Skip to contentSkip to conversation

/proc/mtrr I/O error

BUG Bob Tracy 3 participants 5 messages Aug 29, 2008 – Sep 3, 2008

lore
Share & tools
lore

Bob Tracy

author
#0 lore raw

With 2.6.27-rc4 and possibly earlier kernels, I see the following:

$ cat /proc/mtrr
cat: /proc/mtrr: Input/output error

Processor is an AMD K6-III/450.

Signature · 4 lines
------------------------------------------------------------------------
Bob Tracy          |  "I was a beta tester for dirt.  They never did
[email protected]       |   get all the bugs out." - Steve McGrew on /.
------------------------------------------------------------------------
#1 lore raw
Quoted 7 lines · On Fri, 29 Aug 2008 16:38:24 -0500 Bob Tracy <[email protected]> wrote:

With 2.6.27-rc4 and possibly earlier kernels, I see the following:

$ cat /proc/mtrr
cat: /proc/mtrr: Input/output error

Processor is an AMD K6-III/450.

static int mtrr_open(struct inode *inode, struct file *file)

{
	if (!mtrr_if) 
		return -EIO;

Beats me. Perhaps something broke in mtrr_bp_init().

It would be useful to know if 2.6.26 worked OK.

Thanks.

#2 lore raw
Quoted 18 lines · Andrew Morton
Quoted 7 lines · On Fri, 29 Aug 2008 16:38:24 -0500 Bob Tracy <[email protected]> wrote:

With 2.6.27-rc4 and possibly earlier kernels, I see the following:

$ cat /proc/mtrr
cat: /proc/mtrr: Input/output error

Processor is an AMD K6-III/450.

static int mtrr_open(struct inode *inode, struct file *file)

{
       if (!mtrr_if)
               return -EIO;
Quoted 4 lines

Beats me. Perhaps something broke in mtrr_bp_init().

It would be useful to know if 2.6.26 worked OK.

two patches needed..

http://lkml.org/lkml/2008/9/1/307
[PATCH 4/6] x86: delay early cpu initialization until cpuid is done
http://lkml.org/lkml/2008/9/1/308
[PATCH 5/6] x86: move mtrr cpu cap setting early in early_init_xxxx

YH

author
#3 lore raw
Quoted 24 lines · Yinghai Lu
Quoted 18 lines · Andrew Morton
Quoted 7 lines · On Fri, 29 Aug 2008 16:38:24 -0500 Bob Tracy <[email protected]> wrote:

With 2.6.27-rc4 and possibly earlier kernels, I see the following:

$ cat /proc/mtrr
cat: /proc/mtrr: Input/output error

Processor is an AMD K6-III/450.

static int mtrr_open(struct inode *inode, struct file *file)

{
       if (!mtrr_if)
               return -EIO;
Quoted 4 lines

Beats me. Perhaps something broke in mtrr_bp_init().

It would be useful to know if 2.6.26 worked OK.

Quoted 6 lines

two patches needed..

http://lkml.org/lkml/2008/9/1/307
[PATCH 4/6] x86: delay early cpu initialization until cpuid is done
http://lkml.org/lkml/2008/9/1/308
[PATCH 5/6] x86: move mtrr cpu cap setting early in early_init_xxxx

Krzysztof Helt's quick-and-dirty patch posted a few days ago did the trick, but from the descriptions of the above patches, they address the problem properly. Here's what "cat /proc/mtrr" shows with Helt's patch and the "nv" driver active for my GeForce2 MX 400 card:

$ cat /proc/mtrr
reg00: base=0xd0000000 (3328MB), size=  64MB: write-combining, count=1

MTRR sanitization seems to be working properly: the same command shows nothing when X isn't running.

Signature · 4 lines
------------------------------------------------------------------------
Bob Tracy          |  "I was a beta tester for dirt.  They never did
[email protected]       |   get all the bugs out." - Steve McGrew on /.
------------------------------------------------------------------------
#4 lore raw
Quoted 37 lines · Bob Tracy
Quoted 24 lines · Yinghai Lu
Quoted 18 lines · Andrew Morton
Quoted 7 lines · On Fri, 29 Aug 2008 16:38:24 -0500 Bob Tracy <[email protected]> wrote:

With 2.6.27-rc4 and possibly earlier kernels, I see the following:

$ cat /proc/mtrr
cat: /proc/mtrr: Input/output error

Processor is an AMD K6-III/450.

static int mtrr_open(struct inode *inode, struct file *file)

{
       if (!mtrr_if)
               return -EIO;
Quoted 4 lines

Beats me. Perhaps something broke in mtrr_bp_init().

It would be useful to know if 2.6.26 worked OK.

Quoted 6 lines

two patches needed..

http://lkml.org/lkml/2008/9/1/307
[PATCH 4/6] x86: delay early cpu initialization until cpuid is done
http://lkml.org/lkml/2008/9/1/308
[PATCH 5/6] x86: move mtrr cpu cap setting early in early_init_xxxx

Krzysztof Helt's quick-and-dirty patch posted a few days ago did the trick, but from the descriptions of the above patches, they address the problem properly. Here's what "cat /proc/mtrr" shows with Helt's patch and the "nv" driver active for my GeForce2 MX 400 card:

Quoted 5 lines
$ cat /proc/mtrr
reg00: base=0xd0000000 (3328MB), size=  64MB: write-combining, count=1

MTRR sanitization seems to be working properly: the same command shows nothing when X isn't running.

did you apply them both
[PATCH 4/6] x86: delay early cpu initialization until cpuid is done [PATCH 5/6] x86: move mtrr cpu cap setting early in early_init_xxxx

YH