openwrt/staging/blogic.git
17 years agonetpoll: drivers must not enable IRQ unconditionally in their NAPI handler
Francois Romieu [Sun, 17 Dec 2006 22:03:15 +0000 (23:03 +0100)]
netpoll: drivers must not enable IRQ unconditionally in their NAPI handler

net/core/netpoll.c::netpoll_send_skb() calls the poll handler when
it is available. As netconsole can be used from almost any context,
IRQ must not be enabled blindly in the NAPI handler of a driver which
supports netpoll.

b57bd06655a028aba7b92e1c19c2093e7fcfb341 fixed the issue for the
8139too.c driver.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: No-delay link detection at interface up
Jesse Brandeburg [Fri, 15 Dec 2006 09:42:34 +0000 (10:42 +0100)]
[PATCH] e1000: No-delay link detection at interface up

Currently after an interface up, the link state is detected 2 seconds later
when the first watchdog timer runs. This patch changes that by triggering
the hardware to generate a link-change interrupt from the up() function
instead. This has the result that the link state gets detected immediately
and without races. This has the potential to speed up booting since a normal
distribution boot process waits for a link before DHCP is attempted.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoe1000: 3 new driver stats for managability testing
Jeff Garzik [Fri, 15 Dec 2006 16:16:33 +0000 (11:16 -0500)]
e1000: 3 new driver stats for managability testing

Add 3 extra packet redirect counters for tracking purposes to make sure
we can test that all packets arrive properly.

Originally from Jesse Brandeburg <jesse.brandeburg@intel.com>,
rewritten to use feature flags by me.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: Make the copybreak value a module parameter
Jesse Brandeburg [Fri, 15 Dec 2006 09:40:39 +0000 (10:40 +0100)]
[PATCH] e1000: Make the copybreak value a module parameter

Allow the user to vary the size that copybreak works. Currently cb is enabled
for packets < 256 bytes, but various tests indicate that this should be
configurable for specific use cases. In addition, this parameter allows us
to force never/always during testing to get full and predictable coverage of
both code paths.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: Fix PBA allocation calculations
Bruce Allan [Fri, 15 Dec 2006 09:39:45 +0000 (10:39 +0100)]
[PATCH] e1000: Fix PBA allocation calculations

Assign the PBA to be large enough to contain at least 2 jumbo frames on
all adapters. This dramatically increases performance on several adapters
and fixes TX performance degradation issues where the PBA was misallocated
in the old algorithm.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: narrow down the scope of the tipg timer tweak
Jesse Brandeburg [Fri, 15 Dec 2006 09:38:32 +0000 (10:38 +0100)]
[PATCH] e1000: narrow down the scope of the tipg timer tweak

the driver has (ancient) code for messing with TIPG from the 82542 days.
Unfortunately this code was running on our current adapters and setting
TIPG for fiber to be +1 over the copper value.  This caused 1.45Mpps
to be sent instead of 1.487Mpps.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: fix ethtool reported bus type for older adapters
Jeff Kirsher [Fri, 15 Dec 2006 09:37:32 +0000 (10:37 +0100)]
[PATCH] e1000: fix ethtool reported bus type for older adapters

For older adapters we know that they are of the PCI bus type, so we can
just set this.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: fix to set the new max frame size before resetting the adapter
Bruce Allan [Fri, 15 Dec 2006 09:36:35 +0000 (10:36 +0100)]
[PATCH] e1000: fix to set the new max frame size before resetting the adapter

This bugfix makes sure that the driver data reflects the full new situation
before the adapter is reinitialized.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoe1000: workaround for the ESB2 NIC RX unit issue
Jeff Garzik [Fri, 15 Dec 2006 16:06:17 +0000 (11:06 -0500)]
e1000: workaround for the ESB2 NIC RX unit issue

In rare occasions, ESB2 systems would end up started without the RX
unit being turned on. Add a check that runs post-init to work around
this issue.

