Re: [PATCH v2 00/25] mm: introduce numa_memblks

[Zi Yan <ziy@nvidia.com>]

[-- Attachment #1: Type: text/plain, Size: 6301 bytes --]

On 24 Jul 2024, at 20:35, Zi Yan wrote:

> On 24 Jul 2024, at 18:44, Zi Yan wrote:
>
>> On 23 Jul 2024, at 2:41, Mike Rapoport wrote:
>>
>>> From: "Mike Rapoport (Microsoft)" <rppt@kernel.org>
>>>
>>> Hi,
>>>
>>> Following the discussion about handling of CXL fixed memory windows on
>>> arm64 [1] I decided to bite the bullet and move numa_memblks from x86 to
>>> the generic code so they will be available on arm64/riscv and maybe on
>>> loongarch sometime later.
>>>
>>> While it could be possible to use memblock to describe CXL memory windows,
>>> it currently lacks notion of unpopulated memory ranges and numa_memblks
>>> does implement this.
>>>
>>> Another reason to make numa_memblks generic is that both arch_numa (arm64
>>> and riscv) and loongarch use trimmed copy of x86 code although there is no
>>> fundamental reason why the same code cannot be used on all these platforms.
>>> Having numa_memblks in mm/ will make it's interaction with ACPI and FDT
>>> more consistent and I believe will reduce maintenance burden.
>>>
>>> And with generic numa_memblks it is (almost) straightforward to enable NUMA
>>> emulation on arm64 and riscv.
>>>
>>> The first 9 commits in this series are cleanups that are not strictly
>>> related to numa_memblks.
>>> Commits 10-16 slightly reorder code in x86 to allow extracting numa_memblks
>>> and NUMA emulation to the generic code.
>>> Commits 17-19 actually move the code from arch/x86/ to mm/ and commits 20-22
>>> does some aftermath cleanups.
>>> Commit 23 switches arch_numa to numa_memblks.
>>> Commit 24 enables usage of phys_to_target_node() and
>>> memory_add_physaddr_to_nid() with numa_memblks.
>>> Commit 25 moves the description for numa=fake from x86 to admin-guide
>>>
>>> [1] https://lore.kernel.org/all/20240529171236.32002-1-Jonathan.Cameron@huawei.com/
>>>
>>> v1: https://lore.kernel.org/all/20240716111346.3676969-1-rppt@kernel.org
>>> * add cleanup for arch_alloc_nodedata and HAVE_ARCH_NODEDATA_EXTENSION
>>> * add patch that moves description of numa=fake kernel parameter from
>>>   x86 to admin-guide
>>> * reduce rounding up of node_data allocations from PAGE_SIZE to
>>>   SMP_CACHE_BYTES
>>> * restore single allocation attempt of numa_distance
>>> * fix several comments
>>> * added review tags
>>>
>>> Mike Rapoport (Microsoft) (25):
>>>   mm: move kernel/numa.c to mm/
>>>   MIPS: sgi-ip27: make NODE_DATA() the same as on all other architectures
>>>   MIPS: sgi-ip27: ensure node_possible_map only contains valid nodes
>>>   MIPS: sgi-ip27: drop HAVE_ARCH_NODEDATA_EXTENSION
>>>   MIPS: loongson64: rename __node_data to node_data
>>>   MIPS: loongson64: drop HAVE_ARCH_NODEDATA_EXTENSION
>>>   mm: drop CONFIG_HAVE_ARCH_NODEDATA_EXTENSION
>>>   arch, mm: move definition of node_data to generic code
>>>   arch, mm: pull out allocation of NODE_DATA to generic code
>>>   x86/numa: simplify numa_distance allocation
>>>   x86/numa: use get_pfn_range_for_nid to verify that node spans memory
>>>   x86/numa: move FAKE_NODE_* defines to numa_emu
>>>   x86/numa_emu: simplify allocation of phys_dist
>>>   x86/numa_emu: split __apicid_to_node update to a helper function
>>>   x86/numa_emu: use a helper function to get MAX_DMA32_PFN
>>>   x86/numa: numa_{add,remove}_cpu: make cpu parameter unsigned
>>>   mm: introduce numa_memblks
>>>   mm: move numa_distance and related code from x86 to numa_memblks
>>>   mm: introduce numa_emulation
>>>   mm: numa_memblks: introduce numa_memblks_init
>>>   mm: numa_memblks: make several functions and variables static
>>>   mm: numa_memblks: use memblock_{start,end}_of_DRAM() when sanitizing
>>>     meminfo
>>>   arch_numa: switch over to numa_memblks
>>>   mm: make range-to-target_node lookup facility a part of numa_memblks
>>>   docs: move numa=fake description to kernel-parameters.txt
>>>
>> Hi,
>>
>> I have tested this series on both x86_64 and arm64. It works fine on x86_64.
>> All numa=fake= options work as they did before the series.
>>
>> But I am not able to boot the kernel (no printout at all) on arm64 VM
>> (Mac mini M1 VMWare). By git bisecting, arch_numa: switch over to numa_memblks
>> is the first patch causing the boot failure. I see the warning:
>>
>> WARNING: modpost: vmlinux: section mismatch in reference: numa_add_cpu+0x1c (section: .text) -> early_cpu_to_node (section: .init.text)
>>
>> I am not sure if it is red herring or not, since changing early_cpu_to_node
>> to cpu_to_node in numa_add_cpu() from mm/numa_emulation.c did get rid of the
>> warning, but the system still failed to boot.
>>
>> Please note that you need binutils 2.40 to build the arm64 kernel, since there
>> is a bug(https://sourceware.org/bugzilla/show_bug.cgi?id=31924) in 2.42 preventing
>> arm64 kernel from booting as well.
>>
>> My config is attached.
>
> I get more info after adding earlycon to the boot option.
> pgdat is NULL, causing issues when free_area_init_node() is dereferencing
> it at first WARN_ON.
>
> FYI, my build is this series on top of v6.10 instead of the base commit,
> where the series applies cleanly on top v6.10.

OK, the issue comes from that my arm64 VM has no ACPI but x86_64 VM has it,
thus on arm64 VM numa_init(arch_acpi_numa_ini) failed in arch_numa_init()
and the code falls back to numa_init(dummy_numa_init). In dummy_numa_init(),
before patch 23 "arch_numa: switch over to numa_memblks", numa_add_memblk()
from drivers/base/arch_numa.c is called on arm64, which unconditionally
set 0 to numa_nodes_parsed. This is missing in the x86 version of
numa_add_memblk(), which is now used by all arch. By adding the patch
below, my arm64 kernel boots in the VM.


diff --git a/drivers/base/arch_numa.c b/drivers/base/arch_numa.c
index 806550239d08..354f15b8d9b7 100644
--- a/drivers/base/arch_numa.c
+++ b/drivers/base/arch_numa.c
@@ -279,6 +279,7 @@ static int __init dummy_numa_init(void)
                pr_err("NUMA init failed\n");
                return ret;
        }
+       node_set(0, numa_nodes_parsed);

        numa_off = true;
        return 0;


Feel free to add

Tested-by: Zi Yan <ziy@nvidia.com> # for x86_64 and arm64

after you incorporate the fix.


--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]