Re: [PATCH v2 3/5] x86/mce: add EX_TYPE_EFAULT_REG as in-kernel recovery context to fix copy-from-user operations regression

[Shuai Xue <xueshuai@xxxxxxxxxxxxxxxxx>]

Hi, Peter

在 2025/2/18 20:54, Peter Zijlstra 写道:
> On Mon, Feb 17, 2025 at 02:33:33PM +0800, Shuai Xue wrote:
>
> > diff --git a/arch/x86/kernel/cpu/mce/severity.c b/arch/x86/kernel/cpu/mce/severity.c
> > index dac4d64dfb2a..14c2d71c3ce1 100644
> > --- a/arch/x86/kernel/cpu/mce/severity.c
> > +++ b/arch/x86/kernel/cpu/mce/severity.c
> > @@ -16,6 +16,7 @@
> >   #include <asm/traps.h>
> >   #include <asm/insn.h>
> >   #include <asm/insn-eval.h>
> > +#include <linux/extable.h>
> >   
> >   #include "internal.h"
> >   
> > @@ -285,7 +286,8 @@ static bool is_copy_from_user(struct pt_regs *regs)
> >    */
> >   static noinstr int error_context(struct mce *m, struct pt_regs *regs)
> >   {
> > -	int fixup_type;
> > +	const struct exception_table_entry *e;
> > +	int fixup_type, imm;
> >   	bool copy_user;
> >   
> >   	if ((m->cs & 3) == 3)
> > @@ -294,9 +296,14 @@ static noinstr int error_context(struct mce *m, struct pt_regs *regs)
> >   	if (!mc_recoverable(m->mcgstatus))
> >   		return IN_KERNEL;
> >   
> > +	e = search_exception_tables(m->ip);
> > +	if (!e)
> > +		return IN_KERNEL;
>
> You didn't actually build this, did you? Or did you ignore the extra
> noinstr warnings?

I did build and test this patch set on it. But I did not find any warnings.
Could you provide more details?

> >   	/* Allow instrumentation around external facilities usage. */
> >   	instrumentation_begin();
> > -	fixup_type = ex_get_fixup_type(m->ip);
> > +	fixup_type = FIELD_GET(EX_DATA_TYPE_MASK, e->data);
> > +	imm  = FIELD_GET(EX_DATA_IMM_MASK,  e->data);
> >   	copy_user  = is_copy_from_user(regs);
> >   	instrumentation_end();
> >   
> > @@ -304,9 +311,13 @@ static noinstr int error_context(struct mce *m, struct pt_regs *regs)
> >   	case EX_TYPE_UACCESS:
> >   		if (!copy_user)
> >   			return IN_KERNEL;
> > -		m->kflags |= MCE_IN_KERNEL_COPYIN;
> > -		fallthrough;
> > -
> > +		m->kflags |= MCE_IN_KERNEL_COPYIN | MCE_IN_KERNEL_RECOV;
> > +		return IN_KERNEL_RECOV;
> > +	case EX_TYPE_IMM_REG:
> > +		if (!copy_user || imm != -EFAULT)
> > +			return IN_KERNEL;
> > +		m->kflags |= MCE_IN_KERNEL_COPYIN | MCE_IN_KERNEL_RECOV;
> > +		return IN_KERNEL_RECOV;
>
> Maybe I'm justnot understanding things, but what's wrong with something
> like the below; why do we care about the ex-type if we know its a MOV
> reading from userspace?
>
> The less we muck about with the extable here, the better.

We need to make sure that we have register a fixup handler for the copy_user
case.  If no fixup handler found, the PC accessing posion will trigger #MCE
again and again resulting a hardlock up.

> ---
> diff --git a/arch/x86/kernel/cpu/mce/severity.c b/arch/x86/kernel/cpu/mce/severity.c
> index dac4d64dfb2a..cb021058165f 100644
> --- a/arch/x86/kernel/cpu/mce/severity.c
> +++ b/arch/x86/kernel/cpu/mce/severity.c
> @@ -300,13 +300,12 @@ static noinstr int error_context(struct mce *m, struct pt_regs *regs)
>   	copy_user  = is_copy_from_user(regs);
>   	instrumentation_end();
>   
> -	switch (fixup_type) {
> -	case EX_TYPE_UACCESS:
> -		if (!copy_user)
> -			return IN_KERNEL;
> -		m->kflags |= MCE_IN_KERNEL_COPYIN;
> -		fallthrough;
> +	if (copy_user) {
> +		m->kflags |= MCE_IN_KERNEL_COPYIN | MCE_IN_KERNEL_COPYIN;
> +		return IN_KERNEL_RECOV
> +	}
>   
> +	switch (fixup_type) {
>   	case EX_TYPE_FAULT_MCE_SAFE:
>   	case EX_TYPE_DEFAULT_MCE_SAFE:
>   		m->kflags |= MCE_IN_KERNEL_RECOV;

Thanks.
Shuai