Originally from Jesse Brandeburg <jesse.brandeburg@intel.com>,
rewritten to use feature flags by me.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: disable TSO on the 82544 with slab debugging
Jesse Brandeburg [Fri, 15 Dec 2006 09:34:46 +0000 (10:34 +0100)]
[PATCH] e1000: disable TSO on the 82544 with slab debugging

CONFIG_DEBUG_SLAB changes alignments of the data structures the slab
allocators return. These break certain workarounds for TSO on the 82544.
Since DEBUG_SLAB is relatively rare and not used for performance sensitive
cases, the simplest fix is to disable TSO in this special situation.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: Fix Wake-on-Lan with forced gigabit speed
Jesse Brandeburg [Fri, 15 Dec 2006 09:33:46 +0000 (10:33 +0100)]
[PATCH] e1000: Fix Wake-on-Lan with forced gigabit speed

If the user has forced gigabit speed, phy power management must be disabled;
otherwise the NIC would try to negotiate to a linkspeed of 10/100 mbit on
shutdown, which would lead to a total loss of link. This loss of link breaks
Wake-on-Lan and IPMI.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoe1000: consolidate managability enabling/disabling
Jeff Garzik [Fri, 15 Dec 2006 15:56:10 +0000 (10:56 -0500)]
e1000: consolidate managability enabling/disabling

Several bugs existed in how we handle manageability issues all
over the driver.  This patch consolidates all the managability
release and init code in two single functions and call them from
appropriate locations. This fixes several BMC packet redirect issues
and powerup/down hiccups.

Originally from Jesse Brandeburg <jesse.brandeburg@intel.com>, rewritten
to use feature flags by me.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoe1000: omit stats for broken counter in 82543
Jeff Garzik [Fri, 15 Dec 2006 15:41:15 +0000 (10:41 -0500)]
e1000: omit stats for broken counter in 82543

The 82543 chip does not count tx_carrier_errors properly in FD mode;
report zeros instead of garbage.

Originally from Jesse Brandeburg <jesse.brandeburg@intel.com>, rewritten
to use feature flags by me.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoe1000: For sanity, reformat e1000_set_mac_type(), struct e1000_hw[_stats]
Jeff Garzik [Fri, 15 Dec 2006 15:31:40 +0000 (10:31 -0500)]
e1000: For sanity, reformat e1000_set_mac_type(), struct e1000_hw[_stats]

Makes future changes a bit more readable.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: dynamic itr: take TSO and jumbo into account
Jesse Brandeburg [Fri, 15 Dec 2006 09:30:44 +0000 (10:30 +0100)]
[PATCH] e1000: dynamic itr: take TSO and jumbo into account

The dynamic interrupt rate control patches omitted proper counting
for jumbo's and TSO resulting in suboptimal interrupt mitigation strategies.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years ago[PATCH] e1000: The user-supplied itr setting needs the lower 2 bits masked off
Jesse Brandeburg [Fri, 15 Dec 2006 09:29:31 +0000 (10:29 +0100)]
[PATCH] e1000: The user-supplied itr setting needs the lower 2 bits masked off

The lower 2 bits of a user-supplied itr setting (via ethtool) need to be
masked off: These lower two bits are used as control bits.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoLinux 2.6.20-rc2
Linus Torvalds [Sun, 24 Dec 2006 04:00:32 +0000 (20:00 -0800)]
Linux 2.6.20-rc2

17 years agoFix up CIFS for "test_clear_page_dirty()" removal
Linus Torvalds [Sun, 24 Dec 2006 00:19:07 +0000 (16:19 -0800)]
Fix up CIFS for "test_clear_page_dirty()" removal

This also adds he required page "writeback" flag handling, that cifs
hasn't been doing and that the page dirty flag changes made obvious.

