From: Qu Wenruo <quwenruo.btrfs@gmx.com>
To: dsterba@suse.cz, Qu Wenruo <wqu@suse.com>, linux-btrfs@vger.kernel.org
Subject: Re: [PATCH v2 04/19] btrfs: remove the open-code to read disk-key
Date: Thu, 17 Sep 2020 21:15:31 +0800 [thread overview]
Message-ID: <b6f59a13-0572-01dc-656f-09f1b5eb7935@gmx.com> (raw)
In-Reply-To: <20200917123738.GR1791@twin.jikos.cz>
[-- Attachment #1.1: Type: text/plain, Size: 3719 bytes --]
On 2020/9/17 下午8:37, David Sterba wrote:
> On Thu, Sep 17, 2020 at 04:02:37PM +0800, Qu Wenruo wrote:
>> On 2020/9/17 上午12:01, David Sterba wrote:
>>> On Tue, Sep 15, 2020 at 01:35:17PM +0800, Qu Wenruo wrote:
>>>> generic_bin_search() distinguishes between reading a key which doesn't
>>>> cross a page and one which does. However this distinction is not
>>>> necessary since read_extent_buffer handles both cases transparently.
>>>>
>>>> Just use read_extent_buffer to streamline the code.
>>>>
>>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>>> ---
>>>> fs/btrfs/ctree.c | 13 ++-----------
>>>> 1 file changed, 2 insertions(+), 11 deletions(-)
>>>>
>>>> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
>>>> index cd1cd673bc0b..e204e1320745 100644
>>>> --- a/fs/btrfs/ctree.c
>>>> +++ b/fs/btrfs/ctree.c
>>>> @@ -1697,7 +1697,6 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
>>>> }
>>>>
>>>> while (low < high) {
>>>> - unsigned long oip;
>>>> unsigned long offset;
>>>> struct btrfs_disk_key *tmp;
>>>> struct btrfs_disk_key unaligned;
>>>> @@ -1705,17 +1704,9 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
>>>>
>>>> mid = (low + high) / 2;
>>>> offset = p + mid * item_size;
>>>> - oip = offset_in_page(offset);
>>>>
>>>> - if (oip + key_size <= PAGE_SIZE) {
>>>> - const unsigned long idx = offset >> PAGE_SHIFT;
>>>> - char *kaddr = page_address(eb->pages[idx]);
>>>> -
>>>> - tmp = (struct btrfs_disk_key *)(kaddr + oip);
>>>> - } else {
>>>> - read_extent_buffer(eb, &unaligned, offset, key_size);
>>>> - tmp = &unaligned;
>>>> - }
>>>> + read_extent_buffer(eb, &unaligned, offset, key_size);
>>>> + tmp = &unaligned;
>>>
>>> Reading from the first page is a performance optimization on systems
>>> with 4K pages, ie. the majority. I'm not in favor removing it just to
>>> make the code look nicer.
>>
>> For 4K system, with the optimization it only saves one
>> read_extent_buffer() call cost.
>
> This evaluation is wrong, you missed several things that
> generic_bin_search and read_extent_buffer do.
>
> generic_bin_search is called very often, each search slot so
> optimization is worth here
>
> read_extent_buffer is used _only_ for keys that cross page boundary, so
> we need to read the bytes in two steps and this is wrapped into a
> function that we call in a limited number of cases
Then to me, the better solution is to make read_extent_buffer() to be
split into two part.
Part 1 to handle the same page read, which should be made inline.
The part 1 should be small enough, as it only involves the in-page
offset calculation, which is also already done in current
generic_bin_search.
Then part 2 to handle the cross page case, and that part can be a
function call.
Personally speaking, even generic_bin_search() is a hot-path, I still
don't believe it's worthy, as read_extent_buffer() itself is also
frequently called in other locations, and I never see a special handling
for it in any other location.
Anyway, I will use the get_eb_page_offset()/get_eb_page_index() macros
here first, or subpage will be completely screwed.
And then try to use that two-part solution for read_extent_buffer().
Thanks,
Qu
>
> In all other cases, when the whole key is contained in the page the call
> is inline in generic_bin_search, ie. no function call overhead
>
>> Or we will need to manually call get_eb_page_offset() here to make it
>> work for subpage.
>
> For nodesize that is smaller than PAGE_SIZE there's no page crossing at
> all so using read_extent_buffer would be making things worse.
