Ralf Baechle [Sun, 18 Jun 2006 00:32:22 +0000 (01:32 +0100)]
[MIPS] Cleanup memory managment initialization.
Historically plat_mem_setup did the entire platform initialization. This
was rather impractical because it meant plat_mem_setup had to get away
without any kind of memory allocator. To keep old code from breaking
plat_setup was just renamed to plat_setup and a second platform
initialization hook for anything else was introduced.
Ralf Baechle [Mon, 12 Jun 2006 11:20:09 +0000 (12:20 +0100)]
[MIPS] Drop 0 definition for kern_addr_valid
kern_addr_valid is currently only being used in kmem_ptr_validate which
is making some vague attempt at verfying the validity of an address.
Only IA-64, PARISC and x86-64 actually make some actual effort to verify
the validity of the pointer. Most architecture definitions of
kern_addr_valid() just define it as 1; the Alpha and CONFIG_DISCONTIGMEM
on i386 and MIPS even as 0; the 0-definition will result in
kmem_ptr_validate always failing which in turn will cause d_validate to
always fail. d_validate's only two users are smbfs and ncpfs, so the
0 definition ended breaking those ...
Ralf Baechle [Sun, 11 Jun 2006 22:03:08 +0000 (23:03 +0100)]
[MIPS] Cleanup ARCH_DISCONTIGMEM_ENABLE and NUMA configuration.
IP27 configuration isn't the only NUMA system - it just happens to be
the currently only supported MIPS NUMA system. So move the necessary
options back into the main MIPS Kconfig file.
Sergei Shtylyov [Sat, 27 May 2006 20:04:01 +0000 (00:04 +0400)]
[MIPS] arch/mips/au1000/time.c cleanup
Mark au1xxx_timer_setup() __init, just because it is. Get rid of
unneeded extern's (note that (*do_gettimeoffset)() is already declared by
<asm/time.c>) and an unused variable. Kill some whitespace...
Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com> Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Atsushi Nemoto [Mon, 15 May 2006 16:26:03 +0000 (01:26 +0900)]
[MIPS] Unify mips_fpu_soft_struct and mips_fpu_hard_structs.
The struct mips_fpu_soft_struct and mips_fpu_hard_struct are
completely same now and the kernel fpu emulator assumes that. This
patch unifies them to mips_fpu_struct and get rid of mips_fpu_union.
Atsushi Nemoto [Sun, 11 Jun 2006 14:25:43 +0000 (23:25 +0900)]
[MIPS] Fix futex_atomic_op_inuser.
I found that NPTL's pthread_cond_signal() does not work properly on
kernels compiled by gcc 4.1.x. I suppose inline asm for
__futex_atomic_op() was wrong. I suppose:
1. "=&r" constraint should be used for oldval.
2. Instead of "r" (uaddr), "=R" (*uaddr) for output and "R" (*uaddr)
for input should be used.
3. "memory" should be added to the clobber list.
[PATCH] Fix BCM1480 doubled process accounting times.
Running a UP kernel on a bcm1480 board, I get nonsensical timing
results, like this:
release@unknown:~/tmp$ time ./a.out
real 0m22.906s
user 0m45.792s
sys 0m0.010s
According to my watch, this program took 23 seconds to run, so the real
time clock is OK. It is process accounting that is broken.
I tracked this down to a problem with the function
bcm1480_timer_interrupt in the file sibyte/bcm1480/time.c. This
function calls ll_timer_interrupt for cpu0, and ll_local_timer_interrupt
for all cpus. However, both of these functions do process accounting.
Thus processes running on cpu0 end up with doubled times. This is very
obvious in a UP kernel where all processes run on cpu0.
The correct way to do this is to only call ll_local_timer interrupt if
this is not cpu0. This can be seen in the mips-board/generic/time.c
file, and also in the sibyte/sb1250/time.c file, both of which handle
this correctly. I fixed the bcm1480/time.c file by copying over the
correct code from the sb1250/time.c file.
