megaraid 64bit fix/cleanup (AMD64)

PATCH Andi Kleen 2 participants 2 messages Jan 1, 2004 1 message not in the archive

lore

Share & tools

lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Andi Kleen

author

2004-01-01 13:05 UTC #1 lore raw

Review line 1295

Quoted 13 lines · Brad House

diff -ruN linux-2.6.0-gentoo-r1.old/drivers/scsi/megaraid.c linux-2.6.0-gentoo-r1/drivers/scsi/megaraid.c
--- linux-2.6.0-gentoo-r1.old/drivers/scsi/megaraid.c 2003-12-29
23:51:43.000000000 -0500
+++ linux-2.6.0-gentoo-r1/drivers/scsi/megaraid.c 2003-12-29
23:54:01.005469936 -0500
(quoted diff) +1 -1
@@ -1292,7 +1292,7 @@
 
 			/* Calculate Scatter-Gather info */
 			mbox->m_out.numsgelements = mega_build_sglist(adapter, scb,
-					(u32 *)&mbox->m_out.xferaddr, (u32 *)&seg);
+					(dma_addr_t *)&mbox->m_out.xferaddr, (u32 *)&seg);

I'm pretty sure it's completely broken. You're changing the layout of a data structure that is shared with the firmware. Using a 32bit int here is fine when the driver sets the correct dma mask or only stuffs pci_alloc_consistent() memory in there (i think it's the later here)

Even though the driver prints lots of warnings at compile time it actually works on AMD64 as is. But in many cases you should use megaraid2.c instead of megaraid.c.

-Andi

Brad House

2004-01-01 14:08 UTC #2 lore raw

Well, the point was it _didn't_ work on AMD64 with 2.6.0 at all. Which was what I was trying to fix. Also, it is a megaraid2 series driver in the 2.6.0 kernel (2.00.3), I've since ported the 2.00.9 2.4 driver to 2.6.0, but have not yet had it tested (I did not make the 64bit changes I had posted a patch for).
I guess we'll see, I hope bumping to 2.00.9 works.

-Brad

Quoted 26 lines

Review line 1295
Quoted 13 lines · Brad House
diff -ruN linux-2.6.0-gentoo-r1.old/drivers/scsi/megaraid.c linux-2.6.0-gentoo-r1/drivers/scsi/megaraid.c
--- linux-2.6.0-gentoo-r1.old/drivers/scsi/megaraid.c 2003-12-29
23:51:43.000000000 -0500
+++ linux-2.6.0-gentoo-r1/drivers/scsi/megaraid.c 2003-12-29
23:54:01.005469936 -0500
(quoted diff) +1 -1
@@ -1292,7 +1292,7 @@
 
 			/* Calculate Scatter-Gather info */
 			mbox->m_out.numsgelements = mega_build_sglist(adapter, scb,
-					(u32 *)&mbox->m_out.xferaddr, (u32 *)&seg);
+					(dma_addr_t *)&mbox->m_out.xferaddr, (u32 *)&seg);
I'm pretty sure it's completely broken. You're changing the layout of a data structure that is shared with the firmware. Using a 32bit int here is fine when the driver sets the correct dma mask or only stuffs pci_alloc_consistent() memory in there (i think it's the later here)
Even though the driver prints lots of warnings at compile time it actually works on AMD64 as is. But in many cases you should use megaraid2.c instead of megaraid.c.
-Andi