Re: [PATCH V6] NUMA: optimize detection of memory with no node id assigned by firmware

[Mike Rapoport]

On Sat, Oct 21, 2023 at 09:15:20PM +0800, Liam Ni wrote:
> Sanity check that makes sure the nodes cover all memory loops over
> numa_meminfo to count the pages that have node id assigned by the firmware,
> then loops again over memblock.memory to find the total amount of memory
> and in the end checks that the difference between the total memory and
> memory that covered by nodes is less than some threshold. Worse, the loop
> over numa_meminfo calls __absent_pages_in_range() that also partially
> traverses memblock.memory.
> 
> It's much simpler and more efficient to have a single traversal of
> memblock.memory that verifies that amount of memory not covered by nodes is
> less than a threshold.
> 
> Introduce memblock_validate_numa_coverage() that does exactly that and use
> it instead of numa_meminfo_cover_memory().
> 
> Signed-off-by: Liam Ni <zhiguangni01@xxxxxxxxx>
> ---
>  arch/loongarch/kernel/numa.c | 28 +---------------------------
>  arch/x86/mm/numa.c           | 34 ++--------------------------------
>  include/linux/memblock.h     |  1 +
>  mm/memblock.c                | 34 ++++++++++++++++++++++++++++++++++
>  4 files changed, 38 insertions(+), 59 deletions(-)
> 
> diff --git a/arch/loongarch/kernel/numa.c b/arch/loongarch/kernel/numa.c
> index cb00804826f7..84e1b046ab52 100644
> --- a/arch/loongarch/kernel/numa.c
> +++ b/arch/loongarch/kernel/numa.c
> @@ -226,32 +226,6 @@ static void __init node_mem_init(unsigned int node)
> 
>  #ifdef CONFIG_ACPI_NUMA
> 
> -/*
> - * Sanity check to catch more bad NUMA configurations (they are amazingly
> - * common).  Make sure the nodes cover all memory.
> - */
> -static bool __init numa_meminfo_cover_memory(const struct numa_meminfo *mi)
> -{
> -   int i;
> -   u64 numaram, biosram;
> -
> -   numaram = 0;
> -   for (i = 0; i < mi->nr_blks; i++) {
> -       u64 s = mi->blk[i].start >> PAGE_SHIFT;
> -       u64 e = mi->blk[i].end >> PAGE_SHIFT;
> -
> -       numaram += e - s;
> -       numaram -= __absent_pages_in_range(mi->blk[i].nid, s, e);
> -       if ((s64)numaram < 0)
> -           numaram = 0;
> -   }
> -   max_pfn = max_low_pfn;
> -   biosram = max_pfn - absent_pages_in_range(0, max_pfn);
> -
> -   BUG_ON((s64)(biosram - numaram) >= (1 << (20 - PAGE_SHIFT)));
> -   return true;
> -}
> -
>  static void __init add_node_intersection(u32 node, u64 start, u64
> size, u32 type)
>  {
>     static unsigned long num_physpages;
> @@ -396,7 +370,7 @@ int __init init_numa_memory(void)
>         return -EINVAL;
> 
>     init_node_memblock();
> -   if (numa_meminfo_cover_memory(&numa_meminfo) == false)
> +   if (!memblock_validate_numa_coverage(SZ_1M >> PAGE_SIZE))
>         return -EINVAL;
> 
>     for_each_node_mask(node, node_possible_map) {
> diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
> index 2aadb2019b4f..39772d551924 100644
> --- a/arch/x86/mm/numa.c
> +++ b/arch/x86/mm/numa.c
> @@ -447,37 +447,6 @@ int __node_distance(int from, int to)
>  }
>  EXPORT_SYMBOL(__node_distance);
> 
> -/*
> - * Sanity check to catch more bad NUMA configurations (they are amazingly
> - * common).  Make sure the nodes cover all memory.
> - */
> -static bool __init numa_meminfo_cover_memory(const struct numa_meminfo *mi)
> -{
> -   u64 numaram, e820ram;
> -   int i;
> -
> -   numaram = 0;
> -   for (i = 0; i < mi->nr_blks; i++) {
> -       u64 s = mi->blk[i].start >> PAGE_SHIFT;
> -       u64 e = mi->blk[i].end >> PAGE_SHIFT;