With this fix, I now get sensible results.
release@unknown:~/tmp$ time ./a.out
real 0m22.903s
user 0m22.894s
sys 0m0.006s
[MIPS] Malta: Handle byteswapping hardare bug in big endian mode.
The SOC-it system controller running in big endian mode might forget
byteswapping when DMAing to the last word of physical memory. Fixed by
ignoring the last page of memory.
Arnd Bergmann [Thu, 15 Jun 2006 11:15:44 +0000 (21:15 +1000)]
[PATCH] powerpc: Fix 64k pages on non-partitioned machines
The page size encoding passed to tlbie is incorrect for new-style
large pages. This fixes it. This doesn't affect anything on older
machines because mmu_psize_defs[psize].penc (the page size encoding)
is 0 for 4k and 16M pages (the two are distinguished by a separate "is
a large page" bit).
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Oleg Nesterov [Thu, 15 Jun 2006 16:12:02 +0000 (20:12 +0400)]
[PATCH] arm_timer: remove a racy and obsolete PF_EXITING check
arm_timer() checks PF_EXITING to prevent BUG_ON(->exit_state)
in run_posix_cpu_timers().
However, for some reason it does so only for CPUCLOCK_PERTHREAD
case (which is imho wrong).
Also, this check is not reliable, PF_EXITING could be set on
another cpu without any locks/barriers just after the check,
so it can't prevent from attaching the timer to the exiting
task.
Oleg Nesterov [Thu, 15 Jun 2006 16:11:43 +0000 (20:11 +0400)]
[PATCH] run_posix_cpu_timers: remove a bogus BUG_ON()
do_exit() clears ->it_##clock##_expires, but nothing prevents
another cpu to attach the timer to exiting process after that.
arm_timer() tries to protect against this race, but the check
is racy.
After exit_notify() does 'write_unlock_irq(&tasklist_lock)' and
before do_exit() calls 'schedule() local timer interrupt can find
tsk->exit_state != 0. If that state was EXIT_DEAD (or another cpu
does sys_wait4) interrupted task has ->signal == NULL.
At this moment exiting task has no pending cpu timers, they were
cleanuped in __exit_signal()->posix_cpu_timers_exit{,_group}(),
so we can just return from irq.
Oleg Nesterov [Thu, 15 Jun 2006 16:11:15 +0000 (20:11 +0400)]
[PATCH] check_process_timers: fix possible lockup
If the local timer interrupt happens just after do_exit() sets PF_EXITING
(and before it clears ->it_xxx_expires) run_posix_cpu_timers() will call
check_process_timers() with tasklist_lock + ->siglock held and
check_process_timers:
t = tsk;
do {
....
do {
t = next_thread(t);
} while (unlikely(t->flags & PF_EXITING));
} while (t != tsk);
the outer loop will never stop.
Actually, the window is bigger. Another process can attach the timer
after ->it_xxx_expires was cleared (see the next commit) and the 'if
(PF_EXITING)' check in arm_timer() is racy (see the one after that).
A couple of fixes that should prevent crashes when using netconsole and
suspend/resume. First, netconsole poll routine shouldn't run unless the
device is up; second, the NAPI poll should be disabled during suspend.
This is only an issue on sky2, because it has to have one NAPI poll
routine for both ports on dual port boards. Normal drivers use
netif_rx_schedule_prep and that checks for netif_running.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Jens Axboe [Fri, 16 Jun 2006 11:02:29 +0000 (13:02 +0200)]
[PATCH] Fix missing ret assignment in __bio_map_user() error path
If get_user_pages() returns less pages than what we asked for, we jump
to out_unmap which will return ERR_PTR(ret). But ret can contain a
positive number just smaller than local_nr_pages, so be sure to set it
to -EFAULT always.
