lkml.org 
[lkml]   [2002]   [Sep]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
SubjectRe: [patch] readv/writev rework
From
Hello,

Your readv/writev patch interested me and I checked it.
I found we also have a chance to improve normal writev.

a_ops->prepare_write() and a_ops->commit_write will have a
penalty when I/O size isn't PAGE_SIZE.
With following patch generic_file_write_nolock() will try to
make each I/O size become PAGE_SIZE.

Could you apply it?

> This is Janet Morgan's patch which converts the readv/writev code
> to submit all segments for IO before waiting on them, rather than
> submitting each segment separately.
>
> This is a critical performance fix for O_DIRECT reads and writes.
> Prior to this change, O_DIRECT vectored IO was forced to wait for
> completion against each segment of the iovec rather than submitting all
> segments and waiting on the lot. ie: for ten segments, this code will
> be ten times faster.

Thank you,
Hirokazu Takahashi.



--- linux-2.5.34/mm/filemap.c.ORG Wed Sep 11 19:48:00 2030
+++ linux-2.5.34/mm/filemap.c Thu Sep 12 21:43:02 2030
@@ -2109,14 +2109,18 @@ generic_file_write_nolock(struct file *f
unsigned long index;
unsigned long offset;
long page_fault;
+ unsigned rest;
+ unsigned copy;
+ unsigned off;
+ struct iovec *work_iov;
+ char *work_buf;
+ unsigned work_iov_bytes;

offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
index = pos >> PAGE_CACHE_SHIFT;
bytes = PAGE_CACHE_SIZE - offset;
if (bytes > count)
bytes = count;
- if (bytes + written > iov_bytes)
- bytes = iov_bytes - written;

/*
* Bring in the user page that we will copy from _first_.
@@ -2144,7 +2148,29 @@ generic_file_write_nolock(struct file *f
vmtruncate(inode, inode->i_size);
break;
}
- page_fault = filemap_copy_from_user(page, offset, buf, bytes);
+
+ rest = bytes;
+ off = 0;
+ work_iov = cur_iov;
+ work_buf = buf;
+ work_iov_bytes = iov_bytes;
+ while (rest) {
+ copy = rest;
+ if (written + off == work_iov_bytes) {
+ work_iov++;
+ work_iov_bytes += work_iov->iov_len;
+ work_buf = work_iov->iov_base;
+ }
+ if (copy + written + off > work_iov_bytes)
+ copy = work_iov_bytes - written - off;
+
+ page_fault = filemap_copy_from_user(page, offset+off, work_buf, copy);
+ rest -= copy;
+ off += copy;
+ work_buf += copy;
+ if (unlikely(page_fault))
+ break;
+ }
status = a_ops->commit_write(file, page, offset, offset+bytes);
if (unlikely(page_fault)) {
status = -EFAULT;
@@ -2153,14 +2179,21 @@ generic_file_write_nolock(struct file *f
status = bytes;

if (status >= 0) {
- written += status;
count -= status;
pos += status;
- buf += status;
- if (written == iov_bytes && count) {
- cur_iov++;
- iov_bytes += cur_iov->iov_len;
- buf = cur_iov->iov_base;
+ rest = status;
+ while (rest) {
+ copy = rest;
+ if (written == iov_bytes) {
+ cur_iov++;
+ iov_bytes += cur_iov->iov_len;
+ buf = cur_iov->iov_base;
+ }
+ if (copy + written > iov_bytes)
+ copy = iov_bytes - written;
+ rest -= copy;
+ written += copy;
+ buf += copy;
}
}
}
-
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: 2005-03-22 13:28    [W:0.093 / U:2.048 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site