From: Al Viro <viro@zeniv.linux.org.uk>
To: David Howells <dhowells@redhat.com>
Cc: linux-fsdevel@vger.kernel.org,
Dave Wysochanski <dwysocha@redhat.com>,
Marc Dionne <marc.dionne@auristor.com>,
"Matthew Wilcox (Oracle)" <willy@infradead.org>,
Christoph Hellwig <hch@lst.de>,
linux-mm@kvack.org, linux-cachefs@redhat.com,
linux-afs@lists.infradead.org, linux-nfs@vger.kernel.org,
linux-cifs@vger.kernel.org, ceph-devel@vger.kernel.org,
v9fs-developer@lists.sourceforge.net,
Trond Myklebust <trond.myklebust@hammerspace.com>,
Anna Schumaker <anna.schumaker@netapp.com>,
Steve French <sfrench@samba.org>,
Dominique Martinet <asmadeus@codewreck.org>,
Jeff Layton <jlayton@redhat.com>,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH v7 01/31] iov_iter: Add ITER_XARRAY
Date: Sun, 25 Apr 2021 13:14:13 +0000 [thread overview]
Message-ID: <YIVrJT8GwLI0Wlgx@zeniv-ca.linux.org.uk> (raw)
In-Reply-To: <161918448151.3145707.11541538916600921083.stgit@warthog.procyon.org.uk>
On Fri, Apr 23, 2021 at 02:28:01PM +0100, David Howells wrote:
> diff --git a/include/linux/uio.h b/include/linux/uio.h
> index 27ff8eb786dc..5f5ffc45d4aa 100644
> --- a/include/linux/uio.h
> +++ b/include/linux/uio.h
> @@ -10,6 +10,7 @@
> #include <uapi/linux/uio.h>
>
> struct page;
> +struct address_space;
> struct pipe_inode_info;
>
> struct kvec {
What is that chunk for?
> +#define iterate_all_kinds(i, n, v, I, B, K, X) { \
> if (likely(n)) { \
> size_t skip = i->iov_offset; \
> if (unlikely(i->type & ITER_BVEC)) { \
> @@ -88,6 +125,9 @@
> struct kvec v; \
> iterate_kvec(i, n, v, kvec, skip, (K)) \
> } else if (unlikely(i->type & ITER_DISCARD)) { \
> + } else if (unlikely(i->type & ITER_XARRAY)) { \
> + struct bio_vec v; \
> + iterate_xarray(i, n, v, skip, (X)); \
> } else { \
> const struct iovec *iov; \
> struct iovec v; \
> @@ -96,7 +136,7 @@
> } \
> }
For the record - these forests of macros had been my mistake. I'm trying
to get rid of that crap right now, but your changes don't look likely to be
trouble in that respect.
> @@ -738,6 +783,16 @@ size_t _copy_mc_to_iter(const void *addr, size_t bytes, struct iov_iter *i)
> bytes = curr_addr - s_addr - rem;
> return bytes;
> }
> + }),
> + ({
> + rem = copy_mc_to_page(v.bv_page, v.bv_offset,
> + (from += v.bv_len) - v.bv_len, v.bv_len);
> + if (rem) {
> + curr_addr = (unsigned long) from;
> + bytes = curr_addr - s_addr - rem;
> + rcu_read_unlock();
> + return bytes;
> + }
That's broken, same way as kvec and bvec cases are in the same primitive.
Iterator not advanced on failure halfway through.
> @@ -1246,7 +1349,8 @@ unsigned long iov_iter_alignment(const struct iov_iter *i)
> iterate_all_kinds(i, size, v,
> (res |= (unsigned long)v.iov_base | v.iov_len, 0),
> res |= v.bv_offset | v.bv_len,
> - res |= (unsigned long)v.iov_base | v.iov_len
> + res |= (unsigned long)v.iov_base | v.iov_len,
> + res |= v.bv_offset | v.bv_len
> )
> return res;
> }
Hmm... That looks like a really bad overkill - do you need anything beyond count and
iov_offset in that case + perhaps "do we have the very last page"? IOW, do you need
to iterate anything at all here? What am I missing here?
> @@ -1268,7 +1372,9 @@ unsigned long iov_iter_gap_alignment(const struct iov_iter *i)
> (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
> (size != v.bv_len ? size : 0)),
> (res |= (!res ? 0 : (unsigned long)v.iov_base) |
> - (size != v.iov_len ? size : 0))
> + (size != v.iov_len ? size : 0)),
> + (res |= (!res ? 0 : (unsigned long)v.bv_offset) |
> + (size != v.bv_len ? size : 0))
> );
> return res;
> }
Very limited use; it shouldn't be called for anything other than IOV_ITER case.
