All of lore.kernel.org
 help / color / mirror / Atom feed
From: Arnd Bergmann <arnd@arndb.de>
To: Nick Piggin <npiggin@gmail.com>
Cc: linux-kernel@vger.kernel.org,
	Nick Bowler <nbowler@elliptictech.com>,
	Evgeniy Dushistov <dushistov@mail.ru>
Subject: Re: [PATCH 15/20] ufs: remove the BKL
Date: Thu, 27 Jan 2011 14:13:00 +0100	[thread overview]
Message-ID: <201101271413.01204.arnd@arndb.de> (raw)
In-Reply-To: <AANLkTinKO=HM1BqDdFkdcCAGz85co64Ma79qomw6PDPp@mail.gmail.com>

On Thursday 27 January 2011, Nick Piggin wrote:
> On Wed, Jan 26, 2011 at 9:17 AM, Arnd Bergmann <arnd@arndb.de> wrote:
> > This introduces a new per-superblock mutex in UFS to replace
> > the big kernel lock. I have been careful to avoid nested
> > calls to lock_ufs and to get the lock order right with
> > respect to other mutexes, in particular lock_super.
> 
> When I looked at removing bkl from minix a long time ago,
> I was a bit worried about reclaim and fs/io recursion in some
> of the filesystems with bkl.

Thanks for looking at this, you certainly understand these interactions
much better than me!

> > @@ -436,7 +439,8 @@ int ufs_getfrag_block(struct inode *inode, sector_t fragment, struct buffer_head
> >        ret = 0;
> >        bh = NULL;
> >
> > -       lock_kernel();
> > +       if (needs_lock)
> > +               lock_ufs(sb);
> >
> >        UFSD("ENTER, ino %lu, fragment %llu\n", inode->i_ino, (unsigned long long)fragment);
> >        if (fragment >


> So, get_block can be called for .writepage in page reclaim,
> which takes the lock. ufs_lookup takes the lock and winds
> up calling ufs_alloc_inode. And ufs_alloc_inode does
> GFP_KERNEL, which can enter reclaim with __GFP_FS
> set.

This is one path that I did not consider. My initial thought when you
mentioned it was that this is still fine because ufs_getfrag_block can be
called with the lock held by the current task, but you are of course
right that this won't work when we are called by the writeback task.

> I didn't look through all your filesystem conversions, but it is
> something tricky to watch out for I think.

The only other one where I added locks is isofs, and I'm pretty sure that's
fine because it's read-only and does not do writeback. HPFS with Andi's suggested
solution of making it UP-only is of course also fine.
 
> Changing everything to GFP_NOFS may be an option, for
> such crufty old filesystems...

Yes. All performance optimizations in UFS are pointless anyway as long as we
do most of the disk I/O under a global mutex. The patch below converts all
allocations that are done under a lock in UFS to GFP_NOFS. If nobody comes
up with a better solution, I'll fold that into my other patch.