Problem found and diagnosed by Damien Le Moal <damien@sdl.hitachi.co.jp>
Jens Axboe [Fri, 16 Jun 2006 05:46:37 +0000 (07:46 +0200)]
[PATCH] fix cdrom open
Some time ago the cdrom open routine was changed so that we call the
driver's open routine before checking to see if it is read only. However,
if we discovered that a read write open was not possible and the open
flags required a writable open, we just returned -EROFS without calling
the driver's release routine. This seems to work for most cdrom drivers,
but breaks the Powerpc iSeries virtual cdrom rather badly.
This just inserts the release call in the error path to balance the call
to "->open()" done by "open_for_data()".
Jens Axboe [Wed, 14 Jun 2006 17:11:57 +0000 (19:11 +0200)]
[PATCH] cfq-iosched: fix crash in do_div()
We don't clear the seek stat values in cfq_alloc_io_context(), and if
->seek_mean is unlucky enough to be set to -36 by chance, the first
invocation of cfq_update_io_seektime() will oops with a divide by zero
in do_div().
Just memset the entire cic instead of filling invididual values
independently.
[PATCH] sky2: stop/start hardware idle timer on suspend/resume
The resume bug was caused not by an early interrupt but because the idle
timeout was not being stopped on suspend. Also disable hardware IRQ's
on suspend. Will need to revisit this with hotplug?
Signed-off-by: Stephen Hemminger <shemminger@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The set power state function is cleaner if it doesn't return anything.
The only caller that could fail is in suspend() and it can check the argument
there.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Randy Dunlap [Mon, 12 Jun 2006 22:13:40 +0000 (15:13 -0700)]
[PATCH] alpha: generic hweight build fix
From: Randy Dunlap <rdunlap@xenotime.net>
According to include/asm-alpha/bitops.h, only ALPHA_EV67 has hardware
hweight support, so ALPHA_EV6 needs to use GENERIC_HWEIGHT.
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Ernst Herzberg <earny@net4u.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Sergey Vlasov [Mon, 12 Jun 2006 20:53:23 +0000 (21:53 +0100)]
[PATCH] tmpfs: Decrement i_nlink correctly in shmem_rmdir()
shmem_rmdir() must undo the increment of i_nlink done in
shmem_get_inode() for directories, otherwise at least
IN_DELETE_SELF inotify event generation is broken.
Robin H. Johnson [Mon, 12 Jun 2006 20:50:25 +0000 (21:50 +0100)]
[PATCH] tmpfs: time granularity fix for [acm]time going backwards
I noticed a strange behavior in a tmpfs file system the other day, while
building packages - occasionally, and seemingly at random, make decided to
rebuild a target. However, only on tmpfs.
A file would be created, and if checked, it had a sub-second timestamp.
However, after an utimes related call where sub-seconds should be set, they
were zeroed instead. In the case that a file was created, and utimes(...,NULL)
was used on it in the same second, the timestamp on the file moved backwards.
After some digging, I found that this was being caused by tmpfs not having a
time granularity set, thus inheriting the default 1 second granularity.
Hugh adds: yes, we missed tmpfs when the s_time_gran mods went into 2.6.11.
Unfortunately, the granularity of CURRENT_TIME, often used in filesystems,
does not match the default granularity set by alloc_super. A few more such
discrepancies have been found, but this is the most important to fix now.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org> Acked-by: Andi Kleen <ak@suse.de> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Linus Torvalds [Mon, 12 Jun 2006 19:53:27 +0000 (12:53 -0700)]
[sky2] Fix sky2 network driver suspend/resume
This fixes two independent problems: it would not save the PCI state on
suspend (and thus try to resume a nonexistent state on resume), and
while shut off, if an interrupt happened on the same shared irq, the irq
handler would react very badly to the interrupt status being an invalid
all-ones state.
Acked-by: Jeff Garzik <jgarzik@pobox.com> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Aki M Nyrhinen [Mon, 12 Jun 2006 04:18:56 +0000 (21:18 -0700)]
[TCP]: continued: reno sacked_out count fix
From: Aki M Nyrhinen <anyrhine@cs.helsinki.fi>
IMHO the current fix to the problem (in_flight underflow in reno)
is incorrect. it treats the symptons but ignores the problem. the
problem is timing out packets other than the head packet when we
don't have sack. i try to explain (sorry if explaining the obvious).
with sack, scanning the retransmit queue for timed out packets is
fine because we know which packets in our retransmit queue have been
acked by the receiver.
without sack, we know only how many packets in our retransmit queue the
receiver has acknowledged, but no idea which packets.
think of a "typical" slow-start overshoot case, where for example
every third packet in a window get lost because a router buffer gets
full.
with sack, we check for timeouts on those every third packet (as the
rest have been sacked). the packet counting works out and if there
is no reordering, we'll retransmit exactly the packets that were
lost.
without sack, however, we check for timeout on every packet and end up
retransmitting consecutive packets in the retransmit queue. in our
slow-start example, 2/3 of those retransmissions are unnecessary. these
unnecessary retransmissions eat the congestion window and evetually
prevent fast recovery from continuing, if enough packets were lost.
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrea Bittau [Mon, 12 Jun 2006 03:58:33 +0000 (20:58 -0700)]
[DCCP] Ackvec: fix soft lockup in ackvec handling code
A soft lockup existed in the handling of ack vector records.
Specifically, when a tail of the list of ack vector records was
removed, it was possible to end up iterating infinitely on an element
of the tail.
Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk> Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz> Signed-off-by: David S. Miller <davem@davemloft.net>
Paul Mackerras [Mon, 12 Jun 2006 02:16:26 +0000 (12:16 +1000)]
[PATCH] Fix for the PPTP hangs that have been reported
People have been reporting that PPP connections over ptys, such as
used with PPTP, will hang randomly when transferring large amounts of
data, for instance in http://bugzilla.kernel.org/show_bug.cgi?id=6530.
I have managed to reproduce the problem, and the patch below fixes the
actual cause.
The problem is not in fact in ppp_async.c but in n_tty.c. What
happens is that when pptp reads from the pty, we call read_chan() in
drivers/char/n_tty.c on the master side of the pty. That copies all
the characters out of its buffer to userspace and then calls
check_unthrottle(), which calls the pty unthrottle routine, which
calls tty_wakeup on the slave side, which calls ppp_asynctty_wakeup,
which calls tasklet_schedule. So far so good. Since we are in
process context, the tasklet runs immediately and calls
ppp_async_process(), which calls ppp_async_push, which calls the
tty->driver->write function to send some more output.
However, tty->driver->write() returns zero, because the master
tty->receive_room is still zero. We haven't returned from
check_unthrottle() yet, and read_chan() only updates tty->receive_room
_after_ calling check_unthrottle. That means that the driver->write
call in ppp_async_process() returns 0. That would be fine if we were
going to get a subsequent wakeup call, but we aren't (we just had it,
and the buffer is now empty).
The solution is for n_tty.c to update tty->receive_room _before_
calling the driver unthrottle routine. The patch below does this.
With this patch I was able to transfer a 900MB file over a PPTP
connection (taking about 25 minutes), whereas without the patch the
connection would always stall in under a minute.
Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Bug fix: mv_eng_timeout() calls mv_err_intr() without first grabbing the host lock,
which can lead to all sorts of interesting scenarios.
This whole error-handling portion of sata_mv is nasty (and will get fixed for
the new EH stuff), but for now this patch will help keep it on life-support.
Signed-off-by: Mark Lord <liml@rtr.ca> Signed-off-by: Jeff Garzik <jeff@garzik.org>
Dave Jones [Wed, 19 Apr 2006 04:06:51 +0000 (21:06 -0700)]
[PATCH] PCI: Improve PCI config space writeback
At least one laptop blew up on resume from suspend with a black screen due
to a lack of this patch. By only writing back config space that is
different, we minimise the possibility of accidents like this.
Signed-off-by: Dave Jones <davej@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Jean Delvare [Tue, 18 Apr 2006 12:49:56 +0000 (14:49 +0200)]
[PATCH] PCI: Error handling on PCI device resume
We currently don't handle errors properly when resuming a PCI device:
* In pci_default_resume() we capture the error code returned by
pci_enable_device() but don't pass it up to the caller.
Introduced by commit 95a629657dbe28e44a312c47815b3dc3f1ce0970
* In pci_resume_device(), the errors possibly returned by the driver's
.resume method or by the generic pci_default_resume() function are
ignored.
This patch fixes both issues.
Signed-off-by: Jean Delvare <khali@linux-fr.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Krzysztof Helt [Sun, 11 Jun 2006 05:03:43 +0000 (22:03 -0700)]
[SPARC]: Migration cost tune up in sparc smp.
This patch sets the max_cache_size value required to tune up
scheduler in SMP systems. Otherwise, the calculated
migration_cost is too high and task scheduling may lock up.
Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl> Signed-off-by: David S. Miller <davem@davemloft.net>
* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
[SPARC64]: Avoid JBUS errors on some Niagara systems.
[FUSION]: Fix mptspi.c build with CONFIG_PM not set.
[TG3]: Handle Sun onboard tg3 chips more correctly.
[SPARC64]: Dump local cpu registers in sun4v_log_error()
Milton Miller [Sat, 10 Jun 2006 16:54:16 +0000 (09:54 -0700)]
[PATCH] powerpc: console_initcall ordering issues
From: Milton Miller <miltonm@bga.com>
The add_preferred_console call in rtas_console.c was not causing the
console to be selected. It turns out that the add_preferred_console was
being called after the hvc_console driver was registered. It only works
when it is called before the console driver is registered.
Reorder hvc_console.o after the hvc_console drivers to allow the selection
during console_initcall processing.
Signed-off-by: Milton Miller <miltonm@bga.com> Signed-off-by: Anton Blanchard <anton@samba.org> Cc: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
David Howells [Sat, 10 Jun 2006 16:54:12 +0000 (09:54 -0700)]
[PATCH] Further alterations for memory barrier document
From: David Howells <dhowells@redhat.com>
Apply some alterations to the memory barrier document that I worked out
with Paul McKenney of IBM, plus some of the alterations suggested by Alan
Stern.
The following changes were made:
(*) One of the examples given for what can happen with overlapping memory
barriers was wrong.
(*) The description of general memory barriers said that a general barrier is
a combination of a read barrier and a write barrier. This isn't entirely
true: it implies both, but is more than a combination of both.
(*) The first example in the "SMP Barrier Pairing" section was wrong: the
loads around the read barrier need to touch the memory locations in the
opposite order to the stores around the write barrier.
(*) Added a note to make explicit that the loads should be in reverse order to
the stores.
(*) Adjusted the diagrams in the "Examples Of Memory Barrier Sequences"
section to make them clearer. Added a couple of diagrams to make it more
clear as to how it could go wrong without the barrier.
(*) Added a section on memory speculation.
(*) Dropped any references to memory allocation routines doing memory
barriers. They may do sometimes, but it can't be relied on. This may be
worthy of further documentation later.
(*) Made the fact that a LOCK followed by an UNLOCK should not be considered a
full memory barrier more explicit and gave an example.
Signed-off-by: David Howells <dhowells@redhat.com> Acked-by: Paul E. McKenney <paulmck@us.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Russell King [Sat, 10 Jun 2006 11:42:12 +0000 (12:42 +0100)]
[ARM] Fix Integrator and Versatile interrupt initialisation
Both Integrator and Versatile were using set_irq_handler() and
enable_irq(), and working around the initialisation of the
chained interrupt, instead of the more correct
set_irq_chained_handler() function. Fix Integrator and
Versatile to use the right function, and remove these work-arounds.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
David S. Miller [Sat, 10 Jun 2006 08:06:25 +0000 (01:06 -0700)]
[SPARC64]: Avoid JBUS errors on some Niagara systems.
