On Jul 27, 2005, at 10:33 AM, Michael L. Hitch wrote:

> On Wed, 27 Jul 2005, Jason Thorpe wrote:
>
>
>>> to November 28 sources to get rid of the free extent problems.  I
>>> think
>>> I've narrowed that problem down to a change Jason made.
>>>
>>
>> Oof... what change?
>>
>
>   Version 1.29 of alpha/common/sgmap_typedep.c.  It increments seg
> everytime through the loop, but may not always load that segment.  I'm
> testing a change where seg is incremented only when is loads the  
> segment.
> If that works on my CS20, I'm going to see if it helps on the ES40.

D'oh!  I guess the patch looks something like this:

Index: alpha/common/sgmap_typedep.c
===================================================================
RCS file: /cvsroot/src/sys/arch/alpha/common/sgmap_typedep.c,v
retrieving revision 1.29
diff -u -p -r1.29 sgmap_typedep.c
--- alpha/common/sgmap_typedep.c        28 Nov 2004 17:34:45  
-0000      1.29
+++ alpha/common/sgmap_typedep.c        27 Jul 2005 23:15:34 -0000
@@ -272,11 +272,12 @@ __C(SGMAP_TYPE,_load_mbuf)(bus_dma_tag_t
         seg = 0;
         error = 0;
-       for (m = m0; m != NULL && error == 0; m = m->m_next, seg++) {
+       for (m = m0; m != NULL && error == 0; m = m->m_next) {
                 if (m->m_len == 0)
                         continue;
                 error = __C(SGMAP_TYPE,_load_buffer)(t, map,
                     m->m_data, m->m_len, NULL, flags, seg, sgmap);
+               seg++;
         }
         alpha_mb();


I bet I made that same mistake in a bunch of other bus_dma back-ends,  
too.


-- thorpej