> @@ -1849,7 +2111,12 @@ int iov_iter_for_each_range(struct iov_iter *i, size_t bytes,
> kunmap(v.bv_page);
> err;}), ({
> w = v;
> - err = f(&w, context);})
> + err = f(&w, context);}), ({
> + w.iov_base = kmap(v.bv_page) + v.bv_offset;
> + w.iov_len = v.bv_len;
> + err = f(&w, context);
> + kunmap(v.bv_page);
> + err;})
Would be easier to have that sucker removed first...
Anyway, I can live with that; the only real bug is in sodding _copy_mc_to_iter(),
it's not anything new and it can be fixed at the same time we deal with kvec and
bvec cases there. Which, unfortunately, requires untangling the macro mess ;-/
What I've got in a local branch right now is
* removal of iov_iter_for_each_range() (yours, BTW)
* separation of flavour and direction (and the end of pseudo-bitmaps)
* untangling and optimizing iov_iter_advance(); iovec/kvec cases are
switched to the logics similar to bvec_iter_advance(), get considerably smaller
and should be faster
* fixing ITER_DISCARD iov_iter_advance() - move past the end should
quietly stop at the end.
* getting rid of iterate_all_kinds() in iov_iter_alignment(),
iov_iter_gap_alignment(), iov_iter_get_pages() and iov_iter_get_pages_alloc().
After that the only remaining irregular case of iterate_all_kinds() is in
iov_iter_npages(); that's what I'm trying to sort out right now. With that
done, all remaining uses will be for copying-style primitives, same as for
iterate_and_advance(). What I want to try after that is a separate "tracking"
argument, so that e.g. in _copy_to_iter() we'd have
iterate_and_advance(i, bytes, from, v,
copyout(v.iov_base, from, v.iov_len),
memcpy_to_page(v.bv_page, v.bv_offset, from, v.bv_len),
memcpy(v.iov_base, from, v.iov_len)
)
Next step will be to teach the damn thing about the possibility of short
copies in kvec/bvec cases. We'd get
#define iterate_and_advance(i, n, p, v, I, K, B) \
__iterate_and_advance(i, n, p, v, I, (K, 0), (B, 0))
and AFAICS it can be done in a way that won't screw code generation for
the normal ones. At that point _copy_mc_to_iter() mess gets cleared *AND*
we can merge K and B callbacks, handling B as kmap_atomic + K + kunmap_atomic
(_copy_mc_to_iter() is the main obstacle to that). Your callback (X) would
also fold into that.
After that I want to try and see how well iov_iter_advance() got optimized
and see if we can get e.g. _copy_to_iter() simply to
iterate_all_kinds(i, bytes, from, v,
copyout(v.iov_base, from, v.iov_len),
memcpy(v.iov_base, from, v.iov_len)
)
iov_iter_advance(i, from - addr);
return from - addr;
If iov_iter_advance() ends up being too much overhead - oh, well, we'll keep
iterate_and_advance() along with iterate_all_kinds(). Needs profiling,
obviously.
next prev parent reply other threads:[~2021-04-25 13:14 UTC|newest]
Thread overview: 53+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-23 13:27 [PATCH v7 00/31] Network fs helper library & fscache kiocb API David Howells
2021-04-23 13:28 ` [PATCH v7 01/31] iov_iter: Add ITER_XARRAY David Howells
2021-04-23 14:06 ` Matthew Wilcox
2021-04-23 14:33 ` David Howells
2021-04-25 13:14 ` Al Viro [this message]
2021-04-25 13:58 ` David Howells
2021-04-25 14:16 ` Al Viro
2021-04-26 18:54 ` Al Viro
2021-04-26 19:15 ` Jeff Layton
2021-04-26 19:23 ` David Howells
2021-04-26 19:52 ` Al Viro
2021-04-23 13:28 ` [PATCH v7 02/31] mm: Add set/end/wait functions for PG_private_2 David Howells
2021-04-23 13:28 ` [PATCH v7 03/31] mm/filemap: Pass the file_ra_state in the ractl David Howells
2021-04-23 13:28 ` [PATCH v7 04/31] fs: Document file_ra_state David Howells
2021-04-23 13:28 ` [PATCH v7 05/31] mm/readahead: Handle ractl nr_pages being modified David Howells
