Skip to contentSkip to conversation

fix BUILD_BUG_ON() and a couple of bogus uses of it

PATCH In review Jan Beulich 4 participants 9 messages Aug 18, 2009 – Sep 23, 2009

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fix BUILD_BUG_ON() and a couple of bogus uses of it

gcc permitting variable length arrays makes the current construct used for BUILD_BUG_ON() useless, as that doesn't produce any diagnostic if the controlling expression isn't really constant. Instead, this patch makes it so that a bit field gets used here. Consequently, those uses where the condition isn't really constant now also need fixing.

Note that in the gfp.h, kmemcheck.h, and virtio_config.h cases MAYBE_BUILD_BUG_ON() really just serves documentation purposes - even if the expression is compile time constant (__builtin_constant_p() yields true), the array is still deemed of variable length by gcc, and hence the whole expression doesn't have the intended effect.

---
 drivers/char/tpm/tpm.c        |    2 --
 drivers/net/niu.c             |    2 +-
 include/linux/gfp.h           |    2 +-
 include/linux/kernel.h        |    8 ++++++--
 include/linux/kmemcheck.h     |    2 +-
 include/linux/virtio_config.h |    3 +--
 6 files changed, 10 insertions(+), 9 deletions(-)
drivers/char/tpm/tpm.c +0 -2
@@ -696,7 +696,6 @@ int __tpm_pcr_read(struct tpm_chip *chip
 
 	cmd.header.in = pcrread_header;
 	cmd.params.pcrread_in.pcr_idx = cpu_to_be32(pcr_idx);
-	BUILD_BUG_ON(cmd.header.in.length > READ_PCR_RESULT_SIZE);
 	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
 			  "attempting to read a pcr value");
 
@@ -760,7 +759,6 @@ int tpm_pcr_extend(u32 chip_num, int pcr
 		return -ENODEV;
 
 	cmd.header.in = pcrextend_header;
-	BUILD_BUG_ON(be32_to_cpu(cmd.header.in.length) > EXTEND_PCR_SIZE);
 	cmd.params.pcrextend_in.pcr_idx = cpu_to_be32(pcr_idx);
 	memcpy(cmd.params.pcrextend_in.hash, hash, TPM_DIGEST_SIZE);
 	rc = transmit_cmd(chip, &cmd, cmd.header.in.length,
drivers/net/niu.c +1 -1
@@ -5615,7 +5615,7 @@ static void niu_init_tx_mac(struct niu *
 	/* The XMAC_MIN register only accepts values for TX min which
 	 * have the low 3 bits cleared.
 	 */
-	BUILD_BUG_ON(min & 0x7);
+	BUG_ON(min & 0x7);
 
 	if (np->flags & NIU_FLAGS_XMAC)
 		niu_init_tx_xmac(np, min, max);
include/linux/gfp.h +1 -1
@@ -220,7 +220,7 @@ static inline enum zone_type gfp_zone(gf
 					 ((1 << ZONES_SHIFT) - 1);
 
 	if (__builtin_constant_p(bit))
-		BUILD_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
+		MAYBE_BUILD_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
 	else {
 #ifdef CONFIG_DEBUG_VM
 		BUG_ON((GFP_ZONE_BAD >> bit) & 1);
include/linux/kernel.h +6 -2
@@ -675,13 +675,17 @@ struct sysinfo {
 };
 
 /* Force a compilation error if condition is true */
-#define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)]))
+#define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
+
+/* Force a compilation error if condition is constant and true */
+#define MAYBE_BUILD_BUG_ON(cond) ((void)sizeof(char[1 - 2 * !!(cond)]))
 
 /* Force a compilation error if condition is true, but also produce a
    result (of value 0 and type size_t), so the expression can be used
    e.g. in a structure initializer (or where-ever else comma expressions
    aren't permitted). */
-#define BUILD_BUG_ON_ZERO(e) (sizeof(char[1 - 2 * !!(e)]) - 1)
+#define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
+#define BUILD_BUG_ON_NULL(e) ((void *)sizeof(struct { int:-!!(e); }))
 
 /* Trap pasters of __FUNCTION__ at compile-time */
 #define __FUNCTION__ (__func__)
include/linux/kmemcheck.h +1 -1
@@ -140,7 +140,7 @@ static inline void kmemcheck_mark_initia
 	do if (ptr) {							\
 		int _n = (long) &((ptr)->name##_end)			\
 			- (long) &((ptr)->name##_begin);		\
-		BUILD_BUG_ON(_n < 0);					\
+		MAYBE_BUILD_BUG_ON(_n < 0);				\
 									\
 		kmemcheck_mark_initialized(&((ptr)->name##_begin), _n);	\
 	} while (0)
include/linux/virtio_config.h +1 -2
@@ -109,8 +109,7 @@ static inline bool virtio_has_feature(co
 				      unsigned int fbit)
 {
 	/* Did you forget to fix assumptions on max features? */
-	if (__builtin_constant_p(fbit))
-		BUILD_BUG_ON(fbit >= 32);
+	MAYBE_BUILD_BUG_ON(fbit >= 32);
 
 	if (fbit < VIRTIO_TRANSPORT_F_START)
 		virtio_check_driver_offered_feature(vdev, fbit);
maintainer
#1 lore raw
Quoted 11 lines · On Tue, 18 Aug 2009 16:59:25 +0100 "Jan Beulich" <[email protected]> wrote:

gcc permitting variable length arrays makes the current construct used for BUILD_BUG_ON() useless, as that doesn't produce any diagnostic if the controlling expression isn't really constant. Instead, this patch makes it so that a bit field gets used here. Consequently, those uses where the condition isn't really constant now also need fixing.

Note that in the gfp.h, kmemcheck.h, and virtio_config.h cases MAYBE_BUILD_BUG_ON() really just serves documentation purposes - even if the expression is compile time constant (__builtin_constant_p() yields true), the array is still deemed of variable length by gcc, and hence the whole expression doesn't have the intended effect.

sparc64:

In file included from arch/sparc/kernel/vio.c:17:
/usr/src/devel/arch/sparc/include/asm/vio.h: In function `vio_dring_avail':
/usr/src/devel/arch/sparc/include/asm/vio.h:261: error: bit-field `<anonymous>' width not an integer constant

static inline u32 vio_dring_avail(struct vio_dring_state *dr,
unsigned int ring_size)

{
	BUILD_BUG_ON(!is_power_of_2(ring_size));

	return (dr->pending -
		((dr->prod - dr->cons) & (ring_size - 1)));
}

changing it to MAYBE_BUILD_BUG_ON seems to have fixed it.

maintainer
#2 lore raw

From: Andrew Morton <[email protected]>
Date: Thu, 17 Sep 2009 16:24:37 -0700

Quoted 8 lines

sparc64:

In file included from arch/sparc/kernel/vio.c:17:
/usr/src/devel/arch/sparc/include/asm/vio.h: In function `vio_dring_avail':
/usr/src/devel/arch/sparc/include/asm/vio.h:261: error: bit-field `<anonymous>' width not an integer constant

static inline u32 vio_dring_avail(struct vio_dring_state *dr,
unsigned int ring_size)

Quoted 8 lines
{
	BUILD_BUG_ON(!is_power_of_2(ring_size));

	return (dr->pending -
		((dr->prod - dr->cons) & (ring_size - 1)));
}

changing it to MAYBE_BUILD_BUG_ON seems to have fixed it.

That's completely bogus.

First of all, arch/sparc/kernel/vio.c never calls this function so it should never be evaluated.

Second of all, all the places that do call this function only pass pure constants as the 'ring_size' parameter.

drivers/block/sunvdc.c:

static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)

{
	return vio_dring_avail(dr, VDC_TX_RING_SIZE);
}

drivers/net/sunvnet.c:

static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)

{
	return vio_dring_avail(dr, VNET_TX_RING_SIZE);
}

Making this MAYBE_BUILD_BUG_ON shouldn't be necessary.

maintainer
#3 lore raw
Quoted 35 lines · On Thu, 17 Sep 2009 17:01:56 -0700 (PDT) David Miller <[email protected]>…

From: Andrew Morton <[email protected]>
Date: Thu, 17 Sep 2009 16:24:37 -0700

Quoted 8 lines

sparc64:

In file included from arch/sparc/kernel/vio.c:17:
/usr/src/devel/arch/sparc/include/asm/vio.h: In function `vio_dring_avail':
/usr/src/devel/arch/sparc/include/asm/vio.h:261: error: bit-field `<anonymous>' width not an integer constant

static inline u32 vio_dring_avail(struct vio_dring_state *dr,
unsigned int ring_size)

Quoted 8 lines
{
	BUILD_BUG_ON(!is_power_of_2(ring_size));

	return (dr->pending -
		((dr->prod - dr->cons) & (ring_size - 1)));
}

changing it to MAYBE_BUILD_BUG_ON seems to have fixed it.

That's completely bogus.

First of all, arch/sparc/kernel/vio.c never calls this function so it should never be evaluated.

Second of all, all the places that do call this function only pass pure constants as the 'ring_size' parameter.

drivers/block/sunvdc.c:

static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)

{
	return vio_dring_avail(dr, VDC_TX_RING_SIZE);
}
Quoted 8 lines

drivers/net/sunvnet.c:

static inline u32 vnet_tx_dring_avail(struct vio_dring_state *dr)

{
	return vio_dring_avail(dr, VNET_TX_RING_SIZE);
}

Making this MAYBE_BUILD_BUG_ON shouldn't be necessary.

There's a shortcoming in the current BUILD_BUG_ON() - it silently does nothing if passed a non-constant arg.

I suspect that in the 2.6.31 code, that BUILD_BUG_ON() just does nothing at all, and that Jan's patch is now exposing this. It might be compiler-version dependent too.

<tests it>

Yup, on base 2.6.31, this:

arch/sparc/kernel/vio.c renamed from arch/sparc/kernel/vio.c~a +2 -0
@@ -23,6 +23,8 @@ static const struct vio_device_id *vio_m
 	const char *type, *compat;
 	int len;
 
+	vio_dring_avail(NULL, 33);
+
 	type = dev->type;
 	compat = dev->compat;
 	len = dev->compat_len;

_

compiles without error with gcc-3.4.5.

maintainer
#4 lore raw
Quoted 25 lines · On Thu, 17 Sep 2009 17:15:04 -0700 Andrew Morton <[email protected]>…

There's a shortcoming in the current BUILD_BUG_ON() - it silently does nothing if passed a non-constant arg.

I suspect that in the 2.6.31 code, that BUILD_BUG_ON() just does nothing at all, and that Jan's patch is now exposing this. It might be compiler-version dependent too.

<tests it>

Yup, on base 2.6.31, this:

arch/sparc/kernel/vio.c renamed from arch/sparc/kernel/vio.c~a +2 -0
@@ -23,6 +23,8 @@ static const struct vio_device_id *vio_m
 	const char *type, *compat;
 	int len;
 
+	vio_dring_avail(NULL, 33);
+
 	type = dev->type;
 	compat = dev->compat;
 	len = dev->compat_len;

_

Quoted 1 line

compiles without error with gcc-3.4.5.

And I can't immediately find a way to make any compile-time error occur here, with or without Jan's patch. hm.

maintainer
#5 lore raw
In reply to Andrew Morton

From: Andrew Morton <[email protected]>
Date: Thu, 17 Sep 2009 17:15:04 -0700

Quoted 11 lines

There's a shortcoming in the current BUILD_BUG_ON() - it silently does nothing if passed a non-constant arg.

I suspect that in the 2.6.31 code, that BUILD_BUG_ON() just does nothing at all, and that Jan's patch is now exposing this. It might be compiler-version dependent too.

<tests it>

Yup, on base 2.6.31, this:

Ok, I'll have to either change this function to a macro or get rid of the check.

Thanks.

maintainer
#6 lore raw

From: David Miller <[email protected]>
Date: Thu, 17 Sep 2009 17:31:54 -0700 (PDT)

Quoted 11 lines

From: Andrew Morton <[email protected]>
Date: Thu, 17 Sep 2009 17:15:04 -0700

Quoted 8 lines

There's a shortcoming in the current BUILD_BUG_ON() - it silently does nothing if passed a non-constant arg.

I suspect that in the 2.6.31 code, that BUILD_BUG_ON() just does nothing at all, and that Jan's patch is now exposing this. It might be compiler-version dependent too.

Quoted 6 lines
Quoted 3 lines

<tests it>

Yup, on base 2.6.31, this:

Ok, I'll have to either change this function to a macro or get rid of the check.

I can't even get GCC to see the constant evaluated by is_power_of_2().

I give up, I'll just remove the BUILD_BUG_ON() entirely.

#7 lore raw
In reply to Jan Beulich
Quoted 5 lines · Jan Beulich

Note that in the gfp.h, kmemcheck.h, and virtio_config.h cases MAYBE_BUILD_BUG_ON() really just serves documentation purposes - even if the expression is compile time constant (__builtin_constant_p() yields true), the array is still deemed of variable length by gcc, and hence the whole expression doesn't have the intended effect.

How annoying. A gcc build assert extension would be great here?

Thanks,
Rusty.

#8 lore raw
In reply to Jan Beulich
Quoted 6 lines · Jan Beulich

gcc permitting variable length arrays makes the current construct used for BUILD_BUG_ON() useless, as that doesn't produce any diagnostic if the controlling expression isn't really constant. Instead, this patch makes it so that a bit field gets used here. Consequently, those uses where the condition isn't really constant now also need fixing.

Quoted 7 lines

Note that in the gfp.h, kmemcheck.h, and virtio_config.h cases MAYBE_BUILD_BUG_ON() really just serves documentation purposes - even if the expression is compile time constant (__builtin_constant_p() yields true), the array is still deemed of variable length by gcc, and hence the whole expression doesn't have the intended effect.

We used to use an undefined symbol here; diagnostics are worse but it catches more stuff.

Perhaps a hybrid is the way to go?

#ifndef __OPTIMIZE__
#define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)])) #else
/* If it's a constant, catch it at compile time, otherwise at link time. */ extern int __build_bug_on_failed;

#define BUILD_BUG_ON(condition) \
	do { 								\
		((void)sizeof(char[1 - 2*!!(condition)]));		\
		if (condition) __build_bug_on_failed = 1;		\
	} while(0)

#endif

Thanks,
Rusty.