Acked-by: Steve French <smfltc@us.ibm.com>
Acked-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] arch/i386/pci/mmconfig.c tlb flush fix
OGAWA Hirofumi [Sat, 23 Dec 2006 01:00:43 +0000 (10:00 +0900)]
[PATCH] arch/i386/pci/mmconfig.c tlb flush fix

We use the fixmap for accessing pci config space in pci_mmcfg_read/write().
The problem is in pci_exp_set_dev_base(). It is caching a last
accessed address to avoid calling set_fixmap_nocache() whenever
pci_mmcfg_read/write() is used.

  static inline void pci_exp_set_dev_base(int bus, int devfn)
  {
u32 dev_base = base | (bus << 20) | (devfn << 12);
if (dev_base != mmcfg_last_accessed_device) {
mmcfg_last_accessed_device = dev_base;
set_fixmap_nocache(FIX_PCIE_MCFG, dev_base);
}
  }

            cpu0                                        cpu1
  ---------------------------------------------------------------------------
    pci_mmcfg_read("device-A")
        pci_exp_set_dev_base()
            set_fixmap_nocache()
                                              pci_mmcfg_read("device-B")
                                                  pci_exp_set_dev_base()
                                                      set_fixmap_nocache()
    pci_mmcfg_read("device-B")
        pci_exp_set_dev_base()
            /* doesn't flush tlb */

But if cpus accessed the above order, the second pci_mmcfg_read() on
cpu0 doesn't flush the TLB, because "mmcfg_last_accessed_device" is
device-B.  So, second pci_mmcfg_read() on cpu0 accesses a device-A via
a previous TLB cache. This problem became the cause of several strange
behavior.

This patches fixes this situation by adds "mmcfg_last_accessed_cpu" check.

[ Alternatively, we could make a per-cpu mapping area or something. Not
  that it's probably worth it, but if we wanted to avoid all locking and
  instead just disable preemption, that would be the way to go. --Linus ]

Signed-off-by: OGAWA Hirofumi <hogawa@miraclelinux.com>
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] suspend: fix suspend on single-CPU systems
Ingo Molnar [Sat, 23 Dec 2006 15:55:29 +0000 (16:55 +0100)]
[PATCH] suspend: fix suspend on single-CPU systems

Clark Williams reported that suspend doesnt work on his laptop on
2.6.20-rc1-rt kernels. The bug was introduced by the following cleanup
commit:

 commit 112cecb2cc0e7341db92281ba04b26c41bb8146d
 Author: Siddha, Suresh B <suresh.b.siddha@intel.com>
 Date:   Wed Dec 6 20:34:31 2006 -0800

    [PATCH] suspend: don't change cpus_allowed for task initiating the suspend

because with this change 'error' is not initialized to 0 anymore, if
there are no other online CPUs. (i.e. if the system is single-CPU).

the fix is the initialize it to 0. The really weird thing is that my
version of gcc does not warn about this non-initialized variable
situation ...

(also fix the kernel printk in the error branch, it was missing a
 newline)

Reported-by: Clark Williams <williams@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoFix reiserfs after "test_clear_page_dirty()" removal
Linus Torvalds [Sat, 23 Dec 2006 17:32:45 +0000 (09:32 -0800)]
Fix reiserfs after "test_clear_page_dirty()" removal

Thanks to Len Brown for testing this fix, since while they have in the
past, none of my machines run reiserfs at the moment.

Cc: Vladimir V. Saveliev <vs@namesys.com>
Acked-by: Len Brown <lenb@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoClean up and export cancel_dirty_page() to modules
Linus Torvalds [Sat, 23 Dec 2006 17:25:04 +0000 (09:25 -0800)]
Clean up and export cancel_dirty_page() to modules

Make cancel_dirty_page() act more like all the other dirty and writeback
accounting functions: test for "mapping" being NULL, and do the
NR_FILE_DIRY accounting purely based on mapping_cap_account_dirty()).

Also, add it to the exports, so that modular filesystems can use it.

Acked-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Sat, 23 Dec 2006 02:46:56 +0000 (18:46 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6: (68 commits)
  ACPI: replace kmalloc+memset with kzalloc
  ACPI: Add support for acpi_load_table/acpi_unload_table_id
  fbdev: update after backlight argument change
  ACPI: video: Add dev argument for backlight_device_register
  ACPI: Implement acpi_video_get_next_level()
  ACPI: Kconfig - depend on PM rather than selecting it
  ACPI: fix NULL check in drivers/acpi/osl.c
  ACPI: make drivers/acpi/ec.c:ec_ecdt static
  ACPI: prevent processor module from loading on failures
  ACPI: fix single linked list manipulation
  ACPI: ibm_acpi: allow clean removal
  ACPI: fix git automerge failure
  ACPI: ibm_acpi: respond to workqueue update
  ACPI: dock: add uevent to indicate change in device status
  ACPI: ec: Lindent once again
  ACPI: ec: Change #define to enums there possible.
  ACPI: ec: Style changes.
  ACPI: ec: Acquire Global Lock under EC mutex.
  ACPI: ec: Drop udelay() from poll mode. Loop by reading status field instead.
  ACPI: ec: Rename gpe_bit to gpe
  ...

