On 05/05/2025 19:43, Bjorn Helgaas wrote:
On Mon, May 05, 2025 at 11:30:58AM +0200, Karolina Stolarek wrote:
On 01/05/2025 23:43, Bjorn Helgaas wrote:
I'm (finally) getting back to this series because it really needs to
make v6.16.
It would definitely be nice to determine the log level once instead of
several times, but I'm not sure I like passing "level" through the
whole chain because it seems like a lot of change to get that benefit:
- it changes the prototype for __aer_print_error(),
aer_print_error(), and aer_process_err_devices()
- it removes the info->severity test from aer_print_error(), but
leaves it in __aer_print_error() and pci_print_aer(), which need
it for other reasons
All these functions take a pointer to a struct aer_err_info, and if we
want to compute the log level once, maybe we could stash the result in
struct aer_err_info, similar to what we did with ratelimited[] here:
https://lore.kernel.org/all/20250321015806.954866-7-pandoh@xxxxxxxxxx/
I think that would be a good compromise between these two approaches.
I'm rebasing this series to v6.15-rc1 and will post a v6 proposal
soon.
Do you plan to include changes suggested in the thread or just rebase the
series?
Yes.
Also, it's still unclear to me how to approach the sysfs patch, both in the
context of the ratelimit refactor (which, some of it, is in the next
tree)[1] and the value that should be exposed in the attribute. We have
control only over the burst but not the interval. When we deal with high
rates of errors, we may want to increase the time window to see if the flood
is out of ordinary or is it constant.
Unclear to me, too. Might have to revisit that.
I understand. Let me know if you need any help with the series.
All the best,
Karolina
[1] - https://lore.kernel.org/all/b0883f20-c337-40bb-b564-c535a162bf54@paulmck-laptop/
