu_ras_pre_fini(struct amdgpu_device *adev)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>/* Need disable ras on all IPs here before ip [hw/sw]fini */</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_ras_disable_all_features(adev, 0);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_ras_recovery_fini(adev);</div> <div>+</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>/* remove sysfs before pci_remove to avoid OOPSES from sysfs_remove_groups */</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>list_for_each_entry_safe(obj, tmp, &con->head, node) {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_ras_sysfs_remove(adev, &obj->head);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>put_obj(obj);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div>+</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return 0;</div> <div> }</div> <div> </div> <div>diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div>index 4e7d9cb09a69..0fa806a78e39 100644</div> <div>--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div>+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div>@@ -693,16 +693,35 @@ bool kfd_is_locked(void)</div> <div> </div> <div> void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force)</div> <div> {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>struct kfd_process *p;</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>struct amdkfd_process_info *p_info;</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>unsigned int temp;</div> <div>+</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!kfd->init_complete)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>return;</div> <div> </div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>/* for runtime suspend, skip locking kfd */</div> <div>-<span class="Apple-tab-span" style="white-space:pre"> </span>if (!run_pm) {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>if (!run_pm && !drm_dev_is_unplugged(kfd->ddev)) {</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>/* For first KFD device suspend all the KFD processes */</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (atomic_inc_return(&kfd_locked) == 1)</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>kfd_suspend_all_processes(force);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div> </div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>if (drm_dev_is_unplugged(kfd->ddev)){</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>int idx = srcu_read_lock(&kfd_processes_srcu);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_debug("cancel restore_userptr_wor\n");</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>if ( kfd_process_gpuidx_from_gpuid(p, kfd->id) >= 0 ) {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>p_info = p->kgd_process_info;</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_debug("cancel processes, pid = %d for gpu_id = %d", pid_nr(p_info->pid), kfd->id);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>cancel_delayed_work_sync(&p_info->restore_userptr_work);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>/* block all future restore_userptr_work */</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_inc(&p_info->invalid);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>srcu_read_unlock(&kfd_processes_srcu, idx);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div>+</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>kfd->dqm->ops.stop(kfd->dqm);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>kfd_iommu_suspend(kfd);</div> <div> }</div> <div>diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c</div> <div>index 600ba2a728ea..7e3d1848eccc 100644</div> <div>--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c</div> <div>+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c</div> <div>@@ -669,11 +669,12 @@ static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev)</div> <div> #ifdef HAVE_AMD_IOMMU_PC_SUPPORTED</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>if (dev->kobj_perf) {</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>list_for_each_entry(perf, &dev->perf_props, list) {</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>sysfs_remove_group(dev->kobj_perf, perf->attr_group);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>kfree(perf->attr_group);</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>perf->attr_group = NULL;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>kobject_del(dev->kobj_perf);</div> <div>-<span class="Apple-tab-span" style="white-space:pre"> </span>kobject_put(dev->kobj_perf);</div> <div>+<span class="Apple-tab-span" style="white-space:pre"> </span>/* kobject_put(dev->kobj_perf); */</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>dev->kobj_perf = NULL;</div> <div> <span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div> #endif</div> <div class=""><br class=""> </div> <div class="">Thank you so much! Looking forward to your comments!</div> <div class=""><br class=""> </div> <div class="">Regards,</div> <div class="">Shuotao</div> <blockquote type="cite" class=""> <div class=""> <div class=""> <p class=""></p> <p class="">Andrey</p> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:76C16B06-0539-48EE-AB11-80089490CE3A@microsoft.com" class=""> <div class=""> <div class=""> <div class=""> <div class=""><br class=""> </div> <div class="">Thank you so much!</div> <div class=""><br class=""> </div> <div class="">Best regards,</div> <div class="">Shuotao</div> </div> <blockquote type="cite" class=""> <div class=""> <p class="">Andrey</p> <p class=""><br class=""> </p> <blockquote type="cite" cite="mid:B7970589-ACF6-41F3-8622-1C0F705F3EE0@microsoft.com" class=""> <div class=""> <div class=""><br class=""> </div> <div class=""><br class=""> </div> <div class=""> <div class="">diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c</div> <div class="">index 8fa9b86ac9d2..c0b27f722281 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c</div> <div class="">@@ -188,6 +188,12 @@ void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>kgd2kfd_interrupt(adev->kfd.dev, ih_ring_entry);</div> <div class=""> }</div> <div class=""> </div> <div class="">+void amdgpu_amdkfd_kill_all_processes(struct amdgpu_device *adev)</div> <div class="">+{</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>if (adev->kfd.dev)</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>kgd2kfd_kill_all_user_processes(adev->kfd.dev);</div> <div class="">+}</div> <div class="">+</div> <div class=""> void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool run_pm)</div> <div class=""> {</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>if (adev->kfd.dev)</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div class="">index 27c74fcec455..f4e485d60442 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div class="">+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h</div> <div class="">@@ -141,6 +141,7 @@ struct amdkfd_process_info {</div> <div class=""> int amdgpu_amdkfd_init(void);</div> <div class=""> void amdgpu_amdkfd_fini(void);</div> <div class=""> </div> <div class="">+void amdgpu_amdkfd_kill_all_processes(struct amdgpu_device *adev);</div> <div class=""> void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool run_pm);</div> <div class=""> int amdgpu_amdkfd_resume_iommu(struct amdgpu_device *adev);</div> <div class=""> int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool run_pm, bool sync);</div> <div class="">@@ -405,6 +406,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>const struct kgd2kfd_shared_resources *gpu_resources);</div> <div class=""> void kgd2kfd_device_exit(struct kfd_dev *kfd);</div> <div class=""> void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force);</div> <div class="">+void kgd2kfd_kill_all_user_processes(struct kfd_dev *kfd);</div> <div class=""> int kgd2kfd_resume_iommu(struct kfd_dev *kfd);</div> <div class=""> int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm, bool sync);</div> <div class=""> int kgd2kfd_pre_reset(struct kfd_dev *kfd);</div> <div class="">@@ -443,6 +445,9 @@ static inline void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force)</div> <div class=""> {</div> <div class=""> }</div> <div class=""> </div> <div class="">+void kgd2kfd_kill_all_user_processes(struct kfd_dev *kfd){</div> <div class="">+}</div> <div class="">+</div> <div class=""> static int __maybe_unused kgd2kfd_resume_iommu(struct kfd_dev *kfd)</div> <div class=""> {</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>return 0;</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c</div> <div class="">index 3d5fc0751829..af6fe5080cfa 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c</div> <div class="">@@ -2101,6 +2101,9 @@ amdgpu_pci_remove(struct pci_dev *pdev)</div> <div class=""> {</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>struct drm_device *dev = pci_get_drvdata(pdev);</div> <div class=""> </div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>/* kill all kfd processes before drm_dev_unplug */</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>amdgpu_amdkfd_kill_all_processes(drm_to_adev(dev));</div> <div class="">+</div> <div class=""> #ifdef HAVE_DRM_DEV_UNPLUG</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>drm_dev_unplug(dev);</div> <div class=""> #else</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">index 5504a18b5a45..480c23bef5e2 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c</div> <div class="">@@ -691,6 +691,12 @@ bool kfd_is_locked(void)</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>return (atomic_read(&kfd_locked) > 0);</div> <div class=""> }</div> <div class=""> </div> <div class="">+inline void kgd2kfd_kill_all_user_processes(struct kfd_dev* dev)</div> <div class="">+{</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>kfd_kill_all_user_processes();</div> <div class="">+}</div> <div class="">+</div> <div class="">+</div> <div class=""> void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm, bool force)</div> <div class=""> {</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!kfd->init_complete)</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">index 55c9e1922714..a35a2cb5bb9f 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h</div> <div class="">@@ -1064,6 +1064,7 @@ static inline struct kfd_process_device *kfd_process_device_from_gpuidx(</div> <div class=""> void kfd_unref_process(struct kfd_process *p);</div> <div class=""> int kfd_process_evict_queues(struct kfd_process *p, bool force);</div> <div class=""> int kfd_process_restore_queues(struct kfd_process *p);</div> <div class="">+void kfd_kill_all_user_processes(void);</div> <div class=""> void kfd_suspend_all_processes(bool force);</div> <div class=""> /*</div> <div class=""> * kfd_resume_all_processes:</div> <div class="">diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">index 6cdc855abb6d..17e769e6951d 100644</div> <div class="">--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c</div> <div class="">@@ -46,6 +46,9 @@ struct mm_struct;</div> <div class=""> #include "kfd_trace.h"</div> <div class=""> #include "kfd_debug.h"</div> <div class=""> </div> <div class="">+static atomic_t kfd_process_locked = ATOMIC_INIT(0);</div> <div class="">+static atomic_t kfd_inflight_kills = ATOMIC_INIT(0);</div> <div class="">+</div> <div class=""> /*</div> <div class=""> * List of struct kfd_process (field kfd_process).