On 7/16/2021 4:56 PM, Bjorn Helgaas wrote:
On Thu, May 13, 2021 at 03:03:14AM +0530, Stuart Hayes wrote:
The pcieport driver can fail to attach to a downstream port that doesn't
support bandwidth notification. This can happen when, in
pcie_port_device_register(), pci_init_service_irqs() tries (and fails) to
set up a bandwidth notification IRQ for a device that doesn't support it.
I'm trying to figure out exactly how this fails. The only failure
path in pcie_init_service_irqs() is when
pci_alloc_irq_vectors(PCI_IRQ_LEGACY) fails, which I guess means the
port's dev->irq was zero?
Yes... dev->irq is zero. Here's "lspci -v "for the device in case it helps:
66:00.0 PCI bridge: Broadcom / LSI PEX880xx PCIe Gen 4 Switch (rev b0)
(prog-if 00 [Normal decode])
Flags: bus master, fast devsel, latency 0, NUMA node 0
Bus: primary=66, secondary=67, subordinate=70, sec-latency=0
I/O behind bridge: [disabled]
Memory behind bridge: be000000-bf4fffff [size=21M]
Prefetchable memory behind bridge:
00000d0000000000-00000d00017fffff [size=24M]
Capabilities: [40] Power Management version 3
Capabilities: [68] Express Downstream Port (Slot-), MSI 00
Capabilities: [a4] Subsystem: Broadcom / LSI Device a064
Capabilities: [100] Device Serial Number 00-80-5e-10-56-39-53-50
Capabilities: [fb4] Advanced Error Reporting
Capabilities: [148] Virtual Channel
Capabilities: [b70] Vendor Specific Information: ID=0001 Rev=0
Len=010 <?>
Capabilities: [e00] Multicast
Capabilities: [f24] Access Control Services
And to even attempt legacy IRQs, either we had pcie_pme_no_msi() or
pcie_port_enable_irq_vec() failed?
pcie_port_enable_irq_vec() failed. I don't believe this device has the
ability to generate interrupts.
This patch changes get_port_device_capability() to look at the link
bandwidth notification capability bit in the link capabilities register of
the port, instead of assuming that all downstream ports have that
capability.
I think this needs:
Fixes: e8303bb7a75c ("PCI/LINK: Report degraded links via link bandwidth notification")
because even though b4c7d2076b4e ("PCI/LINK: Remove bandwidth
notification") removed *most* of e8303bb7a75c, it did not remove the
code in get_port_device_capability() that you're fixing here.
I can fix this up locally, no need to resend. I think the patch
itself is fine, just trying to understand the commit log.
Thank you!
Signed-off-by: Stuart Hayes <stuart.w.hayes@xxxxxxxxx>
Reviewed-by: Lukas Wunner <lukas@xxxxxxxxx>
---
changes from v1:
- corrected spelling errors in commit message
- added Lukas' reviewed-by:
---
drivers/pci/pcie/portdrv_core.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/pci/pcie/portdrv_core.c b/drivers/pci/pcie/portdrv_core.c
index e1fed6649c41..3ee63968deaa 100644
--- a/drivers/pci/pcie/portdrv_core.c
+++ b/drivers/pci/pcie/portdrv_core.c
@@ -257,8 +257,13 @@ static int get_port_device_capability(struct pci_dev *dev)
services |= PCIE_PORT_SERVICE_DPC;
if (pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM ||
- pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
- services |= PCIE_PORT_SERVICE_BWNOTIF;
+ pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
+ u32 linkcap;
+
+ pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &linkcap);
+ if (linkcap & PCI_EXP_LNKCAP_LBNC)
+ services |= PCIE_PORT_SERVICE_BWNOTIF;
+ }
return services;
}
--
2.27.0
