Running Adobe Flash triggers PCI bus failure

Running Adobe Flash triggers PCI bus failure

[aw ful]

I have a hardware system that is otherwise extremely stable, however,
after 1 minute to 1 day of viewing Flash videos via Firefox a PCI bus
fails and manifests in several different ways, including massive
ethernet errors and frozen sata devices. This failure can also be
invoked using multiple Live distributions, where the only thing I add
is installing the Flash plugin from Adobe. Initially believed to be a
hardware problem, I now believe it is a long term obscure bug as I
have seen related symptoms in various bug reports which were never
clearly resolved, and never associated with the flash plugin.

I am writing to the kernel devs as I can easily make this occur on two
distinct Linux distributions - standard Fedora 19, 20 Live, and 21 TC6
Live, and Linux Mint 17 Live, experienced across many recent 3.14.17,
3.16.2 kernel versions. The failure occurs on a Asrock AM3+ system
with AMD NB and SB. I was not clear how and where to write a further
bug report because I can now demonstrate it on multiple distributions.
Any assistance is appreciated.

Detailed information can be found here:
https://bugzilla.redhat.com/show_bug.cgi?id=1135801
which includes dmesg and lspci -vv

Thank You for your time.
awfl

Re: Running Adobe Flash triggers PCI bus failure

[Bjorn Helgaas]

[+cc Josh, Myron]

On Wed, Sep 10, 2014 at 10:10 AM, aw ful <awfl.awfl@gmail.com> wrote:
> I have a hardware system that is otherwise extremely stable, however,
> after 1 minute to 1 day of viewing Flash videos via Firefox a PCI bus
> fails and manifests in several different ways, including massive
> ethernet errors and frozen sata devices. This failure can also be
> invoked using multiple Live distributions, where the only thing I add
> is installing the Flash plugin from Adobe. Initially believed to be a
> hardware problem, I now believe it is a long term obscure bug as I
> have seen related symptoms in various bug reports which were never
> clearly resolved, and never associated with the flash plugin.
>
> I am writing to the kernel devs as I can easily make this occur on two
> distinct Linux distributions - standard Fedora 19, 20 Live, and 21 TC6
> Live, and Linux Mint 17 Live, experienced across many recent 3.14.17,
> 3.16.2 kernel versions. The failure occurs on a Asrock AM3+ system
> with AMD NB and SB. I was not clear how and where to write a further
> bug report because I can now demonstrate it on multiple distributions.
> Any assistance is appreciated.
>
> Detailed information can be found here:
> https://bugzilla.redhat.com/show_bug.cgi?id=1135801
> which includes dmesg and lspci -vv

Thanks for the report!  It sounds like you've reproduced the problem
on a non-RedHat kernel, so can you open a report at
http://bugzilla.kernel.org using an unmodified upstream kernel,
please?

Please attach "lspci -vv" output (as root) from before the problem and
another from after, as well as the complete dmesg log.  You mention
several other bugzilla reports; please include URLs to those as well.

You mention "taking down PCI bus"; what specifically do you mean by
that?  I do see some FFFFFFFF data from ata7 and ata8 in your dmesg,
which can be a symptom of a device not responding.  ata7 and ata8 seem
to be on 06:00.0, and your lspci shows UnsupReq+ set in that device's
AER Capability.  Please also attach "lspci -xxxxs06:00.0" output so we
can look at that in more detail.

The atl1c device is at 03:00.0, and your lspci also shows UnsupReq+
for it, so please also attach "lspci -xxxxs03:00.0" output.

Does your kernel have CONFIG_PCIEAER=y?  If not, can you turn that on
and see if you see anything useful in dmesg when the problem occurs?

Bjorn

Re: Running Adobe Flash triggers PCI bus failure

[aw ful]

Hello;

When I mention taking down the PCI bus, what I imply is that when I
compare the before and after status of devices on the PCI buses using
"lspci -vv",  I saw many 0xffffffffs and device memory and ports
marked as "disabled". I (naively and likely wrongly) understood that
if the PCI bus (or devices on them) detects a bus fault, or suffer
failure, the devices on it become disabled as a result? Either way,
that is where I got the idea that a bus is down, not just a single
specific device.

During troubleshooting I removed the ata7 and ata8 devices you
mentioned from the machine (as well as all other ata devices except
for the main disk), and all extraneous pci cards. Still crashed as
evidenced by the onboard ethernet failing. I then disabled the onboard
ethernet and installed a separate PCI ethernet card and the network
still failed (i'll look at my notes, but that is what I recall).
Another reason I was looking at the PCI bus as the failure point.

And all these devices work flawlessly, heavily loaded, by themselves
or in tandem. As well when running heavy graphics and steam games
under fglrx. But only when an impending flash fault hasn't yet created
a failure.

It is outside my experience level, but I will try to git an
appropriate kernel, compile it, and get it to run on my system if
possible - it is my sole working machine.

If I am successful, I will attempt to open a bug report with the
information you have requested.

And yes, on my Fedora 19 3.14.17-100 config the default is
CONFIG_PCIEAER=y so the dmesg you see is complete. I did try a debug
kernels, but the output was always as sparse.

Thank You for your time.