linux-kernel - Re: [PATCH] mm: sparse: pass section_nr to section_mark

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <4550295f-0358-9c41-5655-7274f89f6c0a@redhat.com>
Date:   Thu, 1 Jul 2021 18:12:20 +0200
From:   David Hildenbrand <david@...hat.com>
To:     ohoono.kwon@...sung.com,
        "akpm@...ux-foundation.org" <akpm@...ux-foundation.org>,
        "mhocko@...e.com" <mhocko@...e.com>
Cc:     "bhe@...hat.com" <bhe@...hat.com>,
        "rppt@...ux.ibm.com" <rppt@...ux.ibm.com>,
        "ohkwon1043@...il.com" <ohkwon1043@...il.com>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mm: sparse: pass section_nr to section_mark_present

On 01.07.21 17:41, 권오훈 wrote:
> On Thu, Jul 01, 2021 at 04:34:13PM +0200, David Hildenbrand wrote:
>> On 01.07.21 15:55, 권오훈 wrote:
>>> With CONFIG_SPARSEMEM_EXTREME enabled, __section_nr() which converts
>>> mem_section to section_nr could be costly since it iterates all
>>> sections to check if the given mem_section is in its range.
>>   
>> It actually iterates all section roots.
>>   
>>>
>>> On the other hand, __nr_to_section which converts section_nr to
>>> mem_section can be done in O(1).
>>>
>>> Let's pass section_nr instead of mem_section ptr to section_mark_present
>>> in order to reduce needless iterations.
>>   
>> I'd expect this to be mostly noise, especially as we iterate section
>> roots and for most (smallish) machines we might just work on the lowest
>> section roots only.
>>   
>> Can you actually observe an improvement regarding boot times?
>>   
>> Anyhow, looks straight forward to me, although we might just reintroduce
>> similar patterns again easily if it's really just noise (see
>> find_memory_block() as used by). And it might allow for a nice cleanup
>> (see below).
>>   
>> Reviewed-by: David Hildenbrand <david@...hat.com>
>>   
>>   
>> Can you send 1) a patch to convert find_memory_block() as well and 2) a
>> patch to rip out __section_nr() completely?
>>   
>>>
>>> Signed-off-by: Ohhoon Kwon <ohoono.kwon@...sung.com>
>>> ---
>>>    mm/sparse.c | 9 +++++----
>>>    1 file changed, 5 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/mm/sparse.c b/mm/sparse.c
>>> index 55c18aff3e42..4a2700e9a65f 100644
>>> --- a/mm/sparse.c
>>> +++ b/mm/sparse.c
>>> @@ -186,13 +186,14 @@ void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
>>>     * those loops early.
>>>     */
>>>    unsigned long __highest_present_section_nr;
>>> -static void section_mark_present(struct mem_section *ms)
>>> +static void section_mark_present(unsigned long section_nr)
>>>    {
>>> -        unsigned long section_nr = __section_nr(ms);
>>> +        struct mem_section *ms;
>>>    
>>>            if (section_nr > __highest_present_section_nr)
>>>                    __highest_present_section_nr = section_nr;
>>>    
>>> +        ms = __nr_to_section(section_nr);
>>>            ms->section_mem_map |= SECTION_MARKED_PRESENT;
>>>    }
>>>    
>>> @@ -279,7 +280,7 @@ static void __init memory_present(int nid, unsigned long start, unsigned long en
>>>                    if (!ms->section_mem_map) {
>>>                            ms->section_mem_map = sparse_encode_early_nid(nid) |
>>>                                                            SECTION_IS_ONLINE;
>>> -                        section_mark_present(ms);
>>> +                        section_mark_present(section);
>>>                    }
>>>            }
>>>    }
>>> @@ -933,7 +934,7 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
>>>    
>>>            ms = __nr_to_section(section_nr);
>>>            set_section_nid(section_nr, nid);
>>> -        section_mark_present(ms);
>>> +        section_mark_present(section_nr);
>>>    
>>>            /* Align memmap to section boundary in the subsection case */
>>>            if (section_nr_to_pfn(section_nr) != start_pfn)
>>>
>>   
>>   
>> -- 
>> Thanks,
>>   
>> David / dhildenb
>>   
> Dear David.
> 
> I tried to check on time for memblocks_present, but when I tested with mobile
> phones with 8GB ram, the original binary took 0us either as well as the
> patched binary.
> I'm not sure how the results would differ on huge systems with bigger ram.
> I agree that it could turn out to be just a noise, as you expected.
> 
> However as you also mentioned, the patches will be straight forward when all
> codes using __section_nr() are cleaned up nicely.
> 
> Below are the two patches that you asked for.
> Please tell me if you need me to send the patches in separate e-mails.

Yes, please send them separately. Maybe sent all 3 patches combined in a 
single series, so Andrew can pick them easily, and reviewers can review 
more easily.

Thanks!

-- 
Thanks,

David / dhildenb