Re: [PATCH] x86/mce: Check that memory address is usable for recovery

[Yazen Ghannam <yazen.ghannam@amd.com>]

On 4/18/23 13:27, Tony Luck wrote:
> On Tue, Apr 18, 2023 at 12:41:17PM -0400, Yazen Ghannam wrote:
>> On 3/21/23 20:51, Tony Luck wrote:
>>> uc_decode_notifier() includes a check that "struct mce"
>>> contains a valid address for recovery. But the machine
>>> check recovery code does not include a similar check.
>>>
>>> Use mce_usable_address() to check that there is a valid
>>> address.
>>>
>>> Signed-off-by: Tony Luck <tony.luck@intel.com>
>>> ---
>>>  arch/x86/kernel/cpu/mce/core.c | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/arch/x86/kernel/cpu/mce/core.c b/arch/x86/kernel/cpu/mce/core.c
>>> index 2eec60f50057..fa28b3f7d945 100644
>>> --- a/arch/x86/kernel/cpu/mce/core.c
>>> +++ b/arch/x86/kernel/cpu/mce/core.c
>>> @@ -1533,7 +1533,7 @@ noinstr void do_machine_check(struct pt_regs *regs)
>>>  		/* If this triggers there is no way to recover. Die hard. */
>>>  		BUG_ON(!on_thread_stack() || !user_mode(regs));
>>>  
>>> -		if (kill_current_task)
>>> +		if (kill_current_task || !mce_usable_address(&m))
>>>  			queue_task_work(&m, msg, kill_me_now);
>>>  		else
>>>  			queue_task_work(&m, msg, kill_me_maybe);
>>
>> I think it should be like this:
>>
>> 	if (mce_usable_address(&m))
>> 		queue_task_work(&m, msg, kill_me_maybe);
>> 	else
>> 		queue_task_work(&m, msg, kill_me_now);
>>
>> A usable address should always go through memory_failure() so that the page is
>> marked as poison. If !RIPV, then memory_failure() will get the MF_MUST_KILL
>> flag and try to kill all processes after the page is poisoned.
>>
>> I had a similar patch a while back:
>> https://lore.kernel.org/linux-edac/20210504174712.27675-3-Yazen.Ghannam@amd.com/
>>
>> We could also get rid of kill_me_now() like you had suggested.
> 
> Can we also get rid of "kill_current_task"? It is only set when there is
> no valid return address:
> 
>         if (!(m.mcgstatus & MCG_STATUS_RIPV))
>                 kill_current_task = 1;
> 
> kill_me_maybe() does an equivalent check:
> 
> 	if (!p->mce_ripv)
>                 flags |= MF_MUST_KILL;
> 
> Which only leaves this check to resolve in some way:
> 
>         if (worst != MCE_AR_SEVERITY && !kill_current_task)
>                 goto out;
> 

I agree. And I think all these checks should be baked into the severity.
We'll need additional, fine-grained severity levels though.

The "m.cs" and "m.kflags" checks could also be baked in.

Instead of just one AR severity:
	...
	MCE_AR_SEVERITY,
	MCE_PANIC_SEVERITY,

replace it with specific cases:
	...
	MCE_AR_USER_RECOV,
	MCE_AR_USER_KILL,
	MCE_AR_KERNEL_COPYIN,
	MCE_AR_KERNEL_RECOV,
	MCE_PANIC_SEVERITY,

Then the #MC handler can look like this:

	if (worst < MCE_AR_USER_RECOV)
		goto out;

	if (severity == MCE_AR_USER_RECOV)
		queue_task_work(&m, msg, kill_me_maybe);

	if (severity == MCE_AR_USER_KILL)
		force_sig(SIGBUS);

	if (severity == MCE_AR_KERNEL_COPYIN)
		queue_task_work(&m, msg, kill_me_never);
	
	if (severity == MCE_AR_KERNEL_RECOV) {
		if (!fixup_exception(regs, X86_TRAP_MC, 0, 0))
			mce_panic("Failed kernel mode recovery");
	}

I can take a shot at this if it seems reasonable.

What do you think?

Thanks,
Yazen