On Thu, Jun 04, 2020 at 03:35:48PM +0300, Aya Levin wrote: > Hi > > I am writing to you regarding your commits titled: > 1. a99b646afa8a PCI: Disable PCIe Relaxed Ordering if unsupported > 2. 87e09cdec4da PCI: Disable Relaxed Ordering for some Intel processors > > While adding support to relaxed ordering on Mellanox Ethernet driver I > tried to avoid the Intel's bug with pcie_relaxed_ordering_enabled. > Expecting this to return False on Haswell and Broadwell CPU. I run > this API on: Intel(R) Xeon(R) CPU E5-2650 v3 @ 2.30GHz (I think it is > a Haswell, right?) and the API returned True. What am I missing? The quirk is based on Vendor/Device ID, which of course is a problem as new devices are released. What does "lspci -vn" show on your system? > In addition, I saw your comment in pci_configure_relaxed_ordering > (pasted below) the non-root ports are not handled since Peer-to-Peer > DMA is another can of warms. Could you elaborate on the complexity? > What is the effort to extend this to non-root ports? There's some discussion about this here and in other parts of the same thread: https://lore.kernel.org/linux-arm-kernel/20170809032503.GB7191@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/ > --------------------------------------------------------------------- > static void pci_configure_relaxed_ordering(struct pci_dev *dev) > { > struct pci_dev *root; > > /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */ > if (dev->is_virtfn) > return; > > if (!pcie_relaxed_ordering_enabled(dev)) > return; > > /* > * For now, we only deal with Relaxed Ordering issues with Root > * Ports. Peer-to-Peer DMA is another can of worms. > */ > root = pci_find_pcie_root_port(dev); > if (!root) > return; > > if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) { > pcie_capability_clear_word(dev, PCI_EXP_DEVCTL, > PCI_EXP_DEVCTL_RELAX_EN); > pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n"); > } > } > >
