Re: [Bug] PCI: Enable INTx if BIOS left them disabled - triggers during rescan

[Bjorn Helgaas]

[+cc Yinghai, Rajat]

On Fri, Mar 7, 2014 at 7:33 AM, Andreas Noever <andreas.noever@xxxxxxxxx> wrote:
> Hi,
>
> After upgrading to the latest RC I noticed that suprise removal
> stopped working. Linux did not notice that the devices where gone.
> Bisection points to
>
> 1f42db786b14a31bf807fc41ee5583a00c08fcb1 PCI: Enable INTx if BIOS left
> them disabled
> http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=1f42db786b14a31bf807fc41ee5583a00c08fcb1
>
> It seems that the above patch is triggered on the bridge itself (!)
> when a new device is added below it. At this point the hotplug driver
> for the bridge has already enabled MSI. Reenabling INTX kills MSI and
> prevents later suprise removal notifications.
>
> The following stacktrace is from a "echo 1 > rescan" on the bridge:
> do_pci_enable_device+0x59/0x140
> pci_reenable_device+0x1f/0x30
> pci_assign_unassigned_bridge_resources+0x123/0x160
> pci_rescan_bus_bridge_resize+0x28/0x40
> dev_bus_rescan_store+0x85/0xa0
> dev_attr_store+0x18/0x30
> sysfs_kf_write+0x3d/0x50
> kernfs_fop_write+0xd2/0x140
> vfs_write+0xba/0x1e0
> SyS_write+0x49/0xa0
> system_call_fastpath+0x1a/0x1f
>
> Similarly for a hotplug event:
> do_pci_enable_device+0x59/0x140
> pci_reenable_device+0x1f/0x30
> pci_assign_unassigned_bridge_resources+0x123/0x160
> pciehp_configure_device+0x90/0x160
> pciehp_enable_slot+0x163/0x280
> pciehp_power_thread+0xb8/0xe0
> process_one_work+0x167/0x420
> worker_thread+0x121/0x3a0
>
> In both cases DisINTx is turned off and the hotplug driver stops
> reacting to events.

Thanks a lot for noticing and bisecting this!

I opened a bugzilla report at https://bugzilla.kernel.org/show_bug.cgi?id=71691

It seems like clearing DisINTx has some effect on MSI.  I don't see
anything in the spec that would suggest this (I'm looking at the PCIe
r3.0 spec, sec 7.5.1.1).

Can somebody point out a connection between DisINTx and MSI?  If not,
maybe we'll need some sort of quirk to deal with this.

Andrea, can you attach a complete dmesg log and "lspci -vv" output to
the bugzilla?

Thanks again!

Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/