Skip to contentSkip to conversation

qla2xxxx driver for IA64

DISCUSSION Evan Felix 4 participants 4 messages Nov 19, 2003 – Nov 24, 2003

lore
Share & tools
lore

Evan Felix

author
#0 lore raw

Andrew, I downloaded the qla2xxx-src-v8.00.00b6.tar.bz2 driver from sourceforge, and had a few issues getting it woking under IA64 on a rx2600 system. We have two dual channel qla2312 cards.

- I had to add '#include <vmalloc.h>' to qla_os.c
- the IA64 ioctl.h file did not hav the _IOXX_BAD macros, and i added them, i'm not sure how safe that was, but it compiled.

The driver then built fine, and we have been testing it for a couple days now.

Evan

Signature · 6 lines
-------------------------
Evan Felix
Administrator of Supercomputer #5 in Top 500, Nov 2003
Environmental Molecular Sciences Laboratory
Pacific Northwest National Laboratory
Operated for the U.S. DOE by Battelle
#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Evan Felix

Evan> the IA64 ioctl.h file did not hav the _IOXX_BAD macros, and i Evan> added them, i'm not sure how safe that was, but it compiled.

Last time I checked, the _IOXX_BAD macros were not used anywhere in the standard kernel and that's why I leaned against including them (they seemed like a very temporary thing to me). Perhaps the driver could be fixed instead to not use the bad macros?

	--david
#2 lore raw
In reply to Evan Felix
Quoted 10 lines · David Mosberger
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines · Evan Felix

Evan> the IA64 ioctl.h file did not hav the _IOXX_BAD macros, and i Evan> added them, i'm not sure how safe that was, but it compiled.

Last time I checked, the _IOXX_BAD macros were not used anywhere in the standard kernel and that's why I leaned against including them (they seemed like a very temporary thing to me). Perhaps the driver could be fixed instead to not use the bad macros?

Andrew,

In the v8.00.00b6 version the function pointer for the ioctl() handler is not set up in qla2x00_driver_template in qla_os.c so none of the ioctl()s work anyway.

This patch fixes it.

scsi/qla2xxx/qla_os.c renamed from scsi/qla2xxx.v8.00.00b6.orig/qla_os.c +1 -0
@@ -405,6 +405,7 @@
 	.module			= THIS_MODULE,
 	.name			= DRIVER_NAME,
 	.proc_name		= DRIVER_NAME,
+	.ioctl	    	    	= qla2x00_ioctl,
 	.proc_info		= qla2x00_proc_info,
 	.queuecommand		= qla2x00_queuecommand,
 

Also, possibly one of the contributing factors which required the use of _IOWR_BAD instead of _IOWR is the spurious sizeof() in exioctln.h:

#define QL_IOCTL_BASE(idx) \
_IOWR_BAD(QLMULTIPATH_MAGIC, idx, sizeof(EXT_IOCTL))
^^^^^^

The macro expects a type as the third argument, not a size (despite the name of the parameter in the macro definition).

Greg.

Signature · 2 lines
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
I don't speak for SGI.
#3 lore raw
In reply to Evan Felix

All,

Quoted 5 lines

<[email protected]> said:

Evan> the IA64 ioctl.h file did not hav the _IOXX_BAD macros, and i Evan> added them, i'm not sure how safe that was, but it compiled.

The cooresponding callers of the IOCTLs have not had the changes added to support the 8.x driver -- you should be ok for now.

Quoted 5 lines

Last time I checked, the _IOXX_BAD macros were not used anywhere in the standard kernel and that's why I leaned against including them (they seemed like a very temporary thing to me). Perhaps the driver could be fixed instead to not use the bad macros?

Yes, the next driver drop will have the _BAD() macros usage addressed.

Thanks,
Andrew Vasquez