Doing PCI config space accesses to non-present PCI slots
can result in fatal JBUS errors if the PCI config access
hypervisor call is performed on cpus other than the boot
cpu.
PCI config space accesses to present PCI slots works just
fine.
Recursively traverse the OBP device tree under the PCI
controller node and record all present device IDs into
a small hash table.
Avoid the hypervisor call for any PCI config space access
attempt for a device not recorded in the hash table.
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller [Fri, 9 Jun 2006 18:58:36 +0000 (11:58 -0700)]
[TG3]: Handle Sun onboard tg3 chips more correctly.
Get rid of all the SUN_570X logic and instead:
1) Make sure MEMARB_ENABLE is set when we probe the SRAM
for config information. If that is off we will get
timeouts.
2) Always try to sync with the firmware, if there is no
firmware running do not treat it as an error and instead
just report it the first time we notice this condition.
3) If there is no valid SRAM signature, assume the device
is onboard by setting TG3_FLAG_EEPROM_WRITE_PROT.
Update driver version and release date.
With help from Michael Chan and Fabio Massimo Di Nitto.
Signed-off-by: David S. Miller <davem@davemloft.net>
Arnd Bergmann [Wed, 7 Jun 2006 02:04:18 +0000 (12:04 +1000)]
[PATCH] powerpc: Fix cell blade detection
The IBM Cell blade firmware might confuse the kernel to think it's a
pSeries machine. This fixes it for now. With a bit of luck, the firmware
will be updated to avoid that in the future but currently that patch is
needed.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
[PATCH] powerpc: Fix call to ibm,client-architecture-support
The code in prom_init.c calling the firmware
ibm,client-architecture-support method on pSeries has a bug where it
fails to properly pass the instance handle of the firmware object when
trying to call a method. Result ranges from the call doing nothing to
the firmware crashing. (Found by Segher, thanks !)
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Segher Boessenkool <segher@kernel.crashing.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
Paul Mackerras [Fri, 9 Jun 2006 03:02:59 +0000 (13:02 +1000)]
powerpc: Fix machine check problem on 32-bit kernels
This fixes a bug found by Dave Jones that means that it is possible
for userspace to provoke a machine check on 32-bit kernels. This
also fixes a couple of other places where I found similar problems
by inspection.
__futex_atomic_op needs to do an atomic operation in the user address space,
not the kernel address space. Add the missing sacf 256/sacf 0 to switch to
the secondary mode before doing the compare-and-swap. In addition add
another fixup for catch specification exceptions if the compare-and-swap
address is not aligned.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Jens Axboe [Thu, 8 Jun 2006 08:26:39 +0000 (10:26 +0200)]
[PATCH] debugfs inode leak
Looking at the reiser4 crash, I found a leak in debugfs. In
debugfs_mknod(), we create the inode before checking if the dentry
already has one attached. We don't free it if that is the case.
These bugs happen quite often, I'm starting to think we should disallow
such coding in CodingStyle.
Jens Axboe [Thu, 8 Jun 2006 06:49:06 +0000 (08:49 +0200)]
[PATCH] elevator switching race
There's a race between shutting down one io scheduler and firing up the
next, in which a new io could enter and cause the io scheduler to be
invoked with bad or NULL data.
To fix this, we need to maintain the queue lock for a bit longer.
Unfortunately we cannot do that, since the elevator init requires to be
run without the lock held. This isn't easily fixable, without also
changing the mempool API. So split the initialization into two parts,
and alloc-init operation and an attach operation. Then we can
preallocate the io scheduler and related structures, and run the attach
inside the lock after we detach the old one.
This patch has survived 30 minutes of 1 second io scheduler switching
with a very busy io load.
projects / linux-2.6 / log