Re: 2.6.37-rc7: Regression: b43: crashes in hwrng_register()

On Thu, Jan 06, 2011 at 03:42:55PM +0100, Mario 'BitKoenig' Holbe wrote:

holbe@ideapad ~ % gcc -o via-rng-test via-rng-test.c
holbe@ideapad ~ % ./via-rng-test
0xbff09d40: 00000000
ecx: aa55aa55   edx: ffffff03   edi: 0xbff09d40
ecx: aa55aa55   edx: 00000003   edi: 0xbff09d48
0xbff09d40: 339d4525
holbe@ideapad ~ % 

Thanks, I've applied this patch:

commit 0735ac1f2551d9f9d356126aaf3b1110150918e6
Author: Herbert Xu [off-list ref]
Date:   Fri Jan 7 14:48:57 2011 +1100

    hwrng: via_rng - Fix asm constraints
    
    The inline asm to invoke xstore did not specify the constraints
    correctly.  In particular, dx/di should have been marked as output
    registers as well as input as they're modified by xstore.
    
    Thanks to Mario Holbe for creating this patch and testing it.
    
    Tested-by: Mario 'BitKoenig' Holbe [off-list ref]
    Signed-off-by: Herbert Xu [off-list ref]

diff --git a/drivers/char/hw_random/via-rng.c b/drivers/char/hw_random/via-rng.c
index 794aacb..7f86666 100644
--- a/drivers/char/hw_random/via-rng.c
+++ b/drivers/char/hw_random/via-rng.c

@@ -81,8 +81,7 @@ static inline u32 xstore(u32 *addr, u32 edx_in)
 	ts_state = irq_ts_save();
 
 	asm(".byte 0x0F,0xA7,0xC0 /* xstore %%edi (addr=%0) */"
-		:"=m"(*addr), "=a"(eax_out)
-		:"D"(addr), "d"(edx_in));
+		: "=m" (*addr), "=a" (eax_out), "+d" (edx_in), "+D" (addr));
 
 	irq_ts_restore(ts_state);
 	return eax_out;

Cheers,
-- 
Email: Herbert Xu [off-list ref]
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt