Re: [RFC PATCH v2 3/3] btrfs: Use kmap_local_page() on "in_page" in zlib_compress_pages()
From: Qu Wenruo
Date: Fri Jun 17 2022 - 18:16:40 EST
On 2022/6/18 02:13, Fabio M. De Francesco wrote:
On venerdì 17 giugno 2022 15:09:47 CEST Qu Wenruo wrote:
On 2022/6/17 20:05, Fabio M. De Francesco wrote:
The use of kmap() is being deprecated in favor of kmap_local_page().
With
kmap_local_page(), the mapping is per thread, CPU local and not
globally
visible.
Therefore, use kmap_local_page() / kunmap_local() on "in_page" in
zlib_compress_pages() because in this function the mappings are per
thread
and are not visible in other contexts.
Use an array based stack in order to take note of the order of mappings
and un-mappings to comply to the rules of nesting local mappings.
Any extra comment on the "rules of nesting local mappings" part?
Actually, I'm not sure about what to add. I thought that whoever need more
information about LIFO mappings / un-mappings can look at highmem.rst. I've
changed that document and now it contains information on why and how to use
kmap_local_page() in place of kmap() and kmap_atomic().
Am I misunderstanding what you are trying to say? If so, any specific
suggestions would be greatly appreciated.
Thanks for pointing to the doc, and that doc is enough to answer my
question.
Tested with xfstests on QEMU + KVM 32 bits VM with 4GB of RAM and
HIGHMEM64G enabled. This patch passes 26/26 tests of group "compress".
Suggested-by: Ira Weiny <ira.weiny@xxxxxxxxx>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@xxxxxxxxx>
---
fs/btrfs/zlib.c | 65 +++++++++++++++++++++++++++++++++++++++
+---------
1 file changed, 53 insertions(+), 12 deletions(-)
diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c
index c7c69ce4a1a9..1f16014e8ff3 100644
--- a/fs/btrfs/zlib.c
+++ b/fs/btrfs/zlib.c
@@ -99,6 +99,8 @@ int zlib_compress_pages(struct list_head *ws, struct
address_space *mapping,
int ret;
char *data_in = NULL;
char *cpage_out = NULL;
+ char mstack[2];
+ int sind = 0;
int nr_pages = 0;
struct page *in_page = NULL;
struct page *out_page = NULL;
@@ -126,6 +128,8 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
ret = -ENOMEM;
goto out;
}
+ mstack[sind] = 'A';
+ sind++;
cpage_out = kmap_local_page(out_page);
pages[0] = out_page;
nr_pages = 1;
@@ -148,26 +152,32 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
int i;
for (i = 0; i < in_buf_pages; i++)
{
- if (in_page) {
-
kunmap(in_page);
I don't think we really need to keep @in_page mapped for that long.
We only need the input pages (pages from inode page cache) when we run
out of input.
So what we really need is just to map the input, copy the data to
buffer, unmap the page.
+ if (data_in) {
+ sind--;
+
kunmap_local(data_in);
put_page(in_page);
}
in_page =
find_get_page(mapping,
start >> PAGE_SHIFT);
- data_in = kmap(in_page);
+ mstack[sind] = 'B';
+ sind++;
+ data_in =
kmap_local_page(in_page);
memcpy(workspace->buf + i
* PAGE_SIZE,
data_in,
PAGE_SIZE);
start += PAGE_SIZE;
}
workspace->strm.next_in =
workspace->buf;
} else {
I think we can clean up the code.
In fact the for loop can handle both case, I didn't see any special
reason to do different handling, we can always use workspace->buf,
instead of manually dancing using different paths.
As I said in a recent email, I'm relatively new to kernel development,
especially to Btrfs and other filesystems.
That's not a big deal, that's why we're here to provide help.
However, I noted that this code does different handling depending on how
many "in_page" is going to map. I am not able to say why...
AFAIK the reason is optimization.
The idea is like this, if there are multiple pages left as input, we
copy the pages from page cache into the workspace buffer.
If there is no more than one page left, we use that page from page cache
directly.
I believe that's the problem causing the difficult in converting to
kmap_local_page().
I believe with all these cleanup, it should be much simpler to convert
to kmap_local_page().