>
[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]
next prev parent reply other threads:[~2020-09-17 13:25 UTC|newest]
Thread overview: 58+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-15 5:35 [PATCH v2 00/19] btrfs: add read-only support for subpage sector size Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 01/19] btrfs: extent-io-tests: remove invalid tests Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 02/19] btrfs: remove the unnecessary parameter @start and @len for check_data_csum() Qu Wenruo
2020-09-15 8:39 ` Johannes Thumshirn
2020-09-15 5:35 ` [PATCH v2 03/19] btrfs: calculate inline extent buffer page size based on page size Qu Wenruo
2020-09-15 8:35 ` Nikolay Borisov
2020-09-15 10:05 ` Qu Wenruo
2020-09-15 8:40 ` Johannes Thumshirn
2020-09-15 5:35 ` [PATCH v2 04/19] btrfs: remove the open-code to read disk-key Qu Wenruo
2020-09-15 8:36 ` Nikolay Borisov
2020-09-15 8:40 ` Johannes Thumshirn
2020-09-16 16:01 ` David Sterba
2020-09-17 8:02 ` Qu Wenruo
2020-09-17 12:37 ` David Sterba
2020-09-17 13:15 ` Qu Wenruo [this message]
2020-09-17 22:41 ` David Sterba
2020-09-17 23:26 ` Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 05/19] btrfs: make btrfs_fs_info::buffer_radix to take sector size devided values Qu Wenruo
2020-09-15 8:27 ` Johannes Thumshirn
2020-09-15 10:04 ` Qu Wenruo
2020-09-15 10:12 ` Johannes Thumshirn
2020-09-15 17:40 ` kernel test robot
2020-09-15 5:35 ` [PATCH v2 06/19] btrfs: don't allow tree block to cross page boundary for subpage support Qu Wenruo
2020-09-15 8:37 ` Nikolay Borisov
2020-09-15 10:06 ` Qu Wenruo
2020-09-15 8:44 ` Johannes Thumshirn
2020-09-15 5:35 ` [PATCH v2 07/19] btrfs: update num_extent_pages() to support subpage sized extent buffer Qu Wenruo
2020-09-15 8:42 ` Johannes Thumshirn
2020-09-15 10:07 ` Qu Wenruo
2020-09-15 10:12 ` Johannes Thumshirn
2020-09-15 10:07 ` Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 08/19] btrfs: handle sectorsize < PAGE_SIZE case for extent buffer accessors Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 09/19] btrfs: make csum_tree_block() handle sectorsize smaller than page size Qu Wenruo
2020-09-15 8:47 ` Johannes Thumshirn
2020-09-15 5:35 ` [PATCH v2 10/19] btrfs: add assert_spin_locked() for attach_extent_buffer_page() Qu Wenruo
2020-09-15 8:52 ` Johannes Thumshirn
2020-09-15 5:35 ` [PATCH v2 11/19] btrfs: extract the extent buffer verification from btree_readpage_end_io_hook() Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 12/19] btrfs: extent_io: only require sector size alignment for page read Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 13/19] btrfs: make btrfs_readpage_end_io_hook() follow sector size Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 14/19] btrfs: make btree inode io_tree has its special owner Qu Wenruo
2020-09-16 9:28 ` Johannes Thumshirn
2020-09-16 16:06 ` David Sterba
2020-09-17 0:02 ` Qu Wenruo
2020-09-17 12:50 ` David Sterba
2020-09-18 8:18 ` Qu Wenruo
2020-09-22 14:06 ` David Sterba
2020-09-22 14:14 ` David Sterba
2020-09-15 5:35 ` [PATCH v2 15/19] btrfs: don't set extent_io_tree bits for btree inode at endio time Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 16/19] btrfs: use extent_io_tree to handle subpage extent buffer allocation Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 17/19] btrfs: implement subpage metadata read and its endio function Qu Wenruo
2020-09-16 8:47 ` kernel test robot
2020-09-15 5:35 ` [PATCH v2 18/19] btrfs: implement btree_readpage() and try_release_extent_buffer() for subpage Qu Wenruo
2020-09-15 5:35 ` [PATCH v2 19/19] btrfs: allow RO mount of 4K sector size fs on 64K page system Qu Wenruo
2020-09-16 1:35 ` [PATCH v2 00/19] btrfs: add read-only support for subpage sector size Qu Wenruo
2020-09-16 16:18 ` Neal Gompa
2020-09-17 0:03 ` Qu Wenruo
2020-09-17 0:13 ` Neal Gompa
2020-09-17 0:24 ` Qu Wenruo
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=b6f59a13-0572-01dc-656f-09f1b5eb7935@gmx.com \
--to=quwenruo.btrfs@gmx.com \
--cc=dsterba@suse.cz \
--cc=linux-btrfs@vger.kernel.org \
--cc=wqu@suse.com \
/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
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).