PCI: Work around Intel I210 ROM BAR overlap defect
authorBjorn Helgaas <bhelgaas@google.com>
Tue, 21 Dec 2021 16:45:07 +0000 (10:45 -0600)
committerBjorn Helgaas <bhelgaas@google.com>
Tue, 11 Jan 2022 15:33:10 +0000 (09:33 -0600)
Per PCIe r5, sec 7.5.1.2.4, a device must not claim accesses to its
Expansion ROM unless both the Memory Space Enable and the Expansion ROM
Enable bit are set.  But apparently some Intel I210 NICs don't work
correctly if the ROM BAR overlaps another BAR, even if the Expansion ROM is
disabled.

Michael reported that on a Kontron SMARC-sAL28 ARM64 system with U-Boot
v2021.01-rc3, the ROM BAR overlaps BAR 3, and networking doesn't work at
all:

  BAR 0: 0x40000000 (32-bit, non-prefetchable) [size=1M]
  BAR 3: 0x40200000 (32-bit, non-prefetchable) [size=16K]
  ROM:   0x40200000 (disabled) [size=1M]

  NETDEV WATCHDOG: enP2p1s0 (igb): transmit queue 0 timed out
  Hardware name: Kontron SMARC-sAL28 (Single PHY) on SMARC Eval 2.0 carrier (DT)
  igb 0002:01:00.0 enP2p1s0: Reset adapter

Previously, pci_std_update_resource() wrote the assigned ROM address to the
BAR only when the ROM was enabled.  This meant that the I210 ROM BAR could
be left with an address assigned by firmware, which might overlap with
other BARs.

Quirk these I210 devices so pci_std_update_resource() always writes the
assigned address to the ROM BAR, whether or not the ROM is enabled.

Link: https://lore.kernel.org/r/20211223163754.GA1267351@bhelgaas
Link: https://lore.kernel.org/r/20201230185317.30915-1-michael@walle.cc
Link: https://bugzilla.kernel.org/show_bug.cgi?id=211105
Reported-by: Michael Walle <michael@walle.cc>
Tested-by: Michael Walle <michael@walle.cc>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
drivers/pci/quirks.c
drivers/pci/setup-res.c
include/linux/pci.h

index 003950c738d26e0a584ede2d4dcb075f47789bb1..46ff04091fa3d7dbde3e77445f00a17f431a58e4 100644 (file)
@@ -5857,3 +5857,13 @@ static void nvidia_ion_ahci_fixup(struct pci_dev *pdev)
        pdev->dev_flags |= PCI_DEV_FLAGS_HAS_MSI_MASKING;
 }
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_NVIDIA, 0x0ab8, nvidia_ion_ahci_fixup);
+
+static void rom_bar_overlap_defect(struct pci_dev *dev)
+{
+       pci_info(dev, "working around ROM BAR overlap defect\n");
+       dev->rom_bar_overlap = 1;
+}
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1533, rom_bar_overlap_defect);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1536, rom_bar_overlap_defect);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1537, rom_bar_overlap_defect);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x1538, rom_bar_overlap_defect);
index 7f1acb3918d0c3c8e5304a6848a22ffd02886257..439ac5f5907a6ee5778025b3180ff268e3c3166b 100644 (file)
@@ -75,12 +75,16 @@ static void pci_std_update_resource(struct pci_dev *dev, int resno)
                 * as zero when disabled, so don't update ROM BARs unless
                 * they're enabled.  See
                 * https://lore.kernel.org/r/43147B3D.1030309@vc.cvut.cz/
+                * But we must update ROM BAR for buggy devices where even a
+                * disabled ROM can conflict with other BARs.
                 */
-               if (!(res->flags & IORESOURCE_ROM_ENABLE))
+               if (!(res->flags & IORESOURCE_ROM_ENABLE) &&
+                   !dev->rom_bar_overlap)
                        return;
 
                reg = dev->rom_base_reg;
-               new |= PCI_ROM_ADDRESS_ENABLE;
+               if (res->flags & IORESOURCE_ROM_ENABLE)
+                       new |= PCI_ROM_ADDRESS_ENABLE;
        } else
                return;
 
index 18a75c8e615cdb3ef068ff07cfe1552e2a30eeff..51c4a063f489228d2f92ba91ed843469df4c2f8d 100644 (file)
@@ -455,6 +455,7 @@ struct pci_dev {
        unsigned int    link_active_reporting:1;/* Device capable of reporting link active */
        unsigned int    no_vf_scan:1;           /* Don't scan for VFs after IOV enablement */
        unsigned int    no_command_memory:1;    /* No PCI_COMMAND_MEMORY */
+       unsigned int    rom_bar_overlap:1;      /* ROM BAR disable broken */
        pci_dev_flags_t dev_flags;
        atomic_t        enable_cnt;     /* pci_enable_device has been called */