Re: [PATCH] mm/hugetlb: two-phase hugepage allocation when reservation is high

[David Hildenbrand <david@redhat.com>]

On 27.08.25 06:12, Li,Rongqing wrote:
> 
> .
>>
>> Also, can't we fail lightly during the first attempt and dynamically decide if we
>> should do a second pase?
>>
> 
> 
> Good idea, like below
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 753f99b..425a759 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -3589,6 +3589,7 @@ static unsigned long __init hugetlb_pages_alloc_boot(struct hstate *h)
> 
>          unsigned long jiffies_start;
>          unsigned long jiffies_end;
> +       unsigned long remaining;
> 
>          job.thread_fn   = hugetlb_pages_alloc_boot_node;
>          job.start       = 0;
> @@ -3620,6 +3621,18 @@ static unsigned long __init hugetlb_pages_alloc_boot(struct hstate *h)
> 
>          jiffies_start = jiffies;
>          padata_do_multithreaded(&job);
> +
> +       if (h->nr_huge_pages != h->max_huge_pages && hugetlb_vmemmap_optimizable(h)) {
> +               remaining = h->max_huge_pages - h->nr_huge_pages;
> +               /* vmemmap optimization can save about 1.6% (4/250) memory */
> +               remaining = min(remaining, (h->nr_huge_pages * 4 / 250));

I don't like hard coding that here.

> +
> +               job.start       = h->nr_huge_pages;
> +               job.size        = remaining;
> +               job.min_chunk   = remaining / hugepage_allocation_threads;
> +               padata_do_multithreaded(&job);
> +       }

Thinking out load, can't we try in a loop until either

a) We allocated all we need

b) We don't make any more progress


Not sure if something like the following could fly:

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 1f42186a85ea4..dfb4d717b8a02 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3595,8 +3595,6 @@ static unsigned long __init hugetlb_pages_alloc_boot(struct hstate *h)
         unsigned long jiffies_end;
  
         job.thread_fn   = hugetlb_pages_alloc_boot_node;
-       job.start       = 0;
-       job.size        = h->max_huge_pages;
  
         /*
          * job.max_threads is 25% of the available cpu threads by default.
@@ -3620,10 +3618,24 @@ static unsigned long __init hugetlb_pages_alloc_boot(struct hstate *h)
         }
  
         job.max_threads = hugepage_allocation_threads;
-       job.min_chunk   = h->max_huge_pages / hugepage_allocation_threads;
  
         jiffies_start = jiffies;
-       padata_do_multithreaded(&job);
+       /* TODO: comment why we retry and how it interacts with vmemmap op. */
+       while (h->nr_huge_pages != h->max_huge_pages) {
+               unsigned long remaining = h->max_huge_pages - h->nr_huge_pages;
+
+               job.start       = h->nr_huge_pages;
+               job.size        = remaining;
+               job.min_chunk   = remaining / hugepage_allocation_threads;
+               padata_do_multithreaded(&job);
+
+               if (hugetlb_vmemmap_optimizable(h))
+                       break;
+
+               /* Stop if there is no progress. */
+               if (remaining == h->max_huge_pages - h->nr_huge_pages)
+                       break;
+       }
         jiffies_end = jiffies;
  
         pr_info("HugeTLB: allocation took %dms with hugepage_allocation_threads=%ld\n",


-- 
Cheers

David / dhildenb