Regression caused by commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by broken ACPI tables")

Regression caused by commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by broken ACPI tables")

[Jean Delvare]

Hi all,

Commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by
broken ACPI tables") was tagged for stable and merged in various stable
kernel trees (at least 3.2, 3.16, 3.18, 4.1, 4.4, 4.9 and 4.12.)

However it turns out that this commit introduced a regression on some
systems (in particular HPE Superdome 2, but possibly others.) The fix
for this breakage is:

commit 252714155f04c5d16989cb3aadb85fd1b5772f99
Author: Vikas C Sajjan
Date:   Thu Nov 16 21:43:44 2017 +0530

    x86/acpi: Handle SCI interrupts above legacy space gracefully

(and follow-up cleanup 4ee2ec1b1225 "x86/acpi: Reduce code duplication
in mp_override_legacy_irq()".) Unfortunately the fix-up patch was NOT
tagged for stable and also lacks a Fixes tag referring to the faulty
commit. As a consequence nobody realized they fix a regression, and this
regression is still present in all the aforementioned stable branches.

So I invite the maintainers of these stable kernel branches to backport
commits 252714155f04 and possibly 4ee2ec1b1225 to solve this issue.

Thanks,
-- 
Jean Delvare
SUSE L3 Support

Re: Regression caused by commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by broken ACPI tables")

[Greg KH]

On Wed, Jan 10, 2018 at 07:28:22PM +0100, Jean Delvare wrote:
> Hi all,
> 
> Commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by
> broken ACPI tables") was tagged for stable and merged in various stable
> kernel trees (at least 3.2, 3.16, 3.18, 4.1, 4.4, 4.9 and 4.12.)
> 
> However it turns out that this commit introduced a regression on some
> systems (in particular HPE Superdome 2, but possibly others.) The fix
> for this breakage is:
> 
> commit 252714155f04c5d16989cb3aadb85fd1b5772f99
> Author: Vikas C Sajjan
> Date:   Thu Nov 16 21:43:44 2017 +0530
> 
>     x86/acpi: Handle SCI interrupts above legacy space gracefully
> 
> (and follow-up cleanup 4ee2ec1b1225 "x86/acpi: Reduce code duplication
> in mp_override_legacy_irq()".) Unfortunately the fix-up patch was NOT
> tagged for stable and also lacks a Fixes tag referring to the faulty
> commit. As a consequence nobody realized they fix a regression, and this
> regression is still present in all the aforementioned stable branches.
> 
> So I invite the maintainers of these stable kernel branches to backport
> commits 252714155f04 and possibly 4ee2ec1b1225 to solve this issue.

Many thanks for the report, I've queued up 252714155f04 now.  I don't
think 4ee2ec1b1225 needs to be merged, unless the acpi maintainers think
it is a good idea to do so.

Rafael, any thoughts?

thanks,

greg k-h

Re: Regression caused by commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by broken ACPI tables")

[Jean Delvare]

Hi Greg,

On Wed, 10 Jan 2018 20:10:44 +0100, Greg KH wrote:
> Many thanks for the report, I've queued up 252714155f04 now.  I don't
> think 4ee2ec1b1225 needs to be merged, unless the acpi maintainers think
> it is a good idea to do so.
> 
> Rafael, any thoughts?

FWIW we applied both in SUSE kernels. I did not make the decision but I
approve it. Having duplicated code in our kernel when there's only one
occurrence of the same code upstream looks dangerous. If there ever is
a bug fixed in that function and we backport the fix, you can be sure
that only one of the copies will get fixed and the bug will remain in
the other.

-- 
Jean Delvare
SUSE L3 Support

Re: Regression caused by commit dad5ab0db8de ("x86/acpi: Prevent out of bound access caused by broken ACPI tables")

[Greg KH]

On Wed, Jan 10, 2018 at 09:03:18PM +0100, Jean Delvare wrote:
> Hi Greg,
> 
> On Wed, 10 Jan 2018 20:10:44 +0100, Greg KH wrote:
> > Many thanks for the report, I've queued up 252714155f04 now.  I don't
> > think 4ee2ec1b1225 needs to be merged, unless the acpi maintainers think
> > it is a good idea to do so.
> > 
> > Rafael, any thoughts?
> 
> FWIW we applied both in SUSE kernels. I did not make the decision but I
> approve it. Having duplicated code in our kernel when there's only one
> occurrence of the same code upstream looks dangerous. If there ever is
> a bug fixed in that function and we backport the fix, you can be sure
> that only one of the copies will get fixed and the bug will remain in
> the other.

Good idea, I've now queued it up as well, thanks.

greg k-h