Bug 101932 – Grub doesn't recognize all memory. Boot fails with error 28

• Product: Gentoo Linux
• Status: RESOLVED DUPLICATE

[Description John E. Vincent 2005-08-09 20:17:25 UTC]

I'm building a new EMT64 server. I've got this exact same hardware on another
production server already with no problems. The only difference is a IBM
ServeRAID 6M card installed.

grub-0.96-r2 is not seeing all the memory (4GB) and unless I add mem=4096M to
the boot line, it fails to boot with an Error 28.

The grub menu even shows the memory below 4GB (even if I add the 640k into the mix.

Reproducible: Always
Steps to Reproduce:
1. install grub-0.96-r2
2. boot kernel with initrd
3. fails to boot
4. add mem=4096M to the kernel line
5. machine boots but only sees 3.xGB or so of memory



Expected Results:  
Boot with all the memory and not require memory line.

Hardware - IBM x346 EMT64 with 4GB of memory
Kernel - 2.6.12-gentoo-r6
Grub - 0.96-r2

[Comment 1 Mike Doty (RETIRED) 2005-08-09 20:23:35 UTC]

Most modern boards reserve a portion of memory that is reported by both the BIOS
and grub.  Without the 'mem=4096', I imagine you get somewhere between 3.7 and
3.8G.  This isn't a bug with eithe gentoo or linux, but by design of the
motherboard and expansion cards.  If you are getting significantly less
memory(2G), then feel free to reopen the bug.

[Comment 2 John E. Vincent 2005-08-09 20:25:44 UTC]

I just noticed this line in dmesg:

Memory: 3166624k/4194304k available (2759k kernel code, 109204k reserved, 1020k
data, 584k init)

I also attempted to run grub-0.96-r1 and I had the same problem.

[Comment 3 John E. Vincent 2005-08-09 20:26:44 UTC]

I just saw the response from the dev. What get's me is that this same hardware
MINUS the ServeRAID doesn't exhibit this problem. I've checked the BIOS but
maybe the serveraid really is reserving that much memory?

[Comment 4 John E. Vincent 2005-08-09 20:35:21 UTC]

And as a side note, GRUB is still giving the memory error as referenced in the
link above.

[Comment 5 Mike Doty (RETIRED) 2005-08-09 21:02:58 UTC]

take out the raid card and boot the livecd and see if it is a similar amount of
memory.  To give you an idea, my promise sata controller takes about 30M and my
VIA sata controller takes about 50M.  Other devices may take a trivial amount.

[Comment 6 John E. Vincent 2005-08-09 21:06:01 UTC]

I don't think it's the servraid taking up the memory. I rebooted off the LiveCD
and saw all of my memory just fine.

[Comment 7 John E. Vincent 2005-08-09 21:08:19 UTC]

I'll try taking the servraid out tomorrow and seeing what happens as well. I'm
currently updating the bios on the raid controller. I wonder if it has something
to do with my initrd. As I said the interesting thing is that the LiveCD
actually does see all 4GB. We had this same problem on Windows 2003 Advanced
Server that was living on this box before we repurposed it but I chalked it up
to a Windows problem. 

[Comment 8 Jakub Moc (RETIRED) 2005-08-10 02:50:52 UTC]

(In reply to comment #7)
> As I said the interesting thing is that the LiveCD
> actually does see all 4GB. 

You need to have 

CONFIG_HIGHMEM4G=y
CONFIG_HIGHMEM=y

enabled in kernel. Did you enable this?

[Comment 9 John E. Vincent 2005-08-10 13:15:17 UTC]

(In reply to comment #8)
> (In reply to comment #7)
> > As I said the interesting thing is that the LiveCD
> > actually does see all 4GB. 
> 
> You need to have 
> 
> CONFIG_HIGHMEM4G=y
> CONFIG_HIGHMEM=y
> 
> enabled in kernel. Did you enable this?

That option isn't available in 64bit kernel configurations. If it's SUPPOSED to
be there, then there is a bug with gentoo-sources-2.6.12-r6 that is causing it
NOT to show in the menu anywhere.

[Comment 10 Olivier Crete (RETIRED) 2005-08-10 15:37:59 UTC]

in reply to comments #8/9 .. you dont need HIGHMEM since the addressing is 64bit
and you dont need any of the crap that they do on x86 to compensate for 32bit
addressing. 

[Comment 11 John E. Vincent 2005-08-10 16:04:03 UTC]

(In reply to comment #10)
> in reply to comments #8/9 .. you dont need HIGHMEM since the addressing is 64bit
> and you dont need any of the crap that they do on x86 to compensate for 32bit
> addressing. 

That's what I thought.

[Comment 12 SpanKY 2005-08-19 17:04:20 UTC]

*** This bug has been marked as a duplicate of 99897 ***