[Bug 1223] SD card no longer operational with 2.6.24 (the never-ending saga)

[Mike (mwester)]

Er, I didn't see a commit with this fix yet; so the SD card on the GTA01 
is still being corrupted when folks boot up the 2.6.24 kernel.  On the 
chance that folks perhaps didn't see this email due to over-agressive 
SPAM filters, I've stripped the URLs out of it, and attached the patch 
directly. :(

Mike


Mike (mwester) wrote:
> Mike (mwester) wrote:
>> Mike (mwester) wrote:
>>> Tested.  The write fix is *definitely* needed as well; revision 4080 
>>> fixes both read and write.  Thanks all - we're pretty close to being 
>>> able to call the .24 kernel "stable" for the gta01, I think.
>>
>> No we're not.  I spoke too soon, that'll teach me to always test 
>> thoroughly.  Reads work fine with this change, but writes fail in 
>> strange ways.  My apologies to anyone inconvenienced by my erroneous 
>> declaration of success :(
>>
>> We need to decrement fifo as we write the unaligned data out in the 
>> first loop (s3cmci.c):
>
> Still not there (yikes!).  The commit for svn4082 committed the change 
> from this last email, but it got applied to a different area of the 
> source file.  So that commit didn't fix the problem.  (That's the bad 
> news).
>
> Now the good news: that commit exposed a similar problem in a 
> different area of the code -- same loop structure, needs the same 
> fixes.  This time around I spent a little extra time doing a review of 
> the two functions, and ended up changing the structure of the loops 
> slightly:  "while ((condition1) && fifo--)" is a bad idea if you need 
> to ensure that fifo doesn't go below zero when the loop exits (follow 
> the code with the case where we have an unaligned buffer, and need to 
> read or write 1 byte).
>
> Anyway, this time, to avoid ambiguity, here's a URL [1] to the patch 
> that should be applied to the svn4082 s3cmci.c source file.  The 
> entire source file is also available [2].
>
> With this, I sure hope this undead dragon will finally stay dead.
>
> Mike (mwester)
>

URLS below edited in hopes of avoiding SPAM filters.

> [1] moko-dot-mwester-dot-net/s3cmci_diff_from_svn4082.patch
> [2] moko-dot-mwester-dot-net/s3cmci.c
>
Patch follows - yes, tabs and spaces are screwed up with this mailer, 
see the pseudo-urls above for the
un-damaged patch (I'm just not in a place where I can fix the mailer 
issue at the moment):

--- linux-2.6.24/drivers/mmc/host/s3cmci.c.4082    2008-02-18 
20:26:37.000000000 -0600
+++ linux-2.6.24/drivers/mmc/host/s3cmci.c    2008-02-18 
21:12:59.000000000 -0600
@@ -251,8 +251,10 @@
         host->pio_count += fifo;
 
         /* we might have an unaligned start of data */
-        while (((unsigned long)host->pio_ptr & 0x03) && fifo--)
+        while (((unsigned long)host->pio_ptr & 0x03) && fifo) {
             *(host->pio_ptr++) = readb(host->base + host->sdidata_b);
+            fifo--;
+        }
 
         /* and a major chunk of data in the middle */
         for (; fifo >= 4; fifo -=4) {
@@ -261,8 +263,10 @@
         }
 
         /* as well as some non-modulo-four trailer */
-        while (fifo)
+        while (fifo) {
             *(host->pio_ptr++) = readb(host->base + host->sdidata_b);
+            fifo--;
+        }
     }
 
     if (!host->pio_bytes) {
@@ -315,8 +319,10 @@
         host->pio_count += fifo;
 
         /* we might have an unaligned start of data */
-        while ((unsigned long)host->pio_ptr & 0x03)
+        while (((unsigned long)host->pio_ptr & 0x03) && fifo) {
             writeb(*(host->pio_ptr++), host->base + host->sdidata_b);
+            fifo--;
+        }
 
         /* and a major chunk of data in the middle */
         for (; fifo >= 4; fifo -=4) {
@@ -325,8 +331,10 @@
         }
 
         /* as well as some non-modulo-four trailer */
-        while (fifo--)
+        while (fifo) {
             writeb(*(host->pio_ptr++), host->base + host->sdidata_b);
+            fifo--;
+        }
     }
 
     enable_imask(host, S3C2410_SDIIMSK_TXFIFOHALF);