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

[Peter Zijlstra]

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?

>  	/* 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.

---
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;