On Tue, 17 Nov 2026 12:10:19 +0100 Pierre Gondois <[email protected]> wrote:
> The RAS consumer check removes: > - trace events when no daemon is registered > - CPER console report when one is registered > This ties tracing, PCIe recovery and CXL protocol delivery to daemon > registration, and can hide a fatal record when userspace cannot process > its trace event before a reset. > > Remove the RAS consumer registration condition and: > - always emit trace events > - print the CPER record only for CPER_SEV_FATAL > > [1] additionally suggested to have a cmdline parameter to dynamically > update the above policy. Keep a simple hard coded policy for now. > > [1] > https://lore.kernel.org/lkml/20240521183913.GGZkzqUSN1hdE-_OEY@fat_crate.local/ Use a link tag probably. Link: https://lore.kernel.org/lkml/20240521183913.GGZkzqUSN1hdE-_OEY@fat_crate.local/ #1 I'm not entirely convinced this is a good idea without the policy aspect. Still let us see what others think. > > Signed-off-by: Pierre Gondois <[email protected]> > --- > drivers/acpi/acpi_extlog.c | 6 ++---- > 1 file changed, 2 insertions(+), 4 deletions(-) > > diff --git a/drivers/acpi/acpi_extlog.c b/drivers/acpi/acpi_extlog.c > index e09ea571c32f5..16a16b12d8ccf 100644 > --- a/drivers/acpi/acpi_extlog.c > +++ b/drivers/acpi/acpi_extlog.c > @@ -208,10 +208,9 @@ static int extlog_print(struct notifier_block *nb, > unsigned long val, > if (cper_estatus_len(tmp) > ELOG_ENTRY_LEN || cper_estatus_check(tmp)) > return NOTIFY_DONE; > > - if (!ras_userspace_consumers()) { > + /* Print fatal records regardless of RAS consumer registration. */ > + if (tmp->error_severity == CPER_SEV_FATAL) > print_extlog_rcd(NULL, tmp, cpu); > - goto out; > - } > > /* log event via trace */ > err_seq++; > @@ -255,7 +254,6 @@ static int extlog_print(struct notifier_block *nb, > unsigned long val, > } > } > > -out: > mce->kflags |= MCE_HANDLED_EXTLOG; > return NOTIFY_OK; > }
