Skip to contentSkip to conversation

dma-mapping: work around clang bug

PATCH Superseded Arnd Bergmann 3 participants 5 messages Mar 7, 2019 DMA mapping helpers

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 dma-mapping: work around clang bug

Clang has a rather annoying behavior of checking for integer arithmetic problems in code paths that are discarded by gcc before that perfoms the same checks.

For DMA_BIT_MASK(64), this leads to a warning despite the result of the macro being completely sensible:

arch/arm/plat-iop/adma.c:146:24: error: shift count >= width of type [-Werror,-Wshift-count-overflow]

                .coherent_dma_mask = DMA_BIT_MASK(64),

The best workaround I could come up with is to shift the value twice, which makes the macro way less readable but always has the same result.

---
v2: fix off-by-one error

---
 include/linux/dma-mapping.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/linux/dma-mapping.h +2 -1
index 75e60be91e5f..9e438fe6b130 100644
@@ -138,7 +138,8 @@ struct dma_map_ops {
 extern const struct dma_map_ops dma_virt_ops;
 extern const struct dma_map_ops dma_dummy_ops;
 
-#define DMA_BIT_MASK(n)	(((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
+/* double shift to work around https://bugs.llvm.org/show_bug.cgi?id=38789 */
+#define DMA_BIT_MASK(n)	(((n) == 64) ? ~0ULL : ((1ULL<<((n)-1))<<1)-1)
 
 #define DMA_MASK_NONE	0x0ULL
 
Signature · 1 line
2.20.0
reviewer
#1 lore raw

Hi Arnd,

Quoted 27 lines · Arnd Bergmann

Clang has a rather annoying behavior of checking for integer arithmetic problems in code paths that are discarded by gcc before that perfoms the same checks.

For DMA_BIT_MASK(64), this leads to a warning despite the result of the macro being completely sensible:

arch/arm/plat-iop/adma.c:146:24: error: shift count >= width of type [-Werror,-Wshift-count-overflow]

                 .coherent_dma_mask = DMA_BIT_MASK(64),

The best workaround I could come up with is to shift the value twice, which makes the macro way less readable but always has the same result.

---
v2: fix off-by-one error

---
  include/linux/dma-mapping.h | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
include/linux/dma-mapping.h +0 -0
index 75e60be91e5f..9e438fe6b130 100644

include/linux/dma-mapping.h struct dma_map_ops { · line 142

Quoted 6 lines
include/linux/dma-mapping.h +2 -1
@@ -138,7 +138,8 @@ struct dma_map_ops {
  extern const struct dma_map_ops dma_virt_ops;
  extern const struct dma_map_ops dma_dummy_ops;
 
-#define DMA_BIT_MASK(n)	(((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
+/* double shift to work around https://bugs.llvm.org/show_bug.cgi?id=38789 */
+#define DMA_BIT_MASK(n)	(((n) == 64) ? ~0ULL : ((1ULL<<((n)-1))<<1)-1)

I think that now makes DMA_BIT_MASK(0) undefined - that shouldn't matter in most cases, but it could potentially happen at runtime where callers use a non-constant argument. However, it also means we don't need to special-case 64 any more (since that's there to avoid the same thing anyway), so we could simply flip that to handle 0 instead.

FWIW I'd be very tempted to fold in the second shift as "2ULL<<((n)-1)", but that may not be to everyone's taste.

Robin.

Quoted 4 lines
include/linux/dma-mapping.h +0 -0
  #define DMA_MASK_NONE	0x0ULL
 
 
#2 lore raw
Quoted 11 lines · Robin Murphy

include/linux/dma-mapping.h struct dma_map_ops { · line 142

Quoted 4 lines · Arnd Bergmann
include/linux/dma-mapping.h +2 -1
-#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
+/* double shift to work around https://bugs.llvm.org/show_bug.cgi?id=38789 */
+#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<((n)-1))<<1)-1)

I think that now makes DMA_BIT_MASK(0) undefined - that shouldn't matter in most cases, but it could potentially happen at runtime where callers use a non-constant argument. However, it also means we don't need to special-case 64 any more (since that's there to avoid the same thing anyway), so we could simply flip that to handle 0 instead.

Yes, good idea.

Quoted 2 lines

FWIW I'd be very tempted to fold in the second shift as "2ULL<<((n)-1)", but that may not be to everyone's taste.

I like that. So shall we do this?

/*
* Shifting '2' instead of '1' because of
* https://bugs.llvm.org/show_bug.cgi?id=38789
*/
#define DMA_BIT_MASK(n) (((n) == 0) ? 0ULL : ((2ULL<<((n)-1)))-1)

Arnd

Reviewed-by
#3 lore raw
Quoted 18 lines · Arnd Bergmann
Quoted 11 lines · Robin Murphy

include/linux/dma-mapping.h struct dma_map_ops { · line 142

Quoted 4 lines · Arnd Bergmann
include/linux/dma-mapping.h +2 -1
-#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
+/* double shift to work around https://bugs.llvm.org/show_bug.cgi?id=38789 */
+#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<((n)-1))<<1)-1)

I think that now makes DMA_BIT_MASK(0) undefined - that shouldn't matter in most cases, but it could potentially happen at runtime where callers use a non-constant argument. However, it also means we don't need to special-case 64 any more (since that's there to avoid the same thing anyway), so we could simply flip that to handle 0 instead.

Yes, good idea.

Quoted 2 lines

FWIW I'd be very tempted to fold in the second shift as "2ULL<<((n)-1)", but that may not be to everyone's taste.

Quoted 7 lines

I like that. So shall we do this?

/*
* Shifting '2' instead of '1' because of
* https://bugs.llvm.org/show_bug.cgi?id=38789
*/
#define DMA_BIT_MASK(n) (((n) == 0) ? 0ULL : ((2ULL<<((n)-1)))-1)

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
reviewer Reviewed-by
#4 lore raw
In reply to Arnd Bergmann
Quoted 18 lines · Arnd Bergmann
Quoted 11 lines · Robin Murphy

include/linux/dma-mapping.h struct dma_map_ops { · line 142

Quoted 4 lines · Arnd Bergmann
include/linux/dma-mapping.h +2 -1
-#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<(n))-1))
+/* double shift to work around https://bugs.llvm.org/show_bug.cgi?id=38789 */
+#define DMA_BIT_MASK(n)      (((n) == 64) ? ~0ULL : ((1ULL<<((n)-1))<<1)-1)

I think that now makes DMA_BIT_MASK(0) undefined - that shouldn't matter in most cases, but it could potentially happen at runtime where callers use a non-constant argument. However, it also means we don't need to special-case 64 any more (since that's there to avoid the same thing anyway), so we could simply flip that to handle 0 instead.

Yes, good idea.

Quoted 2 lines

FWIW I'd be very tempted to fold in the second shift as "2ULL<<((n)-1)", but that may not be to everyone's taste.

Quoted 7 lines

I like that. So shall we do this?

/*
* Shifting '2' instead of '1' because of
* https://bugs.llvm.org/show_bug.cgi?id=38789
*/
#define DMA_BIT_MASK(n) (((n) == 0) ? 0ULL : ((2ULL<<((n)-1)))-1)

Neat - it was too early in the morning for me to think of a succinct way to comment it, but that's great. I suspect there may be a redundant set of parentheses around the shift still, but other than that,

Cheers,
Robin.