Recently I went down a rabbit hole looking at a race condition in
panic() on a Hyper-V guest. I assumed, since it was missing from the
command line, that crash_kexec_post_notifiers was disabled. Only after
a rather long reproduction and analysis process did I learn that Hyper-V
actually enables this setting unconditionally.
Users and debuggers alike would like to know when these things happen. I
think it would be good to print a message to the kernel log when this
happens, so that a grep for "crash_kexec_post_notifiers" shows relevant
results.
Signed-off-by: Stephen Brennan <[email protected]>
---
drivers/hv/hv_common.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c
index 181d16bbf49d..c1dd21d0d7ef 100644
--- a/drivers/hv/hv_common.c
+++ b/drivers/hv/hv_common.c
@@ -79,8 +79,10 @@ int __init hv_common_init(void)
* calling crash enlightment interface before running kdump
* kernel.
*/
- if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE)
+ if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
crash_kexec_post_notifiers = true;
+ pr_info("Hyper-V: enabling crash_kexec_post_notifiers\n");
+ }
/*
* Allocate the per-CPU state for the hypercall input arg.
--
2.30.2
From: Stephen Brennan <[email protected]> Sent: Monday, February 14, 2022 5:38 PM
>
> Recently I went down a rabbit hole looking at a race condition in
> panic() on a Hyper-V guest. I assumed, since it was missing from the
> command line, that crash_kexec_post_notifiers was disabled. Only after
> a rather long reproduction and analysis process did I learn that Hyper-V
> actually enables this setting unconditionally.
>
> Users and debuggers alike would like to know when these things happen. I
> think it would be good to print a message to the kernel log when this
> happens, so that a grep for "crash_kexec_post_notifiers" shows relevant
> results.
I'm OK with adding this output line. However, you have probably
seen the two other LKML threads [1] and [2] about reorganizing the
panic notifiers to clearly distinguish between notifiers that always run
vs. those controlled by "crash_kexec_post_notifiers". If the changes
proposed in those threads are submitted and accepted, it is likely that
the kernel log message in this patch would become unnecessary.
But since we don't know when those proposed changes might come
to fruition, adding the message for now makes sense.
Reviewed-by: Michael Kelley <[email protected]>
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://lore.kernel.org/lkml/[email protected]/
>
> Signed-off-by: Stephen Brennan <[email protected]>
> ---
> drivers/hv/hv_common.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c
> index 181d16bbf49d..c1dd21d0d7ef 100644
> --- a/drivers/hv/hv_common.c
> +++ b/drivers/hv/hv_common.c
> @@ -79,8 +79,10 @@ int __init hv_common_init(void)
> * calling crash enlightment interface before running kdump
> * kernel.
> */
> - if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE)
> + if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
> crash_kexec_post_notifiers = true;
> + pr_info("Hyper-V: enabling crash_kexec_post_notifiers\n");
> + }
>
> /*
> * Allocate the per-CPU state for the hypercall input arg.
> --
> 2.30.2
On 2/17/22 08:44, Michael Kelley (LINUX) wrote:
> From: Stephen Brennan <[email protected]> Sent: Monday, February 14, 2022 5:38 PM
>>
>> Recently I went down a rabbit hole looking at a race condition in
>> panic() on a Hyper-V guest. I assumed, since it was missing from the
>> command line, that crash_kexec_post_notifiers was disabled. Only after
>> a rather long reproduction and analysis process did I learn that Hyper-V
>> actually enables this setting unconditionally.
>>
>> Users and debuggers alike would like to know when these things happen. I
>> think it would be good to print a message to the kernel log when this
>> happens, so that a grep for "crash_kexec_post_notifiers" shows relevant
>> results.
>
> I'm OK with adding this output line. However, you have probably
> seen the two other LKML threads [1] and [2] about reorganizing the
> panic notifiers to clearly distinguish between notifiers that always run
> vs. those controlled by "crash_kexec_post_notifiers".
Yeah, I fired this off before seeing those patches. I need to get on top
of the "lore+lei" combination so I can see these discussions quickly, as
there's no subsystem-specific mailing list for the panic/printk stuff.
The patches only surface if you're mentioned or if you trawl through
LKML itself :)
> If the changes
> proposed in those threads are submitted and accepted, it is likely that
> the kernel log message in this patch would become unnecessary.
> But since we don't know when those proposed changes might come
> to fruition, adding the message for now makes sense.
>
> Reviewed-by: Michael Kelley <[email protected]>
Thank you!
Stephen
>
> [1] https://lore.kernel.org/lkml/[email protected]/
> [2] https://lore.kernel.org/lkml/[email protected]/
>
>>
>> Signed-off-by: Stephen Brennan <[email protected]>
>> ---
>> drivers/hv/hv_common.c | 4 +++-
>> 1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c
>> index 181d16bbf49d..c1dd21d0d7ef 100644
>> --- a/drivers/hv/hv_common.c
>> +++ b/drivers/hv/hv_common.c
>> @@ -79,8 +79,10 @@ int __init hv_common_init(void)
>> * calling crash enlightment interface before running kdump
>> * kernel.
>> */
>> - if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE)
>> + if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
>> crash_kexec_post_notifiers = true;
>> + pr_info("Hyper-V: enabling crash_kexec_post_notifiers\n");
>> + }
>>
>> /*
>> * Allocate the per-CPU state for the hypercall input arg.
>> --
>> 2.30.2
>
On Thu, Feb 17, 2022 at 04:44:13PM +0000, Michael Kelley (LINUX) wrote:
> From: Stephen Brennan <[email protected]> Sent: Monday, February 14, 2022 5:38 PM
> >
> > Recently I went down a rabbit hole looking at a race condition in
> > panic() on a Hyper-V guest. I assumed, since it was missing from the
> > command line, that crash_kexec_post_notifiers was disabled. Only after
> > a rather long reproduction and analysis process did I learn that Hyper-V
> > actually enables this setting unconditionally.
> >
> > Users and debuggers alike would like to know when these things happen. I
> > think it would be good to print a message to the kernel log when this
> > happens, so that a grep for "crash_kexec_post_notifiers" shows relevant
> > results.
>
> I'm OK with adding this output line. However, you have probably
> seen the two other LKML threads [1] and [2] about reorganizing the
> panic notifiers to clearly distinguish between notifiers that always run
> vs. those controlled by "crash_kexec_post_notifiers". If the changes
> proposed in those threads are submitted and accepted, it is likely that
> the kernel log message in this patch would become unnecessary.
> But since we don't know when those proposed changes might come
> to fruition, adding the message for now makes sense.
>
> Reviewed-by: Michael Kelley <[email protected]>
Applied to hyperv-next. Thanks.