arm64 crashkernel fails to boot on acpi-only machines due to ACPI regions being no longer mapped as NOMAP

Hi Ard, Akashi

On Mon, Dec 4, 2017 at 7:32 PM, Ard Biesheuvel
<ard.biesheuvel@linaro.org> wrote:
> On 26 November 2017 at 08:29, Bhupesh SHARMA <bhupesh.linux@gmail.com> wrote:
>> Hi Akashi,
>>
>> On Thu, Nov 16, 2017 at 12:30 PM, AKASHI Takahiro
>> <takahiro.akashi@linaro.org> wrote:
>>> Bhupesh,
>>>
>>> On Wed, Nov 15, 2017 at 04:28:55PM +0530, Bhupesh Sharma wrote:
>>>>
>>>   (snip)
>>>
>>>> # dmesg | grep -B 2 -i "ACPI reclaim"
>>>> [    0.000000] efi:   0x000039670000-0x0000396bffff [Runtime Code |RUN|  |
>>>> |  |  |  |  |   |WB|WT|WC|UC]
>>>> [    0.000000] efi:   0x0000396c0000-0x00003970ffff [Boot Code |   |  |  |
>>>> |  |  |  |   |WB|WT|WC|UC]
>>>> [    0.000000] efi:   0x000039710000-0x00003975ffff [ACPI Reclaim Memory|
>>>> |  |  |  |  |  |  |   |WB|WT|WC|UC]
>>>>
>>>> 2. Now, I am not sure which kernel layer does the following changes (I am
>>>> still trying to dig it out more), but I see that the 'Boot Code' and ACPI
>>>> DSDT table regions are somehow merged into one memblock_region and appear as
>>>> range '396c0000-3975ffff' in the '/proc/iomem' interface:
>>>>
>>>> # cat /proc/iomem | grep -A 2 -B 2 39
>>>> 00000000-3961ffff : System RAM
>>>>   00080000-00b6ffff : Kernel code
>>>>   00cb0000-0167ffff : Kernel data
>>>>   0e800000-2e7fffff : Crash kernel
>>>> 39620000-396bffff : reserved
>>>> 396c0000-3975ffff : System RAM
>>>> 39760000-3976ffff : reserved
>>>> 39770000-397affff : reserved
>>>> 397b0000-3989ffff : reserved
>>>> 398a0000-398bffff : reserved
>>>> 398c0000-39d3ffff : reserved
>>>> 39d40000-3ed2ffff : System RAM
>>>>
>>>   (snip)
>>>>
>>>> So, I am looking at what could be causing the 'Boot Code' and 'ACPI DSDT
>>>> table' ranges to be merged into a single region at
>>>> '0x0000396c0000-0x00003970ffff' which cannot be marked as RESERVED using
>>>> 'memblock_is_reserved'.
>>>
>>> Simple:) The short answer is that memblock_add() does.
>>>
>>> The long answer:
>>> First, please note that memblock maintains two type of regions list,
>>> "memory" and "reserved".
>>>
>>> efi_init()
>>>     reserve_regions()
>>>         early_init_dt_add_memory_arch()
>>>             memblock_add()
>>>                 memblock_add_range(memblock.memory)
>>>
>>> The memory regions described in efi.memmap are added to "memory" list
>>> with all the neighboring regions being merged into ones,
>>> in this case, "Runtime Code", "Boot Code", "ACPI Reclaim Memory" and others.
>>>
>>> The secret here is that "Runtime Code" is also marked with "NOMAP" flag in
>>> reserve_regions(), which creates an isolated region since it now has
>>> a different attribute.
>>> Consequently only "Boot Code" and "ACPI Reclaim Memory" are
>>> unified.
>>>
>>> Look at request_standard_resources(). It handles only "memory" list,
>>> and doesn't care about whether any arbitrary part of memory is in
>>> "reserved" list or not.
>>
>> Thanks for the pointers. Now I did some experiments and traversed the
>> whole memblock path and I see
>> how these two regions get merged into a single region which is later
>> on recognized by
>> 'request_standard_resources()' as a System RAM region rather than a
>> RESERVED region.
>>
>> I recently reproduced this on a APM mustang with latest kernel as well
>> when acpi is used to boot the machine, which makes me believe that
>> this is a generic issue for arm64 machines with the 4.14 kernel and if
>> they use acpi=force as the boot method.
>>
>> I am not sure, if a fix/or hack would be suitable for all underlying
>> arm64 machines, but I am trying one on the arm64 machines I have to
>> see if it fixes the issue.
>>
>> @Ard:
>>
>> Hi Ard,
>>
>> I think to create and test a clean solution for all arm64 boards it
>> will take some time, in the meantime should we consider reverting the
>> commit [1] to make sure that acpi enabled arm64 machines can boot with
>> 4.14?
>>
>> Please let me know your opinion.
>>
>> [1] f56ab9a5b73ca2aee777ccdf2d355ae2dd31db5a (efi/arm: Don't mark
>> ACPI reclaim memory as MEMBLOCK_NOMAP)
>>
>
> I don't think that is really going to help tbh.
>
> ACPI reclaim regions are not the only regions that are
> memblock_reserve()d and need to be reserved by the incoming kernel as
> well. So as far as I can tell, this is a symptom of an underlying
> issue that we will need to solve, and reverting the code that exposed
> it will not make the bug go away.
>

