lkml.org 
[lkml]   [2009]   [Sep]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH 2/3] Helper which returns the huge page at a given address (Take 3)
From
On Wed, Sep 16, 2009 at 12:26 AM, Mel Gorman <mel@csn.ul.ie> wrote:
> On Mon, Sep 14, 2009 at 05:18:53PM +1200, Alexey Korolev wrote:
>> This patch provides helper function which returns the huge page at a
>> given address for population before the page has been faulted.
>> It is possible to call hugetlb_get_user_page function in file mmap
>> procedure to get pages before they have been requested by user level.
>>
>
> Worth spelling out that this is similar in principal to get_user_pages()
> but not as painful to use in this specific context.
>

Right. I'll do this. Seems it is important to clearly mention that
this function do not introduce new functionality.

>> include/linux/hugetlb.h |    3 +++
>> mm/hugetlb.c            |   23 +++++++++++++++++++++++
>> 2 files changed, 26 insertions(+)
>>
>> ---
>> Signed-off-by: Alexey Korolev <akorolev@infradead.org>
>
> Patch formatting nit.
>
> diffstat goes below the --- and signed-off-bys go above it.
>
Right. To be fixed.

>>
>> +/*
>> + * hugetlb_get_user_page returns the page at a given address for population
>> + * before the page has been faulted.
>> + */
>> +struct page *hugetlb_get_user_page(struct vm_area_struct *vma,
>> +                                 unsigned long address)
>> +{
>
> Your leader and comments say that the function can be used before the pages
> have been faulted. It would presumably require that this function be called
> from within a mmap() handler.
>
> What is happening because you call follow_hugetlb_page() is that the pages
> get faulted as part of your mmap() operation. This might make the overall
> operation more expensive than you expected. I don't know if what you really
> intended was to allocate the huge page, insert it into the page cache and
> have it faulted later if the process actually references the page.
>
> Similarly the leader and comments imply that you expect this to be
> called as part of the mmap() operation. However, nothing would appear to
> prevent the driver calling this function once the page is already
> faulted. Is this intentional?

The implication was not intende. You are correct, the function can be
called later. The leader and comment can be rewritten to make this
clear.

>> +     int ret;
>> +     int cnt = 1;
>> +     struct page *pg;
>> +     struct hstate *h = hstate_vma(vma);
>> +
>> +     address = address & huge_page_mask(h);
>> +     ret = follow_hugetlb_page(vma->vm_mm, vma, &pg,
>> +                             NULL, &address, &cnt, 0, 0);
>> +     if (ret < 0)
>> +             return ERR_PTR(ret);
>> +     put_page(pg);
>> +
>> +     return pg;
>> +}
>
> I think the caller should be responsible for calling put_page().  Otherwise
> there is an outside chance that the page would disappear from you unexpectedly
> depending on exactly how the driver was implemented. It would also
> behave slightly more like get_user_pages().
>
Correct. Lets have behaviour similar to get_user_pages in order to prevent
misunderstanding. Put_page will be removed.

Thank you very much for review. Now I am about to clear out the
mistakes and will pay a lot more attention to patch descriptions and
comments.

Thanks,
Alexey
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2009-09-16 14:23    [W:0.055 / U:0.276 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site