Re: kern/49120: RPI ld/sdmmc driver fails after a while

[Frank Kardel <kardel%netbsd.org@localhost>]

The following reply was made to PR kern/49120; it has been noted by GNATS.

From: Frank Kardel <kardel%netbsd.org@localhost>
To: Nick Hudson <skrll%netbsd.org@localhost>, gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: kern/49120: RPI ld/sdmmc driver fails after a while
Date: Thu, 21 Aug 2014 13:18:26 +0200

 Not yet. I scanned through the diffs between 2014-07-31 to 2014-08-08 (I 
 hope the culprit is in there at that time).
 
 There was not much relating ARM. but I saw an ARMV6 related item for a 
 speculation (though I have not scanned for any impact):
 
 RCS file: /cvsroot/src/sys/arch/arm/include/types.h,v
 retrieving revision 1.27
 retrieving revision 1.28
 diff -u -r1.27 -r1.28
 --- arch/arm/include/types.h    18 Apr 2014 15:45:46 -0000      1.27
 +++ arch/arm/include/types.h    5 Aug 2014 06:24:56 -0000       1.28
 @@ -1,4 +1,4 @@
 -/*     $NetBSD: types.h,v 1.27 2014/04/18 15:45:46 matt Exp $  */
 +/*     $NetBSD: types.h,v 1.28 2014/08/05 06:24:56 skrll Exp $ */
 
   /*
    * Copyright (c) 1990 The Regents of the University of California.
 @@ -93,9 +93,6 @@
   #if defined(__ARM_EABI__) && defined(_ARM_ARCH_6)
   #define        __HAVE_ATOMIC64_OPS
   #endif
 -#if defined(_ARM_ARCH_6)
 -#define        __NO_STRICT_ALIGNMENT
 -#endif
 
 But that is just a wild guess,
 
 It happens on different RPIs with different cards (16Gb and 32Gb). Older 
 kernels run fine for now 18d with a kernel from July 31st.
 
 I have not started bisecting it as the test takes > 20h per iteration.
 Maybe we need to go through that if nothing else springs to mind.
 
 The strange thing is, that it takes so long and I could not trigger it 
 earlier by running 'find / -ls' over and over.