</div> <div class=""> * Unique/indexed by mm_struct*</div> <div class="">@@ -802,6 +805,9 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>struct kfd_process *process;</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>int ret;</div> <div class=""> </div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>if ( atomic_read(&kfd_process_locked) > 0 )</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>return ERR_PTR(-EINVAL);</div> <div class="">+</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>if (!(thread->mm && mmget_not_zero(thread->mm)))</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>return ERR_PTR(-EINVAL);</div> <div class=""> </div> <div class="">@@ -1126,6 +1132,10 @@ static void kfd_process_wq_release(struct work_struct *work)</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>put_task_struct(p->lead_thread);</div> <div class=""> </div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>kfree(p);</div> <div class="">+</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>if ( atomic_read(&kfd_process_locked) > 0 ){</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_dec(&kfd_inflight_kills);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div class=""> }</div> <div class=""> </div> <div class=""> static void kfd_process_ref_release(struct kref *ref)</div> <div class="">@@ -2186,6 +2196,35 @@ static void restore_process_worker(struct work_struct *work)</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>pr_err("Failed to restore queues of pasid 0x%x\n", p->pasid);</div> <div class=""> }</div> <div class=""> </div> <div class="">+void kfd_kill_all_user_processes(void)</div> <div class="">+{</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>struct kfd_process *p;</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>/* struct amdkfd_process_info *p_info; */</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>unsigned int temp;</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>int idx;</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_inc(&kfd_process_locked);</div> <div class="">+</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>idx = srcu_read_lock(&kfd_processes_srcu);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_info("Killing all processes\n");</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>dev_warn(kfd_device,</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>"Sending SIGBUS to process %d (pasid 0x%x)",</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>p->lead_thread->pid, p->pasid);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>send_sig(SIGBUS, p->lead_thread, 0);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_inc(&kfd_inflight_kills);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>srcu_read_unlock(&kfd_processes_srcu, idx);</div> <div class="">+</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>while ( atomic_read(&kfd_inflight_kills) > 0 ){</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>dev_warn(kfd_device,</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>"kfd_processes_table is not empty, going to sleep for 10ms\n");</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>msleep(10);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>}</div> <div class="">+</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>atomic_dec(&kfd_process_locked);</div> <div class="">+<span class="Apple-tab-span" style="white-space:pre"> </span>pr_info("all processes has been fully released");</div> <div class="">+}</div> <div class="">+</div> <div class=""> void kfd_suspend_all_processes(bool force)</div> <div class=""> {</div> <div class=""> <span class="Apple-tab-span" style="white-space:pre"> </span>struct kfd_process *p;</div> <div class=""><br class=""> </div> <div class=""><br class=""> </div> </div> <div class=""><br class=""> </div> <div class="">Regards,</div> <div class="">Shuotao</div> <br class=""> <blockquote type="cite" class=""> <div class=""> <div class=""> <div class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""></span><br class="webkit-block-placeholder"> </div> <p class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class=""><font class="" size="4">Andrey</font><br class=""> <br class=""> </span></p> <blockquote type="cite" cite="mid:5A64FAEA-CCE8-4EB6-8E7B-852D4F384255@microsoft.com" class=""> <div class="">+ }</div> <div class="">+ srcu_read_unlock(&kfd_processes_srcu, idx);</div> <div class="">+}</div> <div class="">+</div> <div class="">+</div> <div class=""> int kfd_resume_all_processes(bool sync)</div> <div class=""> {</div> <div class=""> struct kfd_process *p;</div> <div class=""><br class=""> </div> <div class=""><br class=""> <blockquote type="cite" class=""> <div class=""><span style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none; float: none; display: inline !important;" class="">Andrey</span><br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br style="caret-color: rgb(0, 0, 0); font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <blockquote type="cite" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: 400; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; text-decoration: none;" class=""> <br class=""> Really appreciate your help!<br class=""> <br class=""> Best,<br class=""> Shuotao<br class=""> <br class=""> <blockquote type="cite" class=""> <blockquote type="cite" class="">2. Remove redudant p2p/io links in sysfs when device is hotplugged<br class=""> out.<br class=""> <br class=""> 3. New kfd node_id is not properly assigned after a new device is<br class=""> added after a gpu is hotplugged out in a system. libhsakmt will<br class=""> find this anomaly, (i.e. node_from != <dev node id> in iolinks),<br class=""> when taking a topology_snapshot, thus returns fault to the rocm<br class=""> stack.<br class=""> <br class=""> -- This patch fixes issue 1; another patch by Mukul fixes issues 2&3.