simd: Allow simd use in kernel threads with softirqs disabled
diff mbox series

Message ID 20141114154315.GA15771@gondor.apana.org.au
State New, archived
Headers show
Series
  • simd: Allow simd use in kernel threads with softirqs disabled
Related show

Commit Message

Herbert Xu Nov. 14, 2014, 3:43 p.m. UTC
While working on the cryptd request reordering problem, I noticed
an anomaly where kernel threads are normally allowed to use simd
per may_use_simd, but as soon as you disable softirqs, they suddenly
lose that ability for no good reason.

The problem is that in_interrupt does not distinguish between
softirq processing and simply having softirqs disabled.  This
patch creates a new helper in_serving_interrupt which makes that
distinction.  It then uses this in all current implementations
of may_use_simd.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>


Thanks,

Comments

Ard Biesheuvel Nov. 14, 2014, 4:17 p.m. UTC | #1
On 14 November 2014 16:43, Herbert Xu <herbert@gondor.apana.org.au> wrote:
> While working on the cryptd request reordering problem, I noticed
> an anomaly where kernel threads are normally allowed to use simd
> per may_use_simd, but as soon as you disable softirqs, they suddenly
> lose that ability for no good reason.
>
> The problem is that in_interrupt does not distinguish between
> softirq processing and simply having softirqs disabled.  This
> patch creates a new helper in_serving_interrupt which makes that
> distinction.  It then uses this in all current implementations
> of may_use_simd.
>

Isn't that a much more widespread problem if in_interrupt() yields
true while no interrupt is being served?

> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
>
> diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
> index a9a4229..6cdaa852 100644
> --- a/arch/x86/kernel/i387.c
> +++ b/arch/x86/kernel/i387.c
> @@ -63,7 +63,7 @@ static inline bool interrupted_user_mode(void)
>   */
>  bool irq_fpu_usable(void)
>  {
> -       return !in_interrupt() ||
> +       return !in_serving_interrupt() ||
>                 interrupted_user_mode() ||
>                 interrupted_kernel_fpu_idle();
>  }
> diff --git a/include/asm-generic/simd.h b/include/asm-generic/simd.h
> index f57eb7b..74e0b05 100644
> --- a/include/asm-generic/simd.h
> +++ b/include/asm-generic/simd.h
> @@ -10,5 +10,5 @@
>   */
>  static __must_check inline bool may_use_simd(void)
>  {
> -       return !in_interrupt();
> +       return !in_serving_interrupt();
>  }
> diff --git a/include/linux/preempt_mask.h b/include/linux/preempt_mask.h
> index dbeec4d..18f3b46 100644
> --- a/include/linux/preempt_mask.h
> +++ b/include/linux/preempt_mask.h
> @@ -65,6 +65,8 @@
>  #define in_softirq()           (softirq_count())
>  #define in_interrupt()         (irq_count())
>  #define in_serving_softirq()   (softirq_count() & SOFTIRQ_OFFSET)
> +#define in_serving_interrupt() (irq_count() & (HARDIRQ_MASK | \
> +                                               SOFTIRQ_OFFSET | NMI_MASK))
>
>  /*
>   * Are we in NMI context?
>
> Thanks,
> --
> Email: Herbert Xu <herbert@gondor.apana.org.au>
> Home Page: http://gondor.apana.org.au/~herbert/
> PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Herbert Xu Nov. 15, 2014, 2:28 a.m. UTC | #2
On Fri, Nov 14, 2014 at 05:17:05PM +0100, Ard Biesheuvel wrote:
> On 14 November 2014 16:43, Herbert Xu <herbert@gondor.apana.org.au> wrote:
> > While working on the cryptd request reordering problem, I noticed
> > an anomaly where kernel threads are normally allowed to use simd
> > per may_use_simd, but as soon as you disable softirqs, they suddenly
> > lose that ability for no good reason.
> >
> > The problem is that in_interrupt does not distinguish between
> > softirq processing and simply having softirqs disabled.  This
> > patch creates a new helper in_serving_interrupt which makes that
> > distinction.  It then uses this in all current implementations
> > of may_use_simd.
> >
> 
> Isn't that a much more widespread problem if in_interrupt() yields
> true while no interrupt is being served?

Good point.  Other users of in_interrupt() are just as likely to
really want the in_serving version.  However, the safe option is
to leave them unchanged until we audit each single one.

If it turns out that all users of in_interrupt() decay into
in_serving_interrupt() then we could remove in_interrupt().

Cheers,

Patch
diff mbox series

diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index a9a4229..6cdaa852 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -63,7 +63,7 @@  static inline bool interrupted_user_mode(void)
  */
 bool irq_fpu_usable(void)
 {
-	return !in_interrupt() ||
+	return !in_serving_interrupt() ||
 		interrupted_user_mode() ||
 		interrupted_kernel_fpu_idle();
 }
diff --git a/include/asm-generic/simd.h b/include/asm-generic/simd.h
index f57eb7b..74e0b05 100644
--- a/include/asm-generic/simd.h
+++ b/include/asm-generic/simd.h
@@ -10,5 +10,5 @@ 
  */
 static __must_check inline bool may_use_simd(void)
 {
-	return !in_interrupt();
+	return !in_serving_interrupt();
 }
diff --git a/include/linux/preempt_mask.h b/include/linux/preempt_mask.h
index dbeec4d..18f3b46 100644
--- a/include/linux/preempt_mask.h
+++ b/include/linux/preempt_mask.h
@@ -65,6 +65,8 @@ 
 #define in_softirq()		(softirq_count())
 #define in_interrupt()		(irq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
+#define in_serving_interrupt()	(irq_count() & (HARDIRQ_MASK | \
+						SOFTIRQ_OFFSET | NMI_MASK))
 
 /*
  * Are we in NMI context?