mbuf LOR

[Andrew Gallatin]

Nate Lawson writes:
 > I was testing some changes to make fxp MPSAFE and got a LOR in allocating
 > the mbuf cluster and then finally a panic when trying to dereference the
 > cluster header.  Is the mbuf system MPSAFE?  Is it ok to call m_getcl
 > with a device lock held (but not Giant)?
 > 
 > The lock reversal was: 1. fxp softc lock, 2. Giant.
 > 
 > Traceback:
 > zalloc...
 > malloc()
 > mb_pop_cont()
 > mb_alloc()
 > m_getcl()
 > fxp_add_rfabuf()
 > fxp_intr_body()
 > fxp_intr() -- locks fxp softc
 > 
 > -Nate


I think the only place it can be coming from is slab_zalloc().
Does the appended (untested) patch help?

BTW, I don't think that there is any need to get Giant for the zone
allocators in the M_NOWAIT case, but I'm not really familar with the
code, and I don't know if the sparc64 uma_small_alloc needs Giant.

BTW, my MPSAFE driver never sees this, but then again, I never
allocate clusters.  I use jumbo frames, and carve out my own recv
buffers, so I'm only allocating mbufs, not clusters.

Drew


Index: uma_core.c
===================================================================
RCS file: /home/ncvs/src/sys/vm/uma_core.c,v
retrieving revision 1.51
diff -u -r1.51 uma_core.c
--- uma_core.c	26 Mar 2003 18:44:53 -0000	1.51
+++ uma_core.c	3 Apr 2003 18:22:14 -0000
@@ -703,10 +703,14 @@
 		wait &= ~M_ZERO;
 
 	if (booted || (zone->uz_flags & UMA_ZFLAG_PRIVALLOC)) {
-		mtx_lock(&Giant);
-		mem = zone->uz_allocf(zone, 
-		    zone->uz_ppera * UMA_SLAB_SIZE, &flags, wait);
-		mtx_unlock(&Giant);
+		if ((wait & M_NOWAIT) == 0) {
+			mtx_lock(&Giant);
+			mem = zone->uz_allocf(zone, 
+			    zone->uz_ppera * UMA_SLAB_SIZE, &flags, wait);
+			mtx_unlock(&Giant);
+		} else {
+			mem = NULL;
+		}
 		if (mem == NULL) {
 			ZONE_LOCK(zone);
 			return (NULL);