<div dir="ltr">On 5 August 2013 19:46, Manish Sharma <span dir="ltr"><<a href="mailto:manishrma@gmail.com" target="_blank">manishrma@gmail.com</a>></span> wrote:<br><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Merged the two for loops. We might get a little gain by overlapping<br>
wait_on_bh and the memcpy operations.<br>
<br>
Signed-off-by: Manish Sharma <<a href="mailto:manishrma@gmail.com">manishrma@gmail.com</a>><br>
---<br>
fs/squashfs/block.c | 9 +++------<br>
1 file changed, 3 insertions(+), 6 deletions(-)<br>
<br>
diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c<br>
index fb50652..5012f98 100644<br>
--- a/fs/squashfs/block.c<br>
+++ b/fs/squashfs/block.c<br>
@@ -169,12 +169,6 @@ int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,<br>
*/<br>
int i, in, pg_offset = 0;<br>
<br>
- for (i = 0; i < b; i++) {<br>
- wait_on_buffer(bh[i]);<br>
- if (!buffer_uptodate(bh[i]))<br>
- goto block_release;<br>
- }<br>
-<br>
for (bytes = length; k < b; k++) {<br>
in = min(bytes, msblk->devblksize - offset);<br>
bytes -= in;<br>
@@ -185,6 +179,9 @@ int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,<br>
}<br>
avail = min_t(int, in, PAGE_CACHE_SIZE -<br>
pg_offset);<br>
+ wait_on_buffer(bh[k]);<br>
+ if (!buffer_uptodate(bh[k]))<br>
+ goto block_release;<br></blockquote><div><br></div><div>Two points:<br><br>1. I understand what you're trying to do here (merging the two loops is a good thing), but this patch is in the wrong place. It should be in the outer loop rather than the inner loop.<br>
<br></div><div>The outer loop cycles through the buffer heads, one buffer head per iteration. The inner loop copies the buffer head to the pages, and this can loop copying the same buffer head to multiple pages in the case there's not enough bytes in the page (if you want to know why, it's because we start off copying from an offset in the first buffer head).<br>
<br></div><div>So it's not a good idea to have the wait_on_buffer() in the inner loop, as we can unnecessarily call it multiple times on the same buffer head. The wait_on_buffer() should be in the outer loop where we know it will only be called once per buffer head.<br>
<br></div><div>I have checked the fixed patch into the "tmp" branch on my squashfs-next repository on <a href="http://git.kernel.org">git.kernel.org</a> here:<br><br><a href="https://git.kernel.org/cgit/linux/kernel/git/pkl/squashfs-next.git/commit/?h=tmp&id=5839f00feea122fb773d8520e5cfb16464fb89d5">https://git.kernel.org/cgit/linux/kernel/git/pkl/squashfs-next.git/commit/?h=tmp&id=5839f00feea122fb773d8520e5cfb16464fb89d5</a><br>
<br></div><div>(As the patch email unfortunately ended up in my gmail account, I'm replying from there, and so it's no point in inlining it, as gmail will corrupt it).<br><br></div><div>Please send a revised v2 patch with this fix. Thanks.<br>
<br></div><div>2. Your emailer corrupted the patch ... This is common occurrence with modern (wysiwyg) emailers. Please see<br><br><a href="https://www.kernel.org/doc/Documentation/email-clients.txt">https://www.kernel.org/doc/Documentation/email-clients.txt</a><br>
<br></div><div>these days it's probably best to use git send-email.<br><br></div><div>In case you're curious, this is how the emailer corrupted the patch. Your patch has<br><br>$ cat -vt /tmp/1-1-Squashfs-Optimized-uncompressed-buffer-loop.patch<br>
diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c<br></div><div>[SNIP]<br>@@ -169,12 +169,6 @@ int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,<br>^I^I */<br>^I^Iint i, in, pg_offset = 0;<br><br>
-^I^Ifor (i = 0; i < b; i++) {<br>-^I^I^Iwait_on_buffer(bh[i]);<br>-^I^I^Iif (!buffer_uptodate(bh[i]))<br>-^I^I^I^Igoto block_release;<br>-^I^I}<br>-<br>^I^Ifor (bytes = length; k < b; k++) {<br>^I^I^Iin = min(bytes, msblk->devblksize - offset);<br>
^I^I^Ibytes -= in;<br></div><div>[SNIP]<br><br></div><div>This should have been<br><br>@@ -169,12 +169,6 @@ int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,<br><space>^I^I */<br><space>^I^Iint i, in, pg_offset = 0;<br>
<br>-^I^Ifor (i = 0; i < b; i++) {<br>-^I^I^Iwait_on_buffer(bh[i]);<br>-^I^I^Iif (!buffer_uptodate(bh[i]))<br>-^I^I^I^Igoto block_release;<br>-^I^I}<br>-<br><space>^I^Ifor (bytes = length; k < b; k++) {<br><space>^I^I^Iin = min(bytes, msblk->devblksize - offset);<br>
<space>^I^I^Ibytes -= in;<br><br></div><div>where <space> should be read as " ", i.e. it has eliminated the leading space before the tabs.<br><br></div><div>Phillip<br></div></div></div></div>