> -       numaram += e - s;
> -       numaram -= __absent_pages_in_range(mi->blk[i].nid, s, e);
> -       if ((s64)numaram < 0)
> -           numaram = 0;
> -   }
> -
> -   e820ram = max_pfn - absent_pages_in_range(0, max_pfn);
> -
> -   /* We seem to lose 3 pages somewhere. Allow 1M of slack. */
> -   if ((s64)(e820ram - numaram) >= (1 << (20 - PAGE_SHIFT))) {
> -       printk(KERN_ERR "NUMA: nodes only cover %LuMB of your %LuMB
> e820 RAM. Not used.\n",
> -              (numaram << PAGE_SHIFT) >> 20,
> -              (e820ram << PAGE_SHIFT) >> 20);
> -       return false;
> -   }
> -   return true;
> -}
> -
>  /*
>   * Mark all currently memblock-reserved physical memory (which covers the
>   * kernel's own memory ranges) as hot-unswappable.
> @@ -583,7 +552,8 @@ static int __init numa_register_memblks(struct
> numa_meminfo *mi)
>             return -EINVAL;
>         }
>     }
> -   if (!numa_meminfo_cover_memory(mi))
> +
> +   if (!memblock_validate_numa_coverage(SZ_1M >> PAGE_SIZE))
>         return -EINVAL;
> 
>     /* Finally register nodes. */
> diff --git a/include/linux/memblock.h b/include/linux/memblock.h
> index 1c1072e3ca06..727242f4b54a 100644
> --- a/include/linux/memblock.h
> +++ b/include/linux/memblock.h
> @@ -120,6 +120,7 @@ int memblock_physmem_add(phys_addr_t base,
> phys_addr_t size);
>  void memblock_trim_memory(phys_addr_t align);
>  bool memblock_overlaps_region(struct memblock_type *type,
>                   phys_addr_t base, phys_addr_t size);
> +bool memblock_validate_numa_coverage(const u64 threshold_pages);
>  int memblock_mark_hotplug(phys_addr_t base, phys_addr_t size);
>  int memblock_clear_hotplug(phys_addr_t base, phys_addr_t size);
>  int memblock_mark_mirror(phys_addr_t base, phys_addr_t size);
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 0863222af4a4..ebea0c204eaa 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -734,6 +734,40 @@ int __init_memblock memblock_add(phys_addr_t
> base, phys_addr_t size)
>     return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
>  }
> 
> +/**
> + * memblock_validate_numa_coverage - check if amount of memory with
> + * no node ID assigned is less than a threshold
> + * @threshold_pages: maximal number of pages that can have unassigned node
> + * ID (in pages).
> + *
> + * A buggy firmware may report memory that does not belong to any node.
> + * Check if amount of such memory is below @threshold_pages.
> + *
> + * Return: true on success, false on failure.
> + */
> +bool __init_memblock memblock_validate_numa_coverage(const u64 threshold_pages)

With threshold in bytes there will be less conversions from pages to bytes.
Also, I don't think const has much value here and unsigned long should
suffice.

> +{
> +   unsigned long nr_pages = 0;
> +   unsigned long start_pfn, end_pfn, mem_size_mb;
> +   int nid, i;
> +
> +   /* calculate lose page */
> +   for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
> +       if (nid == NUMA_NO_NODE)
> +           nr_pages += end_pfn - start_pfn;
> +   }
> +
> +   if (nr_pages >= threshold_pages) {
> +       mem_size_mb = memblock_phys_mem_size() >> 20;
> +       pr_err("NUMA: no nodes coverage for %luMB of %luMB RAM\n",
> +              (nr_pages << PAGE_SHIFT) >> 20, mem_size_mb);
> +       return false;
> +   }
> +
> +   return true;
> +}
> +
> +
>  /**
>   * memblock_isolate_range - isolate given range into disjoint memblocks
>   * @type: memblock type to isolate range for
> -- 
> 2.25.1

-- 
Sincerely yours,
Mike.