17 years ago[PATCH] Call init_timer() for ISDN PPP CCP reset state timer
Marcel Holtmann [Thu, 21 Dec 2006 22:06:24 +0000 (23:06 +0100)]
[PATCH] Call init_timer() for ISDN PPP CCP reset state timer

The function isdn_ppp_ccp_reset_alloc_state() sets ->timer.function
and ->timer.data and later on calls add_timer() with no init_timer()
ever done.

Noted by Al Viro.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Karsten Keil <kkeil@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Fri, 22 Dec 2006 22:14:17 +0000 (14:14 -0800)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [UDP]: Fix reversed logic in udp_get_port().
  [IPV6]: Dumb typo in generic csum_ipv6_magic()
  [SCTP]: make 2 functions static
  [SCTP]: Fix typo adaption -> adaptation as per the latest API draft.
  [SCTP]: Don't export include/linux/sctp.h to userspace.
  [TCP]: Fix ambiguity in the `before' relation.
  [ATM] drivers/atm/fore200e.c: Cleanups.
  [ATM]: Remove dead ATM_TNETA1570 option.
  NetLabel: correctly fill in unused CIPSOv4 level and category mappings
  NetLabel: perform input validation earlier on CIPSOv4 DOI add ops

17 years ago[PATCH] cfq-iosched: tighten allow merge criteria
Jens Axboe [Fri, 22 Dec 2006 08:38:53 +0000 (09:38 +0100)]
[PATCH] cfq-iosched: tighten allow merge criteria

The logic in cfq_allow_merge() wasn't clear enough - basically allow
merging for the same queues only.  Do a fast check for 'rq and bio both
sync/async' before doing the cfqq hash lookup.

This is verified to work with the fixed elv_try_merge() from commit
bb4067e34159648d394943d5e2a011f838bff22f.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[UDP]: Fix reversed logic in udp_get_port().
David S. Miller [Fri, 22 Dec 2006 19:42:26 +0000 (11:42 -0800)]
[UDP]: Fix reversed logic in udp_get_port().

When this code was converted to use sk_for_each() the
logic for the "best hash chain length" code was reversed,
breaking everything.

The original code was of the form:

size = 0;
do {
if (++size >= best_size_so_far)
goto next;
} while ((sk = sk->next) != NULL);
best_size_so_far = size;
best = result;
next:;

and this got converted into:

sk_for_each(sk2, node, head)
if (++size < best_size_so_far) {
best_size_so_far = size;
best = result;
}

Which does something very very different from the original.

Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[IPV6]: Dumb typo in generic csum_ipv6_magic()
Al Viro [Thu, 21 Dec 2006 21:15:18 +0000 (13:15 -0800)]
[IPV6]: Dumb typo in generic csum_ipv6_magic()

... duh

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[SCTP]: make 2 functions static
Adrian Bunk [Thu, 21 Dec 2006 00:08:22 +0000 (16:08 -0800)]
[SCTP]: make 2 functions static

This patch makes the following needlessly global functions static:
- ipv6.c: sctp_inet6addr_event()
- protocol.c: sctp_inetaddr_event()

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[SCTP]: Fix typo adaption -> adaptation as per the latest API draft.
Ivan Skytte Jorgensen [Thu, 21 Dec 2006 00:07:04 +0000 (16:07 -0800)]
[SCTP]: Fix typo adaption -> adaptation as per the latest API draft.

Signed-off-by: Ivan Skytte Jorgensen <isj-sctp@i1.dk>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[SCTP]: Don't export include/linux/sctp.h to userspace.
Sridhar Samudrala [Thu, 21 Dec 2006 00:06:09 +0000 (16:06 -0800)]
[SCTP]: Don't export include/linux/sctp.h to userspace.

