Re: How to reduce PCI initialization from 5 s (1.5 s adding them to IOMMU groups)
On Fri, Nov 05, 2021 at 12:56:09PM +0100, Paul Menzel wrote: > Dear Linux folks, > > > On a PowerEdge T440/021KCD, BIOS 2.11.2 04/22/2021, Linux 5.10.70 takes > almost five seconds to initialize PCI. According to the timestamps, 1.5 s > are from assigning the PCI devices to the 142 IOMMU groups. > > ``` > $ lspci | wc -l > 281 > $ dmesg > […] > [2.918411] PCI: Using host bridge windows from ACPI; if necessary, use > "pci=nocrs" and report a bug > [2.933841] ACPI: Enabled 5 GPEs in block 00 to 7F > [2.973739] ACPI: PCI Root Bridge [PC00] (domain [bus 00-16]) > [2.980398] acpi PNP0A08:00: _OSC: OS supports [ExtendedConfig ASPM > ClockPM Segments MSI HPX-Type3] > [2.989457] acpi PNP0A08:00: _OSC: platform does not support [LTR] > [2.995451] acpi PNP0A08:00: _OSC: OS now controls [PME PCIeCapability] > [3.001394] acpi PNP0A08:00: FADT indicates ASPM is unsupported, using > BIOS configuration > [3.010511] PCI host bridge to bus :00 > […] > [6.233508] system 00:05: [io 0x1000-0x10fe] has been reserved > [6.239420] system 00:05: Plug and Play ACPI device, IDs PNP0c02 (active) > [6.239906] pnp: PnP ACPI: found 6 devices For ~280 PCI devices, (6.24-2.92)/280 = 0.012 s/dev. On my laptop I have about (.66-.37)/36 = 0.008 s/dev (on v5.4), so about the same ballpark. Faster would always be better, of course. I assume this is not really a regression? > [6.989016] pci :d7:05.0: disabled boot interrupts on device > [8086:2034] > [6.996063] PCI: CLS 0 bytes, default 64 > [7.08] Trying to unpack rootfs image as initramfs... > [7.065281] Freeing initrd memory: 5136K > […] > [7.079098] DMAR: dmar7: Using Queued invalidation > [7.083983] pci :00:00.0: Adding to iommu group 0 > […] > [8.537808] pci :d7:17.1: Adding to iommu group 141 I don't have this iommu stuff turned on and don't know what's happening here. > Is there anything that could be done to reduce the time? > > > Kind regards, > > Paul ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: How to reduce PCI initialization from 5 s (1.5 s adding them to IOMMU groups)s
Dear Linux folks, Am 05.11.21 um 12:56 schrieb Paul Menzel: On a PowerEdge T440/021KCD, BIOS 2.11.2 04/22/2021, Linux 5.10.70 takes almost five seconds to initialize PCI. According to the timestamps, 1.5 s are from assigning the PCI devices to the 142 IOMMU groups. ``` $ lspci | wc -l 281 $ dmesg […] [ 2.918411] PCI: Using host bridge windows from ACPI; if necessary, use "pci=nocrs" and report a bug [ 2.933841] ACPI: Enabled 5 GPEs in block 00 to 7F [ 2.973739] ACPI: PCI Root Bridge [PC00] (domain [bus 00-16]) [ 2.980398] acpi PNP0A08:00: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI HPX-Type3] [ 2.989457] acpi PNP0A08:00: _OSC: platform does not support [LTR] [ 2.995451] acpi PNP0A08:00: _OSC: OS now controls [PME PCIeCapability] [ 3.001394] acpi PNP0A08:00: FADT indicates ASPM is unsupported, using BIOS configuration [ 3.010511] PCI host bridge to bus :00 […] [ 6.233508] system 00:05: [io 0x1000-0x10fe] has been reserved [ 6.239420] system 00:05: Plug and Play ACPI device, IDs PNP0c02 (active) [ 6.239906] pnp: PnP ACPI: found 6 devices […] [ 6.989016] pci :d7:05.0: disabled boot interrupts on device [8086:2034] [ 6.996063] PCI: CLS 0 bytes, default 64 [ 7.08] Trying to unpack rootfs image as initramfs... [ 7.065281] Freeing initrd memory: 5136K […] [ 7.079098] DMAR: dmar7: Using Queued invalidation [ 7.083983] pci :00:00.0: Adding to iommu group 0 […] [ 8.537808] pci :d7:17.1: Adding to iommu group 141 [ 8.571191] DMAR: Intel(R) Virtualization Technology for Directed I/O [ 8.577618] PCI-DMA: Using software bounce buffering for IO (SWIOTLB) […] ``` Is there anything that could be done to reduce the time? I created an issue at the Kernel.org Bugzilla, and attached the output of `dmesg` there [1]. Kind regards, Paul [1]: https://bugzilla.kernel.org/show_bug.cgi?id=214953 ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