I'm pretty happy to provide help on this refactor if you don't feel
confident enough on this part of btrfs.
Thanks for any help you can provide, but let me be clear about what my goal
is. I've been assigned with the task to convert kmap() (and possibly also
kmap_atomic()) to kmap_local_page() wherever it can be done across the
entire kernel.
Furthermore, wherever it cannot be done with the API we already have,
changes to the API will be required. One small change has already been
carried out upon David's suggestion to make kunmap_local() to take pointers
to const void. However I'm also talking of much larger changes, if they are
needed.
This is to say that I cannot spend too much on Btrfs. There is lot of work
to be done in other subsystems where I don't yet know which kinds of
difficulties I'm going to find out.
Any help with clean-ups / refactoring of zlib_compress_pages() will be much
appreciated for the reasons I've tried to convey in the paragraphs above.
I'll send out a cleanup for zlib_compress_pages(), mostly to make the
(strm.avail_in == 0) branch to call kmap() and kunmap() in pairs,
without holding @in_page mapped.
Would that make it easier?
Thanks,
Qu
Thank you so much,
Fabio
Thanks,
Qu
- if (in_page) {
- kunmap(in_page);
+ if (data_in) {
+ sind--;
+ kunmap_local(data_in);
put_page(in_page);
}
in_page = find_get_page(mapping,
start
PAGE_SHIFT);
- data_in = kmap(in_page);
+ mstack[sind] = 'B';
+ sind++;
+ data_in = kmap_local_page(in_page);
start += PAGE_SIZE;
workspace->strm.next_in = data_in;
}
@@ -196,23 +206,39 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
* the stream end if required
*/
if (workspace->strm.avail_out == 0) {
+ sind--;
+ kunmap_local(data_in);
+ data_in = NULL;
+
+ sind--;
kunmap_local(cpage_out);
cpage_out = NULL;
+
if (nr_pages == nr_dest_pages) {
out_page = NULL;
+ put_page(in_page);
ret = -E2BIG;
goto out;
}
+
out_page = alloc_page(GFP_NOFS);
if (out_page == NULL) {
+ put_page(in_page);
ret = -ENOMEM;
goto out;
}
+
+ mstack[sind] = 'A';
+ sind++;
cpage_out = kmap_local_page(out_page);
pages[nr_pages] = out_page;
nr_pages++;
workspace->strm.avail_out = PAGE_SIZE;
workspace->strm.next_out = cpage_out;
+
+ mstack[sind] = 'B';
+ sind++;
+ data_in = kmap_local_page(in_page);
}
/* we're all done */
if (workspace->strm.total_in >= len)
@@ -235,10 +261,16 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
goto out;
} else if (workspace->strm.avail_out == 0) {
/* get another page for the stream end */
+ sind--;
+ kunmap_local(data_in);
+ data_in = NULL;
+
+ sind--;
kunmap_local(cpage_out);
cpage_out = NULL;
if (nr_pages == nr_dest_pages) {
out_page = NULL;
+ put_page(in_page);
ret = -E2BIG;
goto out;
}
@@ -247,11 +279,18 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
ret = -ENOMEM;
goto out;
}
+
+ mstack[sind] = 'A';
+ sind++;
cpage_out = kmap_local_page(out_page);
pages[nr_pages] = out_page;
nr_pages++;
workspace->strm.avail_out = PAGE_SIZE;
workspace->strm.next_out = cpage_out;
+
+ mstack[sind] = 'B';
+ sind++;
+ data_in = kmap_local_page(in_page);
}
}
zlib_deflateEnd(&workspace->strm);
@@ -266,13 +305,15 @@ int zlib_compress_pages(struct list_head *ws,
struct address_space *mapping,
*total_in = workspace->strm.total_in;
out:
*out_pages = nr_pages;
- if (cpage_out)
- kunmap_local(cpage_out);
-
- if (in_page) {
- kunmap(in_page);
- put_page(in_page);
+ while (--sind >= 0) {
+ if (mstack[sind] == 'B') {
+ kunmap_local(data_in);
+ put_page(in_page);
+ } else {
+ kunmap_local(cpage_out);
+ }
}
+
return ret;
}