<br class=""> -- Tested on a 4-GPU MI100 gpu nodes with kernel 5.13.0-kfd; kernel<br class=""> 5.16.0-kfd is unstable out of box for MI100.<br class=""> ---<br class=""> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 5 +++++<br class=""> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 7 +++++++<br class=""> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 +<br class=""> drivers/gpu/drm/amd/amdkfd/kfd_device.c | 13 +++++++++++++<br class=""> 4 files changed, 26 insertions(+)<br class=""> <br class=""> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c<br class=""> index c18c4be1e4ac..d50011bdb5c4 100644<br class=""> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c<br class=""> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c<br class=""> @@ -213,6 +213,11 @@ int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool run_pm)<br class=""> return r;<br class=""> }<br class=""> <br class=""> +int amdgpu_amdkfd_resume_processes(void)<br class=""> +{<br class=""> + return kgd2kfd_resume_processes();<br class=""> +}<br class=""> +<br class=""> int amdgpu_amdkfd_pre_reset(struct amdgpu_device *adev)<br class=""> {<br class=""> int r = 0;<br class=""> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h<br class=""> index f8b9f27adcf5..803306e011c3 100644<br class=""> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h<br class=""> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h<br class=""> @@ -140,6 +140,7 @@ void amdgpu_amdkfd_fini(void);<br class=""> void amdgpu_amdkfd_suspend(struct amdgpu_device *adev, bool run_pm);<br class=""> int amdgpu_amdkfd_resume_iommu(struct amdgpu_device *adev);<br class=""> int amdgpu_amdkfd_resume(struct amdgpu_device *adev, bool run_pm);<br class=""> +int amdgpu_amdkfd_resume_processes(void);<br class=""> void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,<br class=""> const void *ih_ring_entry);<br class=""> void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev);<br class=""> @@ -347,6 +348,7 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd);<br class=""> void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm);<br class=""> int kgd2kfd_resume_iommu(struct kfd_dev *kfd);<br class=""> int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm);<br class=""> +int kgd2kfd_resume_processes(void);<br class=""> int kgd2kfd_pre_reset(struct kfd_dev *kfd);<br class=""> int kgd2kfd_post_reset(struct kfd_dev *kfd);<br class=""> void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry);<br class=""> @@ -393,6 +395,11 @@ static inline int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)<br class=""> return 0;<br class=""> }<br class=""> <br class=""> +static inline int kgd2kfd_resume_processes(void)<br class=""> +{<br class=""> + return 0;<br class=""> +}<br class=""> +<br class=""> static inline int kgd2kfd_pre_reset(struct kfd_dev *kfd)<br class=""> {<br class=""> return 0;<br class=""> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br class=""> index fa4a9f13c922..5827b65b7489 100644<br class=""> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br class=""> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c<br class=""> @@ -4004,6 +4004,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)<br class=""> if (drm_dev_is_unplugged(adev_to_drm(adev)))<br class=""> amdgpu_device_unmap_mmio(adev);<br class=""> <br class=""> + amdgpu_amdkfd_resume_processes();<br class=""> }<br class=""> <br class=""> void amdgpu_device_fini_sw(struct amdgpu_device *adev)<br class=""> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> index 62aa6c9d5123..ef05aae9255e 100644<br class=""> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> @@ -714,6 +714,19 @@ int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)<br class=""> return ret;<br class=""> }<br class=""> <br class=""> +/* for non-runtime resume only */<br class=""> +int kgd2kfd_resume_processes(void)<br class=""> +{<br class=""> + int count;<br class=""> +<br class=""> + count = atomic_dec_return(&kfd_locked);<br class=""> + WARN_ONCE(count < 0, "KFD suspend / resume ref. error");<br class=""> + if (count == 0)<br class=""> + return kfd_resume_all_processes();<br class=""> +<br class=""> + return 0;<br class=""> +}<br class=""> </blockquote> <br class=""> It doesn't make sense to me to just increment kfd_locked in<br class=""> kgd2kfd_suspend to only decrement it again a few functions down the<br class=""> road.<br class=""> <br class=""> I suggest this instead - you only incrmemnt if not during PCI remove<br class=""> <br class=""> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> b/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> index 1c2cf3a33c1f..7754f77248a4 100644<br class=""> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c<br class=""> @@ -952,11 +952,12 @@ bool kfd_is_locked(void)<br class=""> <br class=""> void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)<br class=""> {<br class=""> +<br class=""> if (!kfd->init_complete)<br class=""> return;<br class=""> <br class=""> /* for runtime suspend, skip locking kfd */<br class=""> - if (!run_pm) {<br class=""> + if (!run_pm && !drm_dev_is_unplugged(kfd->ddev)) {<br class=""> /* For first KFD device suspend all the KFD processes */<br class=""> if (atomic_inc_return(&kfd_locked) == 1)<br class=""> kfd_suspend_all_processes();<br class=""> <br class=""> <br class=""> Andrey<br class=""> <br class=""> <br class=""> <br class=""> <blockquote type="cite" class="">+<br class=""> int kgd2kfd_resume_iommu(struct kfd_dev *kfd)<br class=""> {<br class=""> int err = 0;</blockquote> </blockquote> </blockquote> </div> </blockquote> </div> <br class=""> </blockquote> </div> </div> </blockquote> </div> <br class=""> </blockquote> </div> </blockquote> </div> <br class=""> </div> </blockquote> </div> </div> </blockquote> </div> <br class=""> </body> </html>