diff mbox

[PATCHv2,0/6] efi: detect erroneous firmware IRQ manipulation

Message ID 20160425105153.GR2829@codeblueprint.co.uk (mailing list archive)
State New, archived
Headers show

Commit Message

Matt Fleming April 25, 2016, 10:51 a.m. UTC
On Mon, 25 Apr, at 11:40:09AM, Mark Rutland wrote:
> 
> It looks like irqs_disabled_flags() will do what you expect, and ignore
> everything but the interrupt flag.
> 
> However, for ARM that will ignore the other exceptions we've seen FW
> erroneously unmask (e.g. FIQ), which is unfortunate, as fiddling with
> those is just as disastrous.
 
Bah, right.

> Would you be happy with an arch_efi_call_check_flags(before, after),
> instead? That way we can make the flags we check arch-specific.

Could we just make the flag mask arch-specific instead of the call
since the rest of efi_call_virt_check_flags() is good?

Something like this (uncompiled, untested, half-baked),

---

Comments

Mark Rutland April 25, 2016, 11:04 a.m. UTC | #1
On Mon, Apr 25, 2016 at 11:51:53AM +0100, Matt Fleming wrote:
> On Mon, 25 Apr, at 11:40:09AM, Mark Rutland wrote:
> > 
> > It looks like irqs_disabled_flags() will do what you expect, and ignore
> > everything but the interrupt flag.
> > 
> > However, for ARM that will ignore the other exceptions we've seen FW
> > erroneously unmask (e.g. FIQ), which is unfortunate, as fiddling with
> > those is just as disastrous.
>  
> Bah, right.
> 
> > Would you be happy with an arch_efi_call_check_flags(before, after),
> > instead? That way we can make the flags we check arch-specific.
> 
> Could we just make the flag mask arch-specific instead of the call
> since the rest of efi_call_virt_check_flags() is good?

Yup, I meant that arch_efi_call_check_flags would only do the flag
comparison, only replacing the bit currently in the WARN_ON_ONCE().

> Something like this (uncompiled, untested, half-baked),
> 
> ---
> 
> diff --git a/drivers/firmware/efi/runtime-wrappers.c b/drivers/firmware/efi/runtime-wrappers.c
> index c38b1cfc26e2..057d00bee7d6 100644
> --- a/drivers/firmware/efi/runtime-wrappers.c
> +++ b/drivers/firmware/efi/runtime-wrappers.c
> @@ -25,9 +25,12 @@
>  static void efi_call_virt_check_flags(unsigned long flags, const char *call)
>  {
>  	unsigned long cur_flags;
> +	bool mismatch;
>  
>  	local_save_flags(cur_flags);
> -	if (!WARN_ON_ONCE(cur_flags != flags))
> +
> +	mismatch = (cur_flags ^ flags) & ARCH_EFI_IRQ_FLAGS_MASK;
> +	if (!WARN_ON_ONCE(mismatch))
>  		return;

This style also works for me.

Should I respin patch 6 as a series doing the above?

I assume that the first 5 patches are fine as-is.

Thanks,
Mark.
Matt Fleming April 25, 2016, 11:19 a.m. UTC | #2
On Mon, 25 Apr, at 12:04:55PM, Mark Rutland wrote:
> On Mon, Apr 25, 2016 at 11:51:53AM +0100, Matt Fleming wrote:
> > On Mon, 25 Apr, at 11:40:09AM, Mark Rutland wrote:
> > > 
> > > It looks like irqs_disabled_flags() will do what you expect, and ignore
> > > everything but the interrupt flag.
> > > 
> > > However, for ARM that will ignore the other exceptions we've seen FW
> > > erroneously unmask (e.g. FIQ), which is unfortunate, as fiddling with
> > > those is just as disastrous.
> >  
> > Bah, right.
> > 
> > > Would you be happy with an arch_efi_call_check_flags(before, after),
> > > instead? That way we can make the flags we check arch-specific.
> > 
> > Could we just make the flag mask arch-specific instead of the call
> > since the rest of efi_call_virt_check_flags() is good?
> 
> Yup, I meant that arch_efi_call_check_flags would only do the flag
> comparison, only replacing the bit currently in the WARN_ON_ONCE().
> 
> > Something like this (uncompiled, untested, half-baked),
> > 
> > ---
> > 
> > diff --git a/drivers/firmware/efi/runtime-wrappers.c b/drivers/firmware/efi/runtime-wrappers.c
> > index c38b1cfc26e2..057d00bee7d6 100644
> > --- a/drivers/firmware/efi/runtime-wrappers.c
> > +++ b/drivers/firmware/efi/runtime-wrappers.c
> > @@ -25,9 +25,12 @@
> >  static void efi_call_virt_check_flags(unsigned long flags, const char *call)
> >  {
> >  	unsigned long cur_flags;
> > +	bool mismatch;
> >  
> >  	local_save_flags(cur_flags);
> > -	if (!WARN_ON_ONCE(cur_flags != flags))
> > +
> > +	mismatch = (cur_flags ^ flags) & ARCH_EFI_IRQ_FLAGS_MASK;
> > +	if (!WARN_ON_ONCE(mismatch))
> >  		return;
> 
> This style also works for me.
 
Cool. One thing that occurred to me after I sent it is that
technically we should either,

  1) make 'mismatch' an int or
  2) do mismatch = !!((cur_flags ^ flags) & ARCH_EFI_IRQ_FLAGS_MASK)

Either is fine with me, I just don't want to leave the implicit
conversion to C's type system.

> Should I respin patch 6 as a series doing the above?
> 
> I assume that the first 5 patches are fine as-is.

Yep, they're fine. Sure, go ahead and respin patch 6.
diff mbox

Patch

diff --git a/drivers/firmware/efi/runtime-wrappers.c b/drivers/firmware/efi/runtime-wrappers.c
index c38b1cfc26e2..057d00bee7d6 100644
--- a/drivers/firmware/efi/runtime-wrappers.c
+++ b/drivers/firmware/efi/runtime-wrappers.c
@@ -25,9 +25,12 @@ 
 static void efi_call_virt_check_flags(unsigned long flags, const char *call)
 {
 	unsigned long cur_flags;
+	bool mismatch;
 
 	local_save_flags(cur_flags);
-	if (!WARN_ON_ONCE(cur_flags != flags))
+
+	mismatch = (cur_flags ^ flags) & ARCH_EFI_IRQ_FLAGS_MASK;
+	if (!WARN_ON_ONCE(mismatch))
 		return;
 
 	add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_NOW_UNRELIABLE);