linux-kernel - Re: [PATCH v3 2/2] PCI/AER: Split the AER stats into multiple sysfs attributes

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 29 Aug 2019 18:18:37 -0500
From:   Bjorn Helgaas <helgaas@...nel.org>
To:     Rajat Jain <rajatja@...gle.com>
Cc:     gregkh@...uxfoundation.com, linux-pci@...r.kernel.org,
        linux-kernel@...r.kernel.org, rajatxjain@...il.com
Subject: Re: [PATCH v3 2/2] PCI/AER: Split the AER stats into multiple sysfs
 attributes

Hi Rajat,

On Tue, Aug 27, 2019 at 03:21:45PM -0700, Rajat Jain wrote:
> Split the AER stats into multiple sysfs atributes. Note that
> this changes the ABI of the AER stats, but hopefully, there
> aren't active users that need to change. This is how the AERs
> are being exposed now:
> 
> localhost /sys/devices/pci0000:00/0000:00:1c.0/aer_stats # ls -l

Possible s/aer_stats/aer/ to make the path shorter?

> -r--r--r--. 1 root root 4096 Aug 20 16:35 correctable_bit0_RxErr
> -r--r--r--. 1 root root 4096 Aug 20 16:35 correctable_bit12_Timeout
> -r--r--r--. 1 root root 4096 Aug 20 16:35 correctable_bit13_NonFatalErr
> ...

> -r--r--r--. 1 root root 4096 Aug 20 16:35 fatal_bit0_Undefined
> -r--r--r--. 1 root root 4096 Aug 20 16:35 fatal_bit12_TLP
> -r--r--r--. 1 root root 4096 Aug 20 16:35 fatal_bit13_FCP
> ...
> -r--r--r--. 1 root root 4096 Aug 20 16:35 nonfatal_bit0_Undefined
> -r--r--r--. 1 root root 4096 Aug 20 16:35 nonfatal_bit12_TLP
> -r--r--r--. 1 root root 4096 Aug 20 16:35 nonfatal_bit13_FCP
> ...

The AER registers are named "Correctable Error Status" and
"Uncorrectable Error Status".  Fatal & nonfatal errors are both
reported in the Uncorrectable Error Status register; the distinction
comes from the Uncorrectable Error Severity register.

E.g., there's only one bit in the Uncorrectable Error Status register
for "Poisoned TLB Received" ("bit12_TLP" above), and it's fatal or
nonfatal depending on the Error Severity setting.

So I propose that you expose "correctable" files and "uncorrectable"
files instead of "correctable", "fatal", and "nonfatal".  Then if you
need the severity information you could add a new file for
"uncorrectable severity".

IIUC these files are counts of how many errors have been logged.
Maybe add "_count" at the end?  I think that would be more informative
than the "bitN" part, although it's not completely obvious how to map
some of these (TLP, FCP, DLP, SDES) to the spec; maybe they could be
expanded?

> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_device_err_cor
> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_device_err_fatal
> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_device_err_nonfatal
> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_rootport_err_cor
> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_rootport_err_fatal
> -r--r--r--. 1 root root 4096 Aug 20 16:35 total_rootport_err_nonfatal
> localhost /sys/devices/pci0000:00/0000:00:1c.0/aer_stats #
> 
> Each file is has a single counter value. Single file containing all

s/is has/contains/

Bjorn