[Bug 193400] [mips] r269577 broke operation on MIPS32

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=193400

--- Comment #2 from Adrian Chadd <adrian at freebsd.org> ---
Ok, here's what fixed it (ignore the error checks):

adrian at adrian-hackbox:~/work/freebsd/embedded/head/src/sys % svn diff kern
Index: kern/subr_sfbuf.c
===================================================================
--- kern/subr_sfbuf.c   (revision 271210)
+++ kern/subr_sfbuf.c   (working copy)
@@ -77,6 +77,14 @@
  */
 static struct mtx sf_buf_lock;

+#ifndef SFBUF
+#error wtf?
+#endif
+
+#ifdef SFBUF_OPTIONAL_DIRECT_MAP
+#error wtf?
+#endif
+
 /*
  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
  */
@@ -127,6 +135,7 @@
            ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
        hash_list = &sf_buf_active[SF_BUF_HASH(m)];
        mtx_lock(&sf_buf_lock);
+#if 0
        LIST_FOREACH(sf, hash_list, list_entry) {
                if (sf->m == m) {
                        sf->ref_count++;
@@ -141,6 +150,7 @@
                        goto done;
                }
        }
+#endif
        while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
                if (flags & SFB_NOWAIT)
                        goto done;

.. the original MIPS sfbuf code didn't do the refcount stuff; it just allocated
a new page for each sfbuf requested. I don't know how correct that was, but it
worked.

So maybe you've exposed a bug in the MIPS pmap code?


-a

-- 
You are receiving this mail because:
You are the assignee for the bug.