cxgb(4): Remove assumption of physically contiguous mbufs.

Investigation of iSCSI target data corruption reports brought me to
discovery that cxgb(4) expects mbufs to be physically contiguous, that
is not true after I've started using m_extaddref() in software iSCSI
for large zero-copy transmissions.  In case of fragmented memory the
driver transmitted garbage from pages following the first one due to
simple use of pmap_kextract() for the first pointer instead of proper
bus_dmamap_load_mbuf_sg().  Seems like it was done as some optimization
many years ago, and at very least it is wrong in a world of IOMMUs.

This patch just removes that optimization, plus limits packet coalescing
for mbufs crossing page boundary, also depending on assumption of one
segment per packet.

MFC after:	3 days
Sponsored by:	iXsystems, Inc.
Reviewed by:	mmacy, np
Differential revision:	https://reviews.freebsd.org/D28428

3 files changed, 5 insertions, 36 deletions

diff --git a/sys/dev/cxgb/cxgb_sge.c b/sys/dev/cxgb/cxgb_sge.c
index 7f456ccff4ca..491d1a751f4a 100644
--- a/sys/dev/cxgb/cxgb_sge.c
+++ b/sys/dev/cxgb/cxgb_sge.c
@@ -322,7 +322,8 @@ coalesce_check(struct mbuf *m, void *arg)
 	int *nbytes = &ci->nbytes;
 
 	if ((*nbytes == 0) || ((*nbytes + m->m_len <= 10500) &&
-		(*count < 7) && (m->m_next == NULL))) {
+	    (*count < 7) && (m->m_next == NULL) &&
+	    ((mtod(m, vm_offset_t) & PAGE_MASK) + m->m_len <= PAGE_SIZE))) {
 		*count += 1;
 		*nbytes += m->m_len;
 		return (1);
diff --git a/sys/dev/cxgb/sys/mvec.h b/sys/dev/cxgb/sys/mvec.h
index bdd0b55c5489..4989bff29ec4 100644
--- a/sys/dev/cxgb/sys/mvec.h
+++ b/sys/dev/cxgb/sys/mvec.h
@@ -33,20 +33,6 @@
 #define _MVEC_H_
 #include <machine/bus.h>
 
-static __inline void
-busdma_map_mbuf_fast(bus_dma_tag_t tag, bus_dmamap_t map,
-    struct mbuf *m, bus_dma_segment_t *seg)
-{
-#if defined(__i386__) || defined(__amd64__)
-	seg->ds_addr = pmap_kextract(mtod(m, vm_offset_t));
-	seg->ds_len = m->m_len;
-#else
-	int nsegstmp;
-
-	bus_dmamap_load_mbuf_sg(tag, map, m, seg, &nsegstmp, 0);
-#endif
-}
-
 int busdma_map_sg_collapse(bus_dma_tag_t tag, bus_dmamap_t map,
     struct mbuf **m, bus_dma_segment_t *segs, int *nsegs);
 void busdma_map_sg_vec(bus_dma_tag_t tag, bus_dmamap_t map,
diff --git a/sys/dev/cxgb/sys/uipc_mvec.c b/sys/dev/cxgb/sys/uipc_mvec.c
index ca31cf1897c7..02f437079468 100644
--- a/sys/dev/cxgb/sys/uipc_mvec.c
+++ b/sys/dev/cxgb/sys/uipc_mvec.c
@@ -65,26 +65,7 @@ busdma_map_sg_collapse(bus_dma_tag_t tag, bus_dmamap_t map,
 retry:
 	psegs = segs;
 	seg_count = 0;
-	if (n->m_next == NULL) {
-		busdma_map_mbuf_fast(tag, map, n, segs);
-		*nsegs = 1;
-		return (0);
-	}
-#if defined(__i386__) || defined(__amd64__)
-	while (n && seg_count < TX_MAX_SEGS) {
-		/*
-		 * firmware doesn't like empty segments
-		 */
-		if (__predict_true(n->m_len != 0)) {
-			seg_count++;
-			busdma_map_mbuf_fast(tag, map, n, psegs);
-			psegs++;
-		}
-		n = n->m_next;
-	}
-#else
 	err = bus_dmamap_load_mbuf_sg(tag, map, *m, segs, &seg_count, 0);
-#endif	
 	if (seg_count == 0) {
 		if (cxgb_debug)
 			printf("empty segment chain\n");
@@ -117,8 +98,9 @@ void
 busdma_map_sg_vec(bus_dma_tag_t tag, bus_dmamap_t map,
     struct mbuf *m, bus_dma_segment_t *segs, int *nsegs)
 {
+	int n = 0;
 
-	for (*nsegs = 0; m != NULL ; segs++, *nsegs += 1, m = m->m_nextpkt)
-		busdma_map_mbuf_fast(tag, map, m, segs);
+	for (*nsegs = 0; m != NULL; segs += n, *nsegs += n, m = m->m_nextpkt)
+		bus_dmamap_load_mbuf_sg(tag, map, m, segs, &n, 0);
 }