kern/169526: [patch] sys/dev/sym broken by r236061 on amd64 with more than 4GB RAM

Mike Watters mike at
Thu Jun 28 17:20:09 UTC 2012

>Number:         169526
>Category:       kern
>Synopsis:       [patch] sys/dev/sym broken by r236061 on amd64 with more than 4GB RAM
>Confidential:   no
>Severity:       serious
>Priority:       low
>Responsible:    freebsd-bugs
>State:          open
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Thu Jun 28 17:20:08 UTC 2012
>Originator:     Mike Watters
>Release:        8-STABLE
FreeBSD 8.3-STABLE #4: Wed Jun 27 00:13:04 MDT 2012
    srcbuild@:/usr/obj/usr/src-8/sys/GENERIC amd64
Related PR: 168928 (different bug in same revision)

real memory  = 8589934592 (8192 MB)
avail memory = 7983443968 (7613 MB)

sym0: <1010-66> port 0xe000-0xe0ff mem 0xfbfffc00-0xfbffffff,0xfbffc000-0xfbffdfff irq 21 at device 7.0 on pci3
sym0: failed to create DMA tag.
device_attach: sym0 attach returned 6

(followed by memory-related errors and boot failure)

In r236061 to sys/dev/sym/sym_hipd.c, the flags parameter to
bus_dma_tag_create() was changed from 0 to BUS_DMA_ALLOCNOW, which
breaks the driver on my amd64 system with 8GB RAM.

If I follow the logic correctly, the BUS_DMA_ALLOCNOW flag combined
with the maxsize parameter being BUS_SPACE_MAXSIZE results in the
system trying to allocate 4GB worth of bounce pages, which I assume
exhausts kernel resources.

Change flags parameter from BUS_DMA_ALLOCNOW back to 0, or change
maxsize parameter to a more appropriate smaller value.  Changing flags to 0
(BUS_DMA_WAITOK) fixes the problem on my system.

Patch attached with submission follows:

--- sym_hipd.c.orig	2012-06-17 03:43:04.000000000 -0600
+++ sym_hipd.c	2012-06-28 11:01:36.335984640 -0600
@@ -8538,7 +8538,7 @@
 	if (bus_dma_tag_create(np->bus_dmat, 1, SYM_CONF_DMA_BOUNDARY,
-	    BUS_DMA_ALLOCNOW, busdma_lock_mutex, &np->mtx, &np->data_dmat)) {
+	    BUS_DMA_WAITOK, busdma_lock_mutex, &np->mtx, &np->data_dmat)) {
 		device_printf(dev, "failed to create DMA tag.\n");
 		goto attach_failed;


More information about the freebsd-bugs mailing list