This file contains protocol definitions and there are no SCTP apps
that use this file.

Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[TCP]: Fix ambiguity in the `before' relation.
Gerrit Renker [Wed, 20 Dec 2006 18:25:55 +0000 (10:25 -0800)]
[TCP]: Fix ambiguity in the `before' relation.

While looking at DCCP sequence numbers, I stumbled over a problem with
the following definition of before in tcp.h:

static inline int before(__u32 seq1, __u32 seq2)
{
        return (__s32)(seq1-seq2) < 0;
}

Problem: This definition suffers from an an ambiguity, i.e. always

           before(a, (a + 2^31) % 2^32)) = 1
           before((a + 2^31) % 2^32), a) = 1

         In text: when the difference between a and b amounts to 2^31,
         a is always considered `before' b, the function can not decide.
         The reason is that implicitly 0 is `before' 1 ... 2^31-1 ... 2^31

Solution: There is a simple fix, by defining before in such a way that
          0 is no longer `before' 2^31, i.e. 0 `before' 1 ... 2^31-1
          By not using the middle between 0 and 2^32, before can be made
          unambiguous.
          This is achieved by testing whether seq2-seq1 > 0 (using signed
          32-bit arithmetic).

I attach a patch to codify this. Also the `after' relation is basically
a redefinition of `before', it is now defined as a macro after before.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM] drivers/atm/fore200e.c: Cleanups.
Adrian Bunk [Wed, 20 Dec 2006 03:36:32 +0000 (19:36 -0800)]
[ATM] drivers/atm/fore200e.c: Cleanups.

This patch contains the following transformations from custom functions
to standard kernel version:
- fore200e_kmalloc() -> kzalloc()
- fore200e_kfree() -> kfree()
- fore200e_swap() -> cpu_to_be32()

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[ATM]: Remove dead ATM_TNETA1570 option.
Adrian Bunk [Wed, 20 Dec 2006 03:35:05 +0000 (19:35 -0800)]
[ATM]: Remove dead ATM_TNETA1570 option.

This patch removes the unconverted ATM_TNETA1570 option that also lacks
any code in the kernel.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoNetLabel: correctly fill in unused CIPSOv4 level and category mappings
Paul Moore [Fri, 15 Dec 2006 21:49:28 +0000 (16:49 -0500)]
NetLabel: correctly fill in unused CIPSOv4 level and category mappings

Back when the original NetLabel patches were being changed to use Netlink
attributes correctly some code was accidentially dropped which set all of the
undefined CIPSOv4 level and category mappings to a sentinel value.  The result
is the mappings data in the kernel contains bogus mappings which always map to
zero.  This patch restores the old/correct behavior by initializing the mapping
data to the correct sentinel value.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Signed-off-by: James Morris <jmorris@namei.org>
17 years agoNetLabel: perform input validation earlier on CIPSOv4 DOI add ops
Paul Moore [Fri, 15 Dec 2006 21:49:27 +0000 (16:49 -0500)]
NetLabel: perform input validation earlier on CIPSOv4 DOI add ops

There are a couple of cases where the user input for a CIPSOv4 DOI add
operation was not being done soon enough; the result was unexpected behavior
which was resulting in oops/panics/lockups on some platforms.  This patch moves
the existing input validation code earlier in the code path to protect against
bogus user input.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Signed-off-by: James Morris <jmorris@namei.org>
17 years ago[PATCH] Fix up page_mkclean_one(): virtual caches, s390
Peter Zijlstra [Fri, 22 Dec 2006 13:25:52 +0000 (14:25 +0100)]
[PATCH] Fix up page_mkclean_one(): virtual caches, s390

 - add flush_cache_page() for all those virtual indexed cache
   architectures.

 - handle s390.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] serial/uartlite: Only enable port if request_port succeeded
Peter Korsgaard [Fri, 22 Dec 2006 15:38:40 +0000 (16:38 +0100)]
[PATCH] serial/uartlite: Only enable port if request_port succeeded

The uartlite driver used to always enable the port even if request_port
failed causing havoc. This patch fixes it.

