On 20/03/2025 19:17, Bjorn Helgaas wrote:
Maybe there's CXL magic that I missed. It looks like Terry's series
changes some of this path. And GHES also currently uses
pci_print_aer(). Some sample logs at [1,2].
Maybe that's it, thanks a lot for the pointers.
Looking at v6.14-rc1, only aer_print_error() logs the "error status"
string, and only pci_print_aer() logs "aer_status", "aer_layer", etc.
The previous path is:
pci_print_aer
pci_err("aer_status: 0x%08x, aer_mask: 0x%08x\n") <--
__aer_print_error
pci_err("aer_layer=%s, aer_agent=%s\n") <--
pcie_print_tlp_log
New path is:
aer_print_error
pci_printk("PCIe Bus Error: severity=%s, type=%s, (%s)\n")
pci_printk(" device [%04x:%04x] error status/mask=%08x/%08x\n)
__aer_print_error
pcie_print_tlp_log
So I expected that the lines I marked in pci_print_aer() would be
different.
Hmm, that seems to be the case. But still, the question is if going with
the new format that matches what's in AER a bad or disruptive thing. I'd
like to try going in the direction of using one way of reporting AER
errors, if possible.
I will send v2 on Monday (with the memset move) and we can keep
discussing other changes in the patch.
All the best,
Karolina
Bjorn
[1] https://lore.kernel.org/lkml/2149597.8uJZFlvqrj@xrated/T/
[2] https://lore.kernel.org/all/e8a58616-aeae-ad78-d496-6dfcef4ddcaa@xxxxxxx/T/
