Re: [PATCH v1] PCI/AER: Handle Multi UnCorrectable/Correctable errors properly

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Mar 14, 2022 at 09:21:46AM -0700, Eric Badger wrote:
> On Sun, Mar 13, 2022 at 02:43:14PM -0700, Raj, Ashok wrote:
> > On Sun, Mar 13, 2022 at 02:52:20PM -0500, Bjorn Helgaas wrote:
> > > On Fri, Mar 11, 2022 at 02:58:07AM +0000, Kuppuswamy Sathyanarayanan wrote:
> > > > Currently the aer_irq() handler returns IRQ_NONE for cases without bits
> > > > PCI_ERR_ROOT_UNCOR_RCV or PCI_ERR_ROOT_COR_RCV are set. But this
> > > > assumption is incorrect.
> > > > 
> > > > Consider a scenario where aer_irq() is triggered for a correctable
> > > > error, and while we process the error and before we clear the error
> > > > status in "Root Error Status" register, if the same kind of error
> > > > is triggered again, since aer_irq() only clears events it saw, the
> > > > multi-bit error is left in tact. This will cause the interrupt to fire
> > > > again, resulting in entering aer_irq() with just the multi-bit error
> > > > logged in the "Root Error Status" register.
> > > > 
> > > > Repeated AER recovery test has revealed this condition does happen
> > > > and this prevents any new interrupt from being triggered. Allow to
> > > > process interrupt even if only multi-correctable (BIT 1) or
> > > > multi-uncorrectable bit (BIT 3) is set.
> > > > 
> > > > Reported-by: Eric Badger <ebadger@xxxxxxxxxxxxxxx>
> > > 
> > > Is there a bug report with any concrete details (dmesg, lspci, etc)
> > > that we can include here?
> > 
> > Eric might have more details to add when he collected numerous logs to get
> > to the timeline of the problem. The test was to stress the links with an
> > automated power off, this will result in some eDPC UC error followed by
> > link down. The recovery worked fine for several cycles and suddenly there
> > were no more interrupts. A manual rescan on pci would probe and device is
> > operational again.
> 
> The problem was originally discovered while performing a looping hot plug
> test. At hot remove time, one or more corrected errors usually appeared:
> 
> [256236.078151] pcieport 0000:89:02.0: AER: Corrected error received: 0000:89:02.0
> [256236.078154] pcieport 0000:89:02.0: AER: PCIe Bus Error: severity=Corrected, type=Physical Layer, (Receiver ID)
> [256236.088606] pcieport 0000:89:02.0: AER:   device [8086:347a] error status/mask=00000001/00000000
> [256236.097857] pcieport 0000:89:02.0: AER:    [ 0] RxErr                 
> [256236.152622] pcieport 0000:89:02.0: pciehp: Slot(400): Link Down
> [256236.152623] pcieport 0000:89:02.0: pciehp: Slot(400): Card not present
> [256236.152631] pcieport 0000:89:02.0: DPC: containment event, status:0x1f01 source:0x0000
> [256236.152632] pcieport 0000:89:02.0: DPC: unmasked uncorrectable error detected reason 0 ext_reason 0
> [256236.152634] pcieport 0000:89:02.0: AER: PCIe Bus Error: severity=Uncorrected (Fatal), type=Transaction Layer, (Receiver ID)
> [256236.164207] pcieport 0000:89:02.0: AER:   device [8086:347a] error status/mask=00000020/00100000
> [256236.173464] pcieport 0000:89:02.0: AER:    [ 5] SDES                   (First)
> [256236.278407] pci 0000:8a:00.0: Removing from iommu group 32
> [256237.500837] pcieport 0000:89:02.0: Data Link Layer Link Active not set in 1000 msec
> [256237.500842] pcieport 0000:89:02.0: link reset at upstream device 0000:89:02.0 failed
> [256237.500865] pcieport 0000:89:02.0: AER: Device recovery failed
> 
> The problematic case arose when 2 corrected errors arrived in a sequence like this:
> 
> 1. Correctable error triggered, bit 0 (ERR_COR) set in Root Error Status,
>    which now has value 0x1.
> 2. aer_irq() triggered, reads Root Error Status, finds value 0x1.
> 3. Second correctable error triggered, bit 1 (multiple ERR_COR) set in Root
>    Error Status, which now has value 0x3.
> 4. aer_irq() writes back 0x1 to Root Error Status, which now has value 0x2.
> 5. aer_irq() triggered again due to the second error, but, finding value 0x2
>    in Root Error Status, takes no action. Future interrupts are now inhibited.

Thanks for the additional details!

After this patch, I guess aer_irq() still reads 0x2
(PCI_ERR_ROOT_MULTI_COR_RCV), but now it writes 0x2 back which clears
PCI_ERR_ROOT_MULTI_COR_RCV.

In addition, aer_irq() will continue on to read PCI_ERR_ROOT_ERR_SRC,
which probably contains either 0 or junk left over from being captured
when PCI_ERR_ROOT_COR_RCV was set.

And aer_irq() will queue an e_src record with status ==
PCI_ERR_ROOT_MULTI_COR_RCV.  But since PCI_ERR_ROOT_COR_RCV is not set
in status, aer_isr_one_error() will do nothing, right?

That might not be *terrible* and is definitely better than not being
able to handle future interrupts.  But we basically threw away the
information that multiple errors occurred, and we queued an e_src
record that occupies space without being used for anything.

Bjorn



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux