Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <a684b0e8-4412-79c8-2926-7822b1163c15@bytedance.com>
Date:   Fri, 28 Jul 2023 14:47:08 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.4.2
Subject: Re: [External] Re: [v1 4/6] memblock: introduce MEMBLOCK_RSRV_NOINIT
 flag
Content-Language: en-US
To:     =?UTF-8?Q?Mika_Penttil=c3=a4?= <mpenttil@redhat.com>,
        linux-mm@kvack.org, muchun.song@linux.dev, mike.kravetz@oracle.com,
        rppt@kernel.org
Cc:     linux-kernel@vger.kernel.org, fam.zheng@bytedance.com,
        liangma@liangbit.com, simon.evans@bytedance.com,
        punit.agrawal@bytedance.com
References: <20230727204624.1942372-1-usama.arif@bytedance.com>
 <20230727204624.1942372-5-usama.arif@bytedance.com>
 <55750855-0029-b10f-3317-e6ae4d89d492@redhat.com>
From:   Usama Arif <usama.arif@bytedance.com>
In-Reply-To: <55750855-0029-b10f-3317-e6ae4d89d492@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 28/07/2023 05:30, Mika Penttilä wrote:
> Hi,
> 
> On 7/27/23 23:46, Usama Arif wrote:
> 
>> For reserved memory regions marked with this flag,
>> reserve_bootmem_region is not called during memmap_init_reserved_pages.
>> This can be used to avoid struct page initialization for
>> regions which won't need them, for e.g. hugepages with
>> HVO enabled.
>>
>> Signed-off-by: Usama Arif <usama.arif@bytedance.com>
>> ---
>>   include/linux/memblock.h |  7 +++++++
>>   mm/memblock.c            | 32 ++++++++++++++++++++++++++------
>>   2 files changed, 33 insertions(+), 6 deletions(-)
>>
>> diff --git a/include/linux/memblock.h b/include/linux/memblock.h
>> index f71ff9f0ec81..7f9d06c08592 100644
>> --- a/include/linux/memblock.h
>> +++ b/include/linux/memblock.h
>> @@ -47,6 +47,7 @@ enum memblock_flags {
>>       MEMBLOCK_MIRROR        = 0x2,    /* mirrored region */
>>       MEMBLOCK_NOMAP        = 0x4,    /* don't add to kernel direct 
>> mapping */
>>       MEMBLOCK_DRIVER_MANAGED = 0x8,    /* always detected via a 
>> driver */
>> +    MEMBLOCK_RSRV_NOINIT    = 0x10,    /* don't call 
>> reserve_bootmem_region for this region */
>>   };
>>   /**
>> @@ -125,6 +126,7 @@ int memblock_clear_hotplug(phys_addr_t base, 
>> phys_addr_t size);
>>   int memblock_mark_mirror(phys_addr_t base, phys_addr_t size);
>>   int memblock_mark_nomap(phys_addr_t base, phys_addr_t size);
>>   int memblock_clear_nomap(phys_addr_t base, phys_addr_t size);
>> +int memblock_rsrv_mark_noinit(phys_addr_t base, phys_addr_t size);
>>   void memblock_free_all(void);
>>   void memblock_free(void *ptr, size_t size);
>> @@ -259,6 +261,11 @@ static inline bool memblock_is_nomap(struct 
>> memblock_region *m)
>>       return m->flags & MEMBLOCK_NOMAP;
>>   }
>> +static inline bool memblock_is_noinit(struct memblock_region *m)
>> +{
>> +    return m->flags & MEMBLOCK_RSRV_NOINIT;
>> +}
>> +
>>   static inline bool memblock_is_driver_managed(struct memblock_region 
>> *m)
>>   {
>>       return m->flags & MEMBLOCK_DRIVER_MANAGED;
>> diff --git a/mm/memblock.c b/mm/memblock.c
>> index 4fd431d16ef2..3a15708af3b6 100644
>> --- a/mm/memblock.c
>> +++ b/mm/memblock.c
>> @@ -997,6 +997,22 @@ int __init_memblock 
>> memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
>>       return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP, 0);
>>   }
>> +/**
>> + * memblock_rsrv_mark_noinit - Mark a reserved memory region with 
>> flag MEMBLOCK_RSRV_NOINIT.
>> + * @base: the base phys addr of the region
>> + * @size: the size of the region
>> + *
>> + * For memory regions marked with %MEMBLOCK_RSRV_NOINIT, 
>> reserve_bootmem_region
>> + * is not called during memmap_init_reserved_pages, hence struct 
>> pages are not
>> + * initialized for this region.
>> + *
>> + * Return: 0 on success, -errno on failure.
>> + */
>> +int __init_memblock memblock_rsrv_mark_noinit(phys_addr_t base, 
>> phys_addr_t size)
>> +{
>> +    return memblock_setclr_flag(base, size, 1, MEMBLOCK_RSRV_NOINIT, 1);
>> +}
>> +
>>   static bool should_skip_region(struct memblock_type *type,
>>                      struct memblock_region *m,
>>                      int nid, int flags)
>> @@ -2113,13 +2129,17 @@ static void __init 
>> memmap_init_reserved_pages(void)
>>           memblock_set_node(start, end, &memblock.reserved, nid);
>>       }
>> -    /* initialize struct pages for the reserved regions */
>> +    /*
>> +     * initialize struct pages for reserved regions that don't have
>> +     * the MEMBLOCK_RSRV_NOINIT flag set
>> +     */
>>       for_each_reserved_mem_region(region) {
>> -        nid = memblock_get_region_node(region);
>> -        start = region->base;
>> -        end = start + region->size;
>> -
>> -        reserve_bootmem_region(start, end, nid);
>> +        if (!memblock_is_noinit(region)) {
>> +            nid = memblock_get_region_node(region);
>> +            start = region->base;
>> +            end = start + region->size;
>> +            reserve_bootmem_region(start, end, nid);
>> +        }
>>       }
>>   }
> 
> There's code like:
> 
> static inline void free_vmemmap_page(struct page *page)
> {
>          if (PageReserved(page))
>                  free_bootmem_page(page);
>          else
>                  __free_page(page);
> }
> 
> which depends on the PageReserved being in vmempages pages, so I think 
> you can't skip that part?
> 

free_vmemmap_page_list (free_vmemmap_page) is called on struct pages 
(refer to as [1]) that point to memory *which contains* the struct pages 
(refer to as [2]) for the hugepage. The above if 
(!memblock_is_noinit(region)) to not reserve_bootmem_region is called 
for the struct pages [2] for the hugepage. struct pages [1] are not 
changed with my patch.

As an experiment if I run the diff at the bottom with and without these 
patches I get the same log "HugeTLB: reserved pages 4096, normal pages 
0", which means those struct pages are treated the same without and 
without these patches. (Its 4096 as 262144 struct pages [2] per hugepage 
* 64 bytes per struct page / PAGE_SIZE = 4096 struct pages [1] )

Also should have mentioned in cover letter, I used cat /proc/meminfo to 
make sure it was working as expected. Reserving 500 1G hugepages with 
and without these patches when hugetlb_free_vmemmap=on
MemTotal:       536207112 kB (511.4G)

when hugetlb_free_vmemmap=off
MemTotal:       528015112 kB (503G)


The expectation is that for 500 1G hugepages, when using HVO we have a 
saving of 16380K*500=~8GB which is what we see with and without those 
patches (511.4G - 503G). These patches didnt affect these numbers.


diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index b5b7834e0f42..bc0ec90552b7 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -208,6 +208,8 @@ static int vmemmap_remap_range(unsigned long start, 
unsigned long end,
         return 0;
  }

+static int i = 0, j = 0;
+
  /*
   * Free a vmemmap page. A vmemmap page can be allocated from the memblock
   * allocator or buddy allocator. If the PG_reserved flag is set, it means
@@ -216,10 +218,14 @@ static int vmemmap_remap_range(unsigned long 
start, unsigned long end,
   */
  static inline void free_vmemmap_page(struct page *page)
  {
-       if (PageReserved(page))
+       if (PageReserved(page)) {
+               i++;
                 free_bootmem_page(page);
-       else
+       }
+       else {
+               j++;
                 __free_page(page);
+       }
  }

  /* Free a list of the vmemmap pages */
@@ -380,6 +386,7 @@ static int vmemmap_remap_free(unsigned long start, 
unsigned long end,

         free_vmemmap_page_list(&vmemmap_pages);

+       pr_err("reserved pages %u, normal pages %u", i, j);
         return ret;
  }


> --Mika
> 
>