Date: Fri, 07 Aug 2026 03:51:30 +0000 From: Kevin Bowling <kbowling@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org Subject: git: 5aab6164161d - main - pci: Ignore SR-IOV VFs when tuning MPS Message-ID: <6a755642.215b7.2c36e3de@gitrepo.freebsd.org>
index | next in thread | raw e-mail
The branch main has been updated by kbowling: URL: https://cgit.FreeBSD.org/src/commit/?id=5aab6164161db397d84e5fff88e1af1a9e405875 commit 5aab6164161db397d84e5fff88e1af1a9e405875 Author: Kevin Bowling <kbowling@FreeBSD.org> AuthorDate: 2026-08-06 06:39:20 +0000 Commit: Kevin Bowling <kbowling@FreeBSD.org> CommitDate: 2026-08-07 03:50:15 +0000 pci: Ignore SR-IOV VFs when tuning MPS The VF Device Control MPS and MRRS fields are reserved and preserved. VF transactions use the PF MPS, so a hardwired VF value must not be used to retune the shared PCIe hierarchy. Document the previously undocumented tuning knob and clarify why a VF may continue to display its reserved hardwired value. This fixes an instant crash/reboot on my Zen3 system with 82599 VFs. MFC after: 1 week --- share/man/man4/pci.4 | 13 ++++++++++++- sys/dev/pci/pci.c | 7 +++++++ 2 files changed, 19 insertions(+), 1 deletion(-) diff --git a/share/man/man4/pci.4 b/share/man/man4/pci.4 index 38a427e64f4f..71ca2f50ed90 100644 --- a/share/man/man4/pci.4 +++ b/share/man/man4/pci.4 @@ -22,7 +22,7 @@ .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF .\" SUCH DAMAGE. .\" -.Dd March 10, 2026 +.Dd August 6, 2026 .Dt PCI 4 .Os .Sh NAME @@ -589,6 +589,17 @@ MSI interrupts can be disabled by setting this tunable to 0. Enable support for extended Message Signalled Interrupts .Pq MSI-X . MSI-X interrupts can be disabled by setting this tunable to 0. +.It Va hw.pci.enable_mps_tune Pq Defaults to 1 +Configure PCI-express Maximum Payload Size +.Pq MPS +during device enumeration. +SR-IOV Virtual Functions are excluded because their MPS fields are reserved +and the Physical Function setting applies to them. +Configuration-space tools may therefore display a Virtual Function's +hardwired reserved value rather than the effective Physical Function MPS. +Maximum Read Request Size +.Pq MRRS +is not modified by this tuner. .It Va hw.pci.enable_pcie_ei Pq Defaults to 0 Enable support for PCI-express Electromechanical Interlock. .It Va hw.pci.enable_pcie_hp Pq Defaults to 1 diff --git a/sys/dev/pci/pci.c b/sys/dev/pci/pci.c index 034f34e2dde2..b690c3412356 100644 --- a/sys/dev/pci/pci.c +++ b/sys/dev/pci/pci.c @@ -4420,6 +4420,13 @@ pcie_setup_mps(device_t dev) device_t root; uint16_t rmps, mmps, mps; + /* + * PCIe r4.0, sec 9.3.5.4 defines the VF MPS and MRRS fields as + * Reserved and Preserved, with the PF settings applying to the VF. + * Do not use the VF's hardwired value to configure the shared path. + */ + if ((dinfo->cfg.flags & PCICFG_VF) != 0) + return; if (dinfo->cfg.pcie.pcie_location == 0) return; root = pci_find_pcie_root_port(dev);home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6a755642.215b7.2c36e3de>