A fedora user recently filed a puzzling bug at
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=169741
The system being reported has exactly 4GB, and its E820
tables seem to concur that there is in fact 4GB.
When run in non-PAE mode, it triggers the
"Warning only 4GB will be used. Use a PAE enabled kernel."
message, which is odd, but the system does actually run.
When run in PAE mode, it seems to lose its mind, and it
fails to run various binaries.
Booting with mem=4G causes the machine to boot fine
(though for some reason, it finds only 3042M of RAM).
The reporter of this bug has tested on 2.6.14-rc3-git4, and found the
same issue exists as he saw on the original FC3 kernel, thus ruling out
any Fedora-specific patches.
Anyone have any ideas what's wrong here?
Dave
Dave Jones wrote:
>A fedora user recently filed a puzzling bug at
>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=169741
>
>The system being reported has exactly 4GB, and its E820
>tables seem to concur that there is in fact 4GB.
>
>When run in non-PAE mode, it triggers the
>"Warning only 4GB will be used. Use a PAE enabled kernel."
>message, which is odd, but the system does actually run.
>
>
>
if there's a hole in the physical address space (for pci devices), you
would need more than 32 bits to address 4GB RAM.
>When run in PAE mode, it seems to lose its mind, and it
>fails to run various binaries.
>
>Booting with mem=4G causes the machine to boot fine
>(though for some reason, it finds only 3042M of RAM).
>
>
>
looks like a 1GB hole.
>The reporter of this bug has tested on 2.6.14-rc3-git4, and found the
>same issue exists as he saw on the original FC3 kernel, thus ruling out
>any Fedora-specific patches.
>
>Anyone have any ideas what's wrong here?
>
>
>
maybe the last 1GB is bad. since it can only be accessed by pae, only
the pae kernel fails.
--
Do not meddle in the internals of kernels, for they are subtle and quick to panic.
>-----Original Message-----
>From: [email protected]
>[mailto:[email protected]] On Behalf Of avi
>Sent: Wednesday, October 05, 2005 2:32 PM
>To: Dave Jones
>Cc: [email protected]
>Subject: Re: PAE causing failure to run various executables.
>
>Dave Jones wrote:
>
>>A fedora user recently filed a puzzling bug at
>>https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=169741
>>
>>The system being reported has exactly 4GB, and its E820
>>tables seem to concur that there is in fact 4GB.
>>
>>When run in non-PAE mode, it triggers the
>>"Warning only 4GB will be used. Use a PAE enabled kernel."
>>message, which is odd, but the system does actually run.
>>
>>When run in PAE mode, it seems to lose its mind, and it
>>fails to run various binaries.
>>
>>Booting with mem=4G causes the machine to boot fine
>>(though for some reason, it finds only 3042M of RAM).
>>
>>
>>
>looks like a 1GB hole.
Yes. E820 map here indeed shows 1GB of memory at higher than 32 bits
range.
>
>>The reporter of this bug has tested on 2.6.14-rc3-git4, and found the
>>same issue exists as he saw on the original FC3 kernel, thus
>ruling out
>>any Fedora-specific patches.
>>
>>Anyone have any ideas what's wrong here?
>>
>>
>>
>maybe the last 1GB is bad. since it can only be accessed by pae, only
>the pae kernel fails.
>
Another possible reason can be NX(Execute Disable) support, which is
active only on a PAE kernel. Trying "noexec=off" on a PAE kernel can be
tried here.
Thanks,
Venki
>
> Another possible reason can be NX(Execute Disable) support, which is
> active only on a PAE kernel. Trying "noexec=off" on a PAE kernel can be
> tried here.
if that were the case them mem=4096M wouldn't solve the hangs..
Arjan van de Ven wrote:
>>Another possible reason can be NX(Execute Disable) support, which is
>>active only on a PAE kernel. Trying "noexec=off" on a PAE kernel can be
>>tried here.
>>
>>
>
>if that were the case them mem=4096M wouldn't solve the hangs..
>
>
>
then bad memory is more likely. does mem= limit memory size to 4096MB,
or does it limit the maximum physical address?
On Thu, 2005-10-06 at 17:26 +0300, Avi Kivity wrote:
> Arjan van de Ven wrote:
>
> >>Another possible reason can be NX(Execute Disable) support, which is
> >>active only on a PAE kernel. Trying "noexec=off" on a PAE kernel can be
> >>tried here.
> >>
> >>
> >
> >if that were the case them mem=4096M wouldn't solve the hangs..
> >
> >
> >
> then bad memory is more likely. does mem= limit memory size to 4096MB,
> or does it limit the maximum physical address?
highest phys address