Re: x86: e820 regression

On Mon, Dec 10, 2018 at 12:15:56PM -0500, Erick Cafferata wrote:

On 12/10 11:58, Sasha Levin wrote:

quoted

On Mon, Dec 10, 2018 at 09:21:52AM -0500, Erick Cafferata wrote:

quoted

On 12/10 10:49, Greg KH wrote:

quoted

On Mon, Dec 10, 2018 at 08:54:21AM +0000, Naoya Horiguchi wrote:

quoted

Hi Erick,

On Mon, Dec 10, 2018 at 03:28:37AM -0500, Erick Cafferata wrote:

quoted

The following commit introduced a regression on my system.

124049decbb121ec32742c94fb5d9d6bed8f24d8
x86/e820: put !E820_TYPE_RAM regions into memblock.reserved

and it was backported to stable, stopping the kernel to boot on my system since around 4.17.4.
It was reverted on upstream a couple months ago.
commit 2a5bda5a624d6471d25e953b9adba5182ab1b51f upstream

This commit seems not a correct pointer.
In mainline, commit 124049decbb was reverted by

    commit 9fd61bc95130d4971568b89c9548b5e0a4e18e0e
    Author: Masayoshi Mizuma [off-list ref]
    Date:   Fri Oct 26 15:10:24 2018 -0700

        Revert "x86/e820: put !E820_TYPE_RAM regions into memblock.reserved"

and, the original problem was finally fixed by

    commit 907ec5fca3dc38d37737de826f06f25b063aa08e
    Author: Naoya Horiguchi [off-list ref]
    Date:   Fri Oct 26 15:10:15 2018 -0700

        mm: zero remaining unavailable struct pages

        Patch series "mm: Fix for movable_node boot option", v3.

so I think both patches should be backported onto v4.17.z.

4.17.y and 4.18.y are long end-of-life, there's nothing I can do there.

I can apply the above patches to the 4.19.y tree, is that sufficient?

thanks,

greg k-h

If it were possible to backport it to 4.14 as well. It would be better,
but 4.19 is already good.
Also, would you port only the revert commit, or also the correct fix for
the previous issue?

PD: also, as it was pointed out previously, the correct commit is
9fd61bc95130d4971568b89c9548b5e0a4e18e0e.
PD2: sorry about removing the context in the previous mail.

9fd61bc95130d4971568b89c9548b5e0a4e18e0e looks like the commit that
reverts the patch in question, not an additional fix.

--
Thanks,
Sasha

That's right, that commit is the revert. The commit I'm most interested
in getting backported. However, I was referring to the other 3 commits
affecting arch/x86/kernel/e820.c:

7e1c4e27928e memblock: stop using implicit alignment to SMP_CACHE_BYTES
57c8a661d95d mm: remove include/linux/bootmem.h
2a5bda5a624d memblock: replace alloc_bootmem with memblock_alloc

This 3 probably fixed the original issue, for which

124049decbb1 x86/e820: put !E820_TYPE_RAM regions into memblock.reserved

was pushed. I was asking if those 3(or more, if needed) would get
backported as well.
regards

+ linux-mm@

These commits touch quite a lot of code, and even though they look
simple they are quite invasive, so I wouldn't want to take them without
a proper backport someone tested and acked by the mm folks.

--
Thanks,
Sasha