How to reduce PCI initialization from 5 s (1.5 s adding them to IOMMU groups)
Dear Linux folks, On a PowerEdge T440/021KCD, BIOS 2.11.2 04/22/2021, Linux 5.10.70 takes almost five seconds to initialize PCI. According to the timestamps, 1.5 s are from assigning the PCI devices to the 142 IOMMU groups. ``` $ lspci | wc -l 281 $ dmesg […] [2.918411] PCI: Using host bridge windows from ACPI; if necessary, use "pci=nocrs" and report a bug [2.933841] ACPI: Enabled 5 GPEs in block 00 to 7F [2.973739] ACPI: PCI Root Bridge [PC00] (domain [bus 00-16]) [2.980398] acpi PNP0A08:00: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI HPX-Type3] [2.989457] acpi PNP0A08:00: _OSC: platform does not support [LTR] [2.995451] acpi PNP0A08:00: _OSC: OS now controls [PME PCIeCapability] [3.001394] acpi PNP0A08:00: FADT indicates ASPM is unsupported, using BIOS configuration [3.010511] PCI host bridge to bus :00 […] [6.233508] system 00:05: [io 0x1000-0x10fe] has been reserved [6.239420] system 00:05: Plug and Play ACPI device, IDs PNP0c02 (active) [6.239906] pnp: PnP ACPI: found 6 devices […] [6.989016] pci :d7:05.0: disabled boot interrupts on device [8086:2034] [6.996063] PCI: CLS 0 bytes, default 64 [7.08] Trying to unpack rootfs image as initramfs... [7.065281] Freeing initrd memory: 5136K […] [7.079098] DMAR: dmar7: Using Queued invalidation [7.083983] pci :00:00.0: Adding to iommu group 0 […] [8.537808] pci :d7:17.1: Adding to iommu group 141 [8.571191] DMAR: Intel(R) Virtualization Technology for Directed I/O [8.577618] PCI-DMA: Using software bounce buffering for IO (SWIOTLB) […] ``` Is there anything that could be done to reduce the time? Kind regards, Paul ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [git pull] IOMMU Updates for Linux v5.16
On 2021/11/5 15:35, Joerg Roedel wrote: On Thu, Nov 04, 2021 at 11:10:53AM -0700, Linus Torvalds wrote: That too seems to be stale and unused since commit 9bfecd058339 ("x86/cpufeatures: Force disable X86_FEATURE_ENQCMD and remove update_pasid()") I left it alone because I didn't want to do any extra development in the merge commit itself, but it looks to me like there's some room for more cleanips. No? Yes, looks like this define is unused now. I talk to Baolu whether this can be cleaned up too, Baolu? Yes. PASID_DISABLED is unused after commit 9bfecd058339 ("x86/cpufeatures: Force disable X86_FEATURE_ENQCMD and remove update_pasid()"). and commit 00ecd5401349a ("iommu/vt-d: Clean up unused PASID updating functions") Best regards, baolu ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [git pull] IOMMU Updates for Linux v5.16
On Thu, Nov 04, 2021 at 11:10:53AM -0700, Linus Torvalds wrote: > That too seems to be stale and unused since commit 9bfecd058339 > ("x86/cpufeatures: Force disable X86_FEATURE_ENQCMD and remove > update_pasid()") > > I left it alone because I didn't want to do any extra development in > the merge commit itself, but it looks to me like there's some room for > more cleanips. No? Yes, looks like this define is unused now. I talk to Baolu whether this can be cleaned up too, Baolu? Thanks, Joerg ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu