Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-Id: <348c2d3f19ffcff8abe50d52513f989c4581d000.1603375524.git.christophe.leroy@csgroup.eu>
From:   Christophe Leroy <christophe.leroy@csgroup.eu>
Subject: [PATCH] powerpc/bitops: Fix possible undefined behaviour with fls()
 and fls64()
To:     Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Michael Ellerman <mpe@ellerman.id.au>, jakub@redhat.com,
        segher@kernel.crashing.org
Cc:     linux-kernel@vger.kernel.org, linuxppc-dev@lists.ozlabs.org
Date:   Thu, 22 Oct 2020 14:05:46 +0000 (UTC)
Precedence: bulk

fls() and fls64() are using __builtin_ctz() and _builtin_ctzll().
On powerpc, those builtins trivially use ctlzw and ctlzd power
instructions.

Allthough those instructions provide the expected result with
input argument 0, __builtin_ctz() and __builtin_ctzll() are
documented as undefined for value 0.

The easiest fix would be to use fls() and fls64() functions
defined in include/asm-generic/bitops/builtin-fls.h and
include/asm-generic/bitops/fls64.h, but GCC output is not optimal:

00000388 <testfls>:
 388:   2c 03 00 00     cmpwi   r3,0
 38c:   41 82 00 10     beq     39c <testfls+0x14>
 390:   7c 63 00 34     cntlzw  r3,r3
 394:   20 63 00 20     subfic  r3,r3,32
 398:   4e 80 00 20     blr
 39c:   38 60 00 00     li      r3,0
 3a0:   4e 80 00 20     blr

000003b0 <testfls64>:
 3b0:   2c 03 00 00     cmpwi   r3,0
 3b4:   40 82 00 1c     bne     3d0 <testfls64+0x20>
 3b8:   2f 84 00 00     cmpwi   cr7,r4,0
 3bc:   38 60 00 00     li      r3,0
 3c0:   4d 9e 00 20     beqlr   cr7
 3c4:   7c 83 00 34     cntlzw  r3,r4
 3c8:   20 63 00 20     subfic  r3,r3,32
 3cc:   4e 80 00 20     blr
 3d0:   7c 63 00 34     cntlzw  r3,r3
 3d4:   20 63 00 40     subfic  r3,r3,64
 3d8:   4e 80 00 20     blr

When the input of fls(x) is a constant, just check x for nullity and
return either 0 or __builtin_clz(x). Otherwise, use cntlzw instruction
directly.

For fls64() on PPC64, do the same but with __builtin_clzll() and
cntlzd instruction. On PPC32, lets take the generic fls64() which
will use our fls(). The result is as expected:

00000388 <testfls>:
 388:   7c 63 00 34     cntlzw  r3,r3
 38c:   20 63 00 20     subfic  r3,r3,32
 390:   4e 80 00 20     blr

000003a0 <testfls64>:
 3a0:   2c 03 00 00     cmpwi   r3,0
 3a4:   40 82 00 10     bne     3b4 <testfls64+0x14>
 3a8:   7c 83 00 34     cntlzw  r3,r4
 3ac:   20 63 00 20     subfic  r3,r3,32
 3b0:   4e 80 00 20     blr
 3b4:   7c 63 00 34     cntlzw  r3,r3
 3b8:   20 63 00 40     subfic  r3,r3,64
 3bc:   4e 80 00 20     blr

Fixes: 2fcff790dcb4 ("powerpc: Use builtin functions for fls()/__fls()/fls64()")
Cc: stable@vger.kernel.org
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
---
 arch/powerpc/include/asm/bitops.h | 23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/include/asm/bitops.h b/arch/powerpc/include/asm/bitops.h
index 4a4d3afd5340..299ab33505a6 100644
--- a/arch/powerpc/include/asm/bitops.h
+++ b/arch/powerpc/include/asm/bitops.h
@@ -216,15 +216,34 @@ static inline void arch___clear_bit_unlock(int nr, volatile unsigned long *addr)
  */
 static inline int fls(unsigned int x)
 {
-	return 32 - __builtin_clz(x);
+	int lz;
+
+	if (__builtin_constant_p(x))
+		return x ? 32 - __builtin_clz(x) : 0;
+	asm("cntlzw %0,%1" : "=r" (lz) : "r" (x));
+	return 32 - lz;
 }
 
 #include <asm-generic/bitops/builtin-__fls.h>
 
+/*
+ * 64-bit can do this using one cntlzd (count leading zeroes doubleword)
+ * instruction; for 32-bit we use the generic version, which does two
+ * 32-bit fls calls.
+ */
+#ifdef CONFIG_PPC64
 static inline int fls64(__u64 x)
 {
-	return 64 - __builtin_clzll(x);
+	int lz;
+
+	if (__builtin_constant_p(x))
+		return x ? 64 - __builtin_clzll(x) : 0;
+	asm("cntlzd %0,%1" : "=r" (lz) : "r" (x));
+	return 64 - lz;
 }
+#else
+#include <asm-generic/bitops/fls64.h>
+#endif
 
 #ifdef CONFIG_PPC64
 unsigned int __arch_hweight8(unsigned int w);
-- 
2.25.0