LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Faiyaz Mohammed <faiyazm@codeaurora.org>
To: David Hildenbrand <david@redhat.com>,
	akpm@linux-foundation.org, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org
Cc: guptap@codeaurora.org
Subject: Re: [PATCH v1] mm: page_alloc: Add debug log in free_reserved_area for static memory
Date: Mon, 11 Oct 2021 12:19:58 +0530	[thread overview]
Message-ID: <c2f9cbf5-ff81-f36e-68f9-ba3110b1ddd3@codeaurora.org> (raw)
In-Reply-To: <aee9523c-4930-7980-e498-00f671b7d336@redhat.com>



On 10/6/2021 5:48 PM, David Hildenbrand wrote:
> On 06.10.21 14:13, Faiyaz Mohammed wrote:
>> Hi,
>>
>> Sorry for delayed response.
>>
>> On 9/29/2021 10:33 PM, David Hildenbrand wrote:
>>> On 29.09.21 10:58, Faiyaz Mohammed wrote:
>>>>
>>>>
>>>> On 9/28/2021 4:46 PM, David Hildenbrand wrote:
>>>>> On 28.09.21 12:53, Faiyaz Mohammed wrote:
>>>>>>
>>>>>>
>>>>>> On 9/28/2021 4:09 PM, David Hildenbrand wrote:
>>>>>>> On 28.09.21 11:04, Faiyaz Mohammed wrote:
>>>>>>>> For INITRD and initmem memory is reserved through
>>>>>>>> "memblock_reserve"
>>>>>>>> during boot up but it is free via "free_reserved_area" instead
>>>>>>>> of "memblock_free".
>>>>>>>> For example:
>>>>>>>> [    0.294848] Freeing initrd memory: 12K.
>>>>>>>> [    0.696688] Freeing unused kernel memory: 4096K.
>>>>>>>>
>>>>>>>> To get the start and end address of the above freed memory and to
>>>>>>>> account
>>>>>>>> proper memblock added memblock_dbg log in "free_reserved_area".
>>>>>>>> After adding log:
>>>>>>>> [    0.294837] memblock_free: [0x00000083600000-0x00000083603000]
>>>>>>>> free_initrd_mem+0x20/0x28
>>>>>>>> [    0.294848] Freeing initrd memory: 12K.
>>>>>>>> [    0.695246] memblock_free: [0x00000081600000-0x00000081a00000]
>>>>>>>> free_initmem+0x70/0xc8
>>>>>>>> [    0.696688] Freeing unused kernel memory: 4096K.
>>>>>>>>
>>>>>>>> Signed-off-by: Faiyaz Mohammed <faiyazm@codeaurora.org>
>>>>>>>> ---
>>>>>>>>      mm/page_alloc.c | 5 +++++
>>>>>>>>      1 file changed, 5 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>>>>>>> index b37435c..f85c3b2 100644
>>>>>>>> --- a/mm/page_alloc.c
>>>>>>>> +++ b/mm/page_alloc.c
>>>>>>>> @@ -8129,6 +8129,11 @@ unsigned long free_reserved_area(void
>>>>>>>> *start,
>>>>>>>> void *end, int poison, const char
>>>>>>>>              pr_info("Freeing %s memory: %ldK\n",
>>>>>>>>                  s, pages << (PAGE_SHIFT - 10));
>>>>>>>>      +#ifdef CONFIG_HAVE_MEMBLOCK
>>>>>>>> +        memblock_dbg("memblock_free: [%#016llx-%#016llx] %pS\n",
>>>>>>>> +            __pa(start), __pa(end), (void *)_RET_IP_);
>>>>>>>> +#endif
>>>>>>>
>>>>>>> IMHO, the "memblock_free" part is misleading. Something was
>>>>>>> allocated
>>>>>>> early via memblock, then we transitioned to the buddy, now we're
>>>>>>> freeing
>>>>>>> that early allocation via the buddy.
>>>>>>> Yes, we're freeing the early allocation via buddy, but for proper
>>>>>> memblock accounting we need this debug print.
>>>>>>
>>>>>
>>>>> What do you mean with "accounting" ? These are debug statements.
>>>>>
>>>>>
>>>> Yes, these are debug statements, which help to know the a-b address
>>>> belongs to x callsite. This info is required when memblock=debug is
>>>> passed through command line and CONFIG_HAVE_MEMBLOCK is enabled.
>>>
>>> The issue I'm having is talking in the name of memblock "memblock_dbg,
>>> memblock_free", when memblock might no longer be around. We have other
>>> places where we free early memblock allocations back to the buddy.
>> I didn't find place where we free early memblock allocation back to the
>> buddy.
> 
> One example I know is
> 
> section_deactivate()->free_map_bootmem()->vmemmap_free()-> ...
> free_pagetable()->free_reserved_page().
> 
> when we free the vmemmap allocated via memblock back to the buddy.
> 
>>
>> Why "memblock_dbg" print with "memblock_free" string?.
>> - After buddy took over, buddy will free memblock reserved memory
>> through free_reserved_area and it will print the freed memory size, but
>> the freed memory through buddy still be part of
>> memblock.reserved.regions.
>> - To know the address ranges, added the "memblock_dbg" print along with
>> "membloc_free" string.
>> - If it is misleading or confusing, we can remove the "memblock_free"
>> string from the "memblock_dgb" print and we can just print the address
>> range when "memlock=debug" pass through command line.
> 
> That would be better, but do we really have to depend on
> "memlock=debug"? Can't we do pr_debug() ?
> 
Yes, I think we can do pr_debug. I will update the patch and push again.

Thanks and regards,
Mohammed Faiyaz

      parent reply	other threads:[~2021-10-11  6:50 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-28  9:04 Faiyaz Mohammed
2021-09-28 10:39 ` David Hildenbrand
2021-09-28 10:53   ` Faiyaz Mohammed
2021-09-28 11:16     ` David Hildenbrand
2021-09-29  8:58       ` Faiyaz Mohammed
2021-09-29 17:03         ` David Hildenbrand
2021-10-06 12:13           ` Faiyaz Mohammed
2021-10-06 12:18             ` David Hildenbrand
2021-10-06 14:22               ` Faiyaz Mohammed
2021-10-11  6:49               ` Faiyaz Mohammed [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=c2f9cbf5-ff81-f36e-68f9-ba3110b1ddd3@codeaurora.org \
    --to=faiyazm@codeaurora.org \
    --cc=akpm@linux-foundation.org \
    --cc=david@redhat.com \
    --cc=guptap@codeaurora.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --subject='Re: [PATCH v1] mm: page_alloc: Add debug log in free_reserved_area for static memory' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).