2021-04-23 13:29 ` [PATCH v7 06/31] mm: Implement readahead_control pageset expansion David Howells
2021-04-23 13:29 ` [PATCH v7 07/31] netfs: Make a netfs helper module David Howells
2021-04-29 8:04 ` Geert Uytterhoeven
2021-04-29 8:41 ` David Howells
2021-04-29 8:43 ` Dominique Martinet
2021-04-23 13:29 ` [PATCH v7 08/31] netfs: Documentation for helper library David Howells
2021-04-23 13:29 ` [PATCH v7 09/31] netfs, mm: Move PG_fscache helper funcs to linux/netfs.h David Howells
2021-04-23 13:29 ` [PATCH v7 10/31] netfs, mm: Add set/end/wait_on_page_fscache() aliases David Howells
2021-04-23 13:30 ` [PATCH v7 11/31] netfs: Provide readahead and readpage netfs helpers David Howells
2021-04-23 13:30 ` [PATCH v7 12/31] netfs: Add tracepoints David Howells
2021-04-23 13:30 ` [PATCH v7 13/31] netfs: Gather stats David Howells
2021-04-23 13:30 ` [PATCH v7 14/31] netfs: Add write_begin helper David Howells
2021-04-23 13:31 ` [PATCH v7 15/31] netfs: Define an interface to talk to a cache David Howells
2021-04-23 13:31 ` [PATCH v7 16/31] netfs: Add a tracepoint to log failures that would be otherwise unseen David Howells
2021-04-23 13:31 ` [PATCH v7 17/31] fscache, cachefiles: Add alternate API to use kiocb for read/write to cache David Howells
2021-04-23 13:32 ` [PATCH v7 18/31] afs: Disable use of the fscache I/O routines David Howells
2021-04-23 13:32 ` [PATCH v7 19/31] afs: Pass page into dirty region helpers to provide THP size David Howells
2021-04-23 13:32 ` [PATCH v7 20/31] afs: Print the operation debug_id when logging an unexpected data version David Howells
2021-04-23 13:32 ` [PATCH v7 21/31] afs: Move key to afs_read struct David Howells
2021-04-23 13:33 ` [PATCH v7 22/31] afs: Don't truncate iter during data fetch David Howells
2021-04-23 13:33 ` [PATCH v7 23/31] afs: Log remote unmarshalling errors David Howells
2021-04-23 13:33 ` [PATCH v7 24/31] afs: Set up the iov_iter before calling afs_extract_data() David Howells
2021-04-23 13:33 ` [PATCH v7 25/31] afs: Use ITER_XARRAY for writing David Howells
2021-04-23 13:33 ` [PATCH v7 26/31] afs: Wait on PG_fscache before modifying/releasing a page David Howells
2021-04-23 13:34 ` [PATCH v7 27/31] afs: Extract writeback extension into its own function David Howells
2021-04-23 13:34 ` [PATCH v7 28/31] afs: Prepare for use of THPs David Howells
2021-04-23 13:34 ` [PATCH v7 29/31] afs: Use the fs operation ops to handle FetchData completion David Howells
2021-04-23 13:34 ` [PATCH v7 30/31] afs: Use new netfs lib read helper API David Howells
2021-04-23 13:34 ` [PATCH v7 31/31] afs: Use the netfs_write_begin() helper David Howells
2021-04-25 23:14 ` [PATCH] iov_iter: Four fixes for ITER_XARRAY David Howells
2021-04-26 17:14 ` Al Viro
2021-04-26 19:02 ` Jeff Layton
2021-04-26 21:38 ` David Wysochanski
2021-04-26 21:06 ` [PATCH] netfs: Miscellaneous fixes David Howells
2021-04-26 21:09 ` Matthew Wilcox
2021-04-26 21:20 ` [PATCH v2] " David Howells
2021-04-26 22:06 ` Matthew Wilcox
2021-04-26 22:17 ` Jeff Layton
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=YIVrJT8GwLI0Wlgx@zeniv-ca.linux.org.uk \
--to=viro@zeniv.linux.org.uk \
--cc=anna.schumaker@netapp.com \
--cc=asmadeus@codewreck.org \
--cc=ceph-devel@vger.kernel.org \
--cc=dhowells@redhat.com \
--cc=dwysocha@redhat.com \
--cc=hch@lst.de \
--cc=jlayton@redhat.com \
--cc=linux-afs@lists.infradead.org \
--cc=linux-cachefs@redhat.com \
--cc=linux-cifs@vger.kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=linux-nfs@vger.kernel.org \
--cc=marc.dionne@auristor.com \
--cc=sfrench@samba.org \
--cc=trond.myklebust@hammerspace.com \
--cc=v9fs-developer@lists.sourceforge.net \
--cc=willy@infradead.org \
/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).