	Arnd
---

 fs/ufs/super.c |    8 ++++----
 fs/ufs/util.c  |    2 +-
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/ufs/super.c b/fs/ufs/super.c
index e4e06ef..7693d62 100644
--- a/fs/ufs/super.c
+++ b/fs/ufs/super.c
@@ -539,7 +539,7 @@ static int ufs_read_cylinder_structures(struct super_block *sb)
 	 */
 	size = uspi->s_cssize;
 	blks = (size + uspi->s_fsize - 1) >> uspi->s_fshift;
-	base = space = kmalloc(size, GFP_KERNEL);
+	base = space = kmalloc(size, GFP_NOFS);
 	if (!base)
 		goto failed; 
 	sbi->s_csp = (struct ufs_csum *)space;
@@ -564,7 +564,7 @@ static int ufs_read_cylinder_structures(struct super_block *sb)
 	 * Read cylinder group (we read only first fragment from block
 	 * at this time) and prepare internal data structures for cg caching.
 	 */
-	if (!(sbi->s_ucg = kmalloc (sizeof(struct buffer_head *) * uspi->s_ncg, GFP_KERNEL)))
+	if (!(sbi->s_ucg = kmalloc (sizeof(struct buffer_head *) * uspi->s_ncg, GFP_NOFS)))
 		goto failed;
 	for (i = 0; i < uspi->s_ncg; i++) 
 		sbi->s_ucg[i] = NULL;
@@ -582,7 +582,7 @@ static int ufs_read_cylinder_structures(struct super_block *sb)
 		ufs_print_cylinder_stuff(sb, (struct ufs_cylinder_group *) sbi->s_ucg[i]->b_data);
 	}
 	for (i = 0; i < UFS_MAX_GROUP_LOADED; i++) {
-		if (!(sbi->s_ucpi[i] = kmalloc (sizeof(struct ufs_cg_private_info), GFP_KERNEL)))
+		if (!(sbi->s_ucpi[i] = kmalloc (sizeof(struct ufs_cg_private_info), GFP_NOFS)))
 			goto failed;
 		sbi->s_cgno[i] = UFS_CGNO_EMPTY;
 	}
@@ -1415,7 +1415,7 @@ static struct kmem_cache * ufs_inode_cachep;
 static struct inode *ufs_alloc_inode(struct super_block *sb)
 {
 	struct ufs_inode_info *ei;
-	ei = (struct ufs_inode_info *)kmem_cache_alloc(ufs_inode_cachep, GFP_KERNEL);
+	ei = (struct ufs_inode_info *)kmem_cache_alloc(ufs_inode_cachep, GFP_NOFS);
 	if (!ei)
 		return NULL;
 	ei->vfs_inode.i_version = 1;
diff --git a/fs/ufs/util.c b/fs/ufs/util.c
index d2c36d5..95425b5 100644
--- a/fs/ufs/util.c
+++ b/fs/ufs/util.c
@@ -27,7 +27,7 @@ struct ufs_buffer_head * _ubh_bread_ (struct ufs_sb_private_info * uspi,
 	if (count > UFS_MAXFRAG)
 		return NULL;
 	ubh = (struct ufs_buffer_head *)
-		kmalloc (sizeof (struct ufs_buffer_head), GFP_KERNEL);
+		kmalloc (sizeof (struct ufs_buffer_head), GFP_NOFS);
 	if (!ubh)
 		return NULL;
 	ubh->fragment = fragment;

  reply	other threads:[~2011-01-27 13:13 UTC|newest]

Thread overview: 64+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-01-25 22:17 [RFC 00/20] Proposal for remaining BKL users Arnd Bergmann
2011-01-25 22:17 ` Arnd Bergmann
2011-01-25 22:17 ` [PATCH 01/20] drm/i810: remove the BKL Arnd Bergmann
2011-01-25 22:17 ` [PATCH 02/20] drm: remove i830 driver Arnd Bergmann
2011-01-25 22:17 ` [PATCH 03/20] staging/usbip: convert to kthread Arnd Bergmann
2011-01-28 17:53   ` Max Vozeler
2011-01-28 18:48     ` Arnd Bergmann
2011-03-01 22:15     ` Arnd Bergmann
2011-01-25 22:17 ` [PATCH 04/20] staging/cx25721: serialize access to devlist Arnd Bergmann
2011-01-26 16:23   ` Palash Bandyopadhyay
2011-01-31 21:37   ` Greg KH
2011-01-25 22:17 ` [PATCH 05/20] staging/go7007: remove the BKL Arnd Bergmann
2011-01-25 22:17 ` [PATCH 06/20] staging: Remove autofs3 Arnd Bergmann
2011-01-26  7:41   ` H. Peter Anvin
2011-01-25 22:17 ` [PATCH 07/20] staging: remove smbfs Arnd Bergmann
2011-01-25 22:17 ` [PATCH 08/20] adfs: remove the big kernel lock Arnd Bergmann
2011-01-25 22:20   ` Russell King
2011-01-25 22:17 ` [PATCH 09/20] hpfs: rename big kernel lock to hpfs_lock Arnd Bergmann
2011-01-25 22:17 ` [PATCH 10/20] hpfs: replace BKL with a global mutex Arnd Bergmann
2011-01-26  0:15   ` Andi Kleen
2011-01-26  0:19   ` Andi Kleen
2011-01-26 12:48     ` [PATCH v2] hpfs: remove the BKL Arnd Bergmann
2011-01-26 12:50     ` [PATCH 10/20] hpfs: replace BKL with a global mutex Arnd Bergmann
2011-01-26 16:52       ` Andi Kleen
2011-01-27  5:01         ` Nick Piggin
2011-01-27 10:57           ` Miklos Szeredi
2011-01-25 22:17 ` [PATCH 11/20] hpfs: move to drivers/staging Arnd Bergmann
2011-02-07 16:17   ` Mikulas Patocka
2011-02-07 19:31     ` Arnd Bergmann
2011-01-25 22:17 ` [PATCH 12/20] x25: remove the BKL Arnd Bergmann
2011-01-27 10:07   ` Andrew Hendry
2011-01-27 12:17     ` Arnd Bergmann
2011-01-27 12:38       ` [PATCH v2] " Arnd Bergmann
2011-01-27 13:20         ` Eric Dumazet
2011-01-27 13:43           ` Arnd Bergmann
2011-01-25 22:17 ` [PATCH 13/20] appletalk: move to staging Arnd Bergmann
2011-01-25 22:17 ` [PATCH 14/20] staging/appletalk: remove the BKL Arnd Bergmann
2011-01-25 22:29   ` David Miller
2011-01-26 12:57     ` Arnd Bergmann
2011-01-25 22:17 ` [PATCH 15/20] ufs: " Arnd Bergmann
2011-01-26  2:30   ` Nick Bowler
2011-01-26 12:53     ` Arnd Bergmann
2011-01-27  5:47   ` Nick Piggin
2011-01-27 13:13     ` Arnd Bergmann [this message]
2011-01-25 22:17 ` [PATCH 16/20] ipx: " Arnd Bergmann
2011-01-25 22:17 ` [PATCH 17/20] tracing: don't trace " Arnd Bergmann
2011-01-25 22:28   ` Frederic Weisbecker
2011-01-25 22:17 ` [PATCH 18/20] rtmutex-tester: remove BKL tests Arnd Bergmann
2011-01-26 15:00   ` [tip:core/locking] rtmutex-tester: Remove " tip-bot for Arnd Bergmann
2011-02-22 20:57   ` [tip:irq/core] rtmutex: tester: " tip-bot for Arnd Bergmann
2011-01-25 22:17 ` [PATCH 19/20] drivers: remove extraneous includes of smp_lock.h Arnd Bergmann
2011-01-25 22:17 ` [PATCH 20/20] BKL: That's all, folks Arnd Bergmann
2011-01-26  6:19   ` Ingo Molnar
2011-01-26  8:47     ` Alan Cox
2011-01-26 11:01       ` Ingo Molnar
2011-01-26 11:22   ` Thomas Gleixner
2011-01-26  2:22 ` [RFC 00/20] Proposal for remaining BKL users Greg KH
2011-01-26  2:22   ` Greg KH
2011-01-26 11:31   ` Arnd Bergmann
2011-01-26 11:31     ` Arnd Bergmann
2011-01-26 11:58     ` Mauro Carvalho Chehab
2011-01-26 13:45       ` Arnd Bergmann
2011-01-26 13:45         ` Arnd Bergmann
2011-01-26 16:24         ` Palash Bandyopadhyay

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=201101271413.01204.arnd@arndb.de \
    --to=arnd@arndb.de \
    --cc=dushistov@mail.ru \
    --cc=linux-kernel@vger.kernel.org \
    --cc=nbowler@elliptictech.com \
    --cc=npiggin@gmail.com \
    --subject='Re: [PATCH 15/20] ufs: remove the BKL' \
    /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

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.