Looking deeper into the issue, since the arm64 kexec-tools uses the
'linux,usable-memory-range' dt property to allow crash dump kernel to
identify its own usable memory and exclude, at its boot time, any
other memory areas that are part of the panicked kernel's memory.
(see https://www.kernel.org/doc/Documentation/devicetree/bindings/chosen.txt
, for details)

1). Now when 'kexec -p' is executed, this node is patched up only
with the crashkernel memory range:

                /* add linux,usable-memory-range */
                nodeoffset = fdt_path_offset(new_buf, "/chosen");
                result = fdt_setprop_range(new_buf, nodeoffset,
                                PROP_USABLE_MEM_RANGE, &crash_reserved_mem,
                                address_cells, size_cells);

(see https://git.kernel.org/pub/scm/utils/kernel/kexec/kexec-tools.git/tree/kexec/arch/arm64/kexec-arm64.c#n465
, for details)

2). This excludes the ACPI reclaim regions irrespective of whether
they are marked as System RAM or as RESERVED. As,
'linux,usable-memory-range' dt node is patched up only with
'crash_reserved_mem' and not 'system_memory_ranges'

3). As a result when the crashkernel boots up it doesn't find this
ACPI memory and crashes while trying to access the same:

# kexec -p /boot/vmlinuz-`uname -r` --initrd=/boot/initramfs-`uname
-r`.img --reuse-cmdline -d

[snip..]

Reserved memory range
000000000e800000-000000002e7fffff (0)

Coredump memory ranges
0000000000000000-000000000e7fffff (0)
000000002e800000-000000003961ffff (0)
0000000039d40000-000000003ed2ffff (0)
000000003ed60000-000000003fbfffff (0)
0000001040000000-0000001ffbffffff (0)
0000002000000000-0000002ffbffffff (0)
0000009000000000-0000009ffbffffff (0)
000000a000000000-000000affbffffff (0)

4). So if we revert Ard's patch or just comment the fixing up of the
memory cap'ing passed to the crash kernel inside
'arch/arm64/mm/init.c' (see below):

static void __init fdt_enforce_memory_region(void)
{
        struct memblock_region reg = {
                .size = 0,
        };

        of_scan_flat_dt(early_init_dt_scan_usablemem, &reg);

        if (reg.size)
                //memblock_cap_memory_range(reg.base, reg.size); /*
comment this out */
}

5). Both the above temporary solutions fix the problem.

6). However exposing all System RAM regions to the crashkernel is not
advisable and may cause the crashkernel or some crashkernel drivers to
fail.

6a). I am trying an approach now, where the ACPI reclaim regions are
added to '/proc/iomem' separately as ACPI reclaim regions by the
kernel code and on the other hand the user-space 'kexec-tools' will
pick up the ACPI reclaim regions from '/proc/iomem' and add it to the
dt node 'linux,usable-memory-range'

6b). The kernel code currently looks like the following:

After this change the ACPI reclaim regions are properly recognized in
'/proc/iomem':

# cat /proc/iomem | grep -i ACPI
396c0000-3975ffff : ACPI reclaim region
39770000-397affff : ACPI reclaim region
398a0000-398bffff : ACPI reclaim region

6c). I am currently changing the 'kexec-tools' and will finish the
testing over the next few days.

I just wanted to know your opinion on this issue, so that I will be
able to propose a fix on the above lines.

Also Cc'ing kexec mailing list for more inputs on changes proposed to
kexec-tools.

Thanks,
Bhupesh

arm64 crashkernel fails to boot on acpi-only machines due to ACPI regions being no longer mapped as NOMAP

Commit Message

Comments

Patch