Thanks, Adrian, for the bisection and detailed debugging! ----- Forwarded message from bugzilla-daemon@xxxxxxxxxx ----- https://bugzilla.kernel.org/show_bug.cgi?id=216644 Summary: Host OS hangs when enabling VMD in UEFI setup Kernel Version: 6.1-rc2 Regression: No Created attachment 303108 --> https://bugzilla.kernel.org/attachment.cgi?id=303108&action=edit OS Log (Serial Log) When enabling VMD in BIOS setup, the host OS cannot boot successfully with the following error message: [ 8.986310] vmd 0000:64:05.5: PCI host bridge to bus 10000:00 ... [ 9.674113] vmd 0000:64:05.5: Bound to PCI domain 10000 ... [ 33.592638] DMAR: VT-d detected Invalidation Queue Error: Reason f [ 33.592640] DMAR: VT-d detected Invalidation Time-out Error: SID ffff [ 33.599853] DMAR: VT-d detected Invalidation Completion Error: SID ffff [ 33.607339] DMAR: QI HEAD: UNKNOWN qw0 = 0x0, qw1 = 0x0 [ 33.621143] DMAR: QI PRIOR: UNKNOWN qw0 = 0x0, qw1 = 0x0 [ 33.627366] DMAR: Invalidation Time-out Error (ITE) cleared *** Hardware Info *** Platform: skylake-D purley platform VMD: 8086:201d # lspci -s 0000:64:05.5 -nn 0000:64:05.5 RAID bus controller [0104]: Intel Corporation Volume Management Device NVMe RAID Controller [8086:201d] (rev 04) *** Detail Info *** `git bisect` points the following offending patch (commit: 6aab5622296b): commit 6aab5622296b990024ee67dd7efa7d143e7558d0 Author: Nirmal Patel <nirmal.patel@xxxxxxxxxxxxxxx> Date: Tue Nov 16 15:11:36 2021 -0700 PCI: vmd: Clean up domain before enumeration During VT-d pass-through, the VMD driver occasionally fails to enumerate underlying NVMe devices when repetitive reboots are performed in the guest OS. The issue can be resolved by resetting VMD root ports for proper enumeration and triggering secondary bus reset which will also propagate reset through downstream bridges. Link: https://lore.kernel.org/r/20211116221136.85134-1-nirmal.patel@xxxxxxxxxxxxxxx Signed-off-by: Nirmal Patel <nirmal.patel@xxxxxxxxxxxxxxx> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx> Reviewed-by: Jon Derrick <jonathan.derrick@xxxxxxxxx> *** Debugging Info *** 1. Reverting 6aab5622296b on top of 6.1-rc2 can fix the issue. 2. Comment out for calling vmd_domain_reset() can also fix the issue. So, it looks like the function memset_io() causes the issue. static void vmd_domain_reset(struct vmd_dev *vmd) { ... for (bus = 0; bus < max_buses; bus++) { for (dev = 0; dev < 32; dev++) { ... memset_io(base + PCI_IO_BASE, 0, PCI_ROM_ADDRESS1 - PCI_IO_BASE); } } } } 3. pci_reset_bus() returns -25 because 'slot' or 'bus->self' is NULL. 4. We have 4 disks attached to VMD: # nvme list Node Generic SN Model Namespace Usage Format FW Rev --------------------- --------------------- -------------------- ---------------------------------------- --------- -------------------------- ---------------- -------- /dev/nvme3n1 /dev/ng3n1 222639A46A39 Micron_7450_MTFDKBA960TFR 1 11.48 GB / 960.20 GB 512 B + 0 B E2MU111 /dev/nvme2n1 /dev/ng2n1 222639A46A30 Micron_7450_MTFDKBA960TFR 1 4.18 GB / 960.20 GB 512 B + 0 B E2MU111 /dev/nvme1n1 /dev/ng1n1 BTLJ849201CE1P0I SSDPELKX010T8L 1 1.00 TB / 1.00 TB 512 B + 0 B VCV1LZ37 /dev/nvme0n1 /dev/ng0n1 BTLJ849201BS1P0I SSDPELKX010T8L 1 1.00 TB / 1.00 TB 512 B + 0 B VCV1LZ37 Any thoughts? Thanks for the help. ----- End forwarded message -----
