[+cc Grant, Rajat] On Tue, Jan 17, 2023 at 06:15:28PM +0000, Alexey Bogoslavsky wrote: > >From: Keith Busch <kbusch@xxxxxxxxxx> > >Sent: Tuesday, January 17, 2023 5:55 PM > >To: Alexey Bogoslavsky <Alexey.Bogoslavsky@xxxxxxx> > >Cc: linux-pci@xxxxxxxxxxxxxxx; bhelgaas@xxxxxxxxxx; 'hch@xxxxxx' <hch@xxxxxx> > >Subject: Re: [PATCH 1/1] PCI/AER: Ignore correctable error reports for SN730 WD SSD > > >On Mon, Jan 16, 2023 at 06:32:54PM +0000, Alexey Bogoslavsky wrote: > >> From: Alexey Bogoslavsky <mailto:Alexey.Bogoslavsky@xxxxxxx> > >> > >> A bug was found in SN730 WD SSD that causes occasional false AER reporting > >> of correctable errors. While functionally harmless, this causes error > >> messages to appear in the system log (dmesg) which, in turn, causes > >> problems in automated platform validation tests. Since the issue can not > >> be fixed by FW, customers asked for correctable error reporting to be > >> quirked out in the kernel for this particular device. > > > >> The patch was manually verified. It was checked that correctable errors > >> are still detected but ignored for the target device (SN730), and are both > >> detected and reported for devices not affected by this quirk. > > >If you're just going to have the kernel ignore these, are you not able > >to suppress the ERR_COR message at the source? Have the following > >options been tried? > > > a. Disabling Correctable Error Reporting Enable in Device Control > > Register; i.e. mask out PCI_EXP_DEVCTL_CERE. > > b. Setting AER Correctable Error Mask Register to all 1's > > >I think it's usually possible for firmware to hardwire these. If the > > I believe these options were discussed but deemed non-viable. I'll > double check anyway > > >If firmware can't do that, quirking the kernel to always disable > >reporting sounds like a better option. If either of the above fail > >to suppress the error messages, then I guess having the kernel > >ignore it is the only option. > > This could probably work. I'll discuss this with our FW team to make > sure the issue can be resolved this way. Thank you Any resolution on this FW possibility? We have patches in progress to rate-limit correctable error messages and make them KERN_INFO instead of KERN_WARN [1], but I don't think that's going to be a good enough solution for you because nobody wants to see even an informational message every 5 seconds if the message is useless. If firmware on the device can turn off these errors, that would be the best solution. If not, I think your quirk is a reasonable approach and just needs a litle polishing per the previous comments. Bjorn [1] https://lore.kernel.org/r/20230317175109.3859943-1-grundler@xxxxxxxxxxxx