Signed-off-by: Peter Korsgaard <jacmet@sunsite.dk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix reparenting to the same thread group. (take 2)
Eric W. Biederman [Fri, 22 Dec 2006 04:28:40 +0000 (21:28 -0700)]
[PATCH] Fix reparenting to the same thread group. (take 2)

This patch fixes the case when we reparent to a different thread in the
same thread group.  This modifies the code so that we do not send
signals and do not change the signal to send to SIGCHLD unless we have
change the thread group of our parents.  It also suppresses sending
pdeath_sig in this cas as well since the result of geppid doesn't
change.

Thanks to Oleg for spotting my bug of only fixing this for non-ptraced
tasks.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Albert Cahalan <acahalan@gmail.com>
Cc: Andrew Morton <akpm@osdl.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Coywolf Qi Hunt <qiyong@fc-cn.com>
Acked-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] build compile.h earlier
Andrew Morton [Fri, 22 Dec 2006 09:12:01 +0000 (01:12 -0800)]
[PATCH] build compile.h earlier

compile.h is created super-late in the build.  But proc_misc.c want to include
it, and it's generally not sane to have a header file in include/linux be
created at the end of the build: it's either not present or, worse, wrong for
most of the build.

So the patch arranges for compile.h to be built at the start of the build
process.  It also consolidates the compile.h rules with those for version.h
and utsname.h, so they all get built together.

I hope.  My chances of having got this right are about 2%.

Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: fix bad missed wakeups in the i386, x86_64, ia64, ACPI and APM idle...
Ingo Molnar [Fri, 22 Dec 2006 09:11:56 +0000 (01:11 -0800)]
[PATCH] sched: fix bad missed wakeups in the i386, x86_64, ia64, ACPI and APM idle code

Fernando Lopez-Lezcano reported frequent scheduling latencies and audio
xruns starting at the 2.6.18-rt kernel, and those problems persisted all
until current -rt kernels. The latencies were serious and unjustified by
system load, often in the milliseconds range.

After a patient and heroic multi-month effort of Fernando, where he
tested dozens of kernels, tried various configs, boot options,
test-patches of mine and provided latency traces of those incidents, the
following 'smoking gun' trace was captured by him:

                 _------=> CPU#
                / _-----=> irqs-off
               | / _----=> need-resched
               || / _---=> hardirq/softirq
               ||| / _--=> preempt-depth
               |||| /
               |||||     delay
   cmd     pid ||||| time  |   caller
      \   /    |||||   \   |   /
  IRQ_19-1479  1D..1    0us : __trace_start_sched_wakeup (try_to_wake_up)
  IRQ_19-1479  1D..1    0us : __trace_start_sched_wakeup <<...>-5856> (37 0)
  IRQ_19-1479  1D..1    0us : __trace_start_sched_wakeup (c01262ba 0 0)
  IRQ_19-1479  1D..1    0us : resched_task (try_to_wake_up)
  IRQ_19-1479  1D..1    0us : __spin_unlock_irqrestore (try_to_wake_up)
  ...
  <idle>-0     1...1   11us!: default_idle (cpu_idle)
  ...
  <idle>-0     0Dn.1  602us : smp_apic_timer_interrupt (c0103baf 1 0)
  ...
   <...>-5856  0D..2  618us : __switch_to (__schedule)
   <...>-5856  0D..2  618us : __schedule <<idle>-0> (20 162)
   <...>-5856  0D..2  619us : __spin_unlock_irq (__schedule)
   <...>-5856  0...1  619us : trace_stop_sched_switched (__schedule)
   <...>-5856  0D..1  619us : trace_stop_sched_switched <<...>-5856> (37 0)

what is visible in this trace is that CPU#1 ran try_to_wake_up() for
PID:5856, it placed PID:5856 on CPU#0's runqueue and ran resched_task()
for CPU#0. But it decided to not send an IPI that no CPU - due to
TS_POLLING. But CPU#0 never woke up after its NEED_RESCHED bit was set,
and only rescheduled to PID:5856 upon the next lapic timer IRQ. The
result was a 600+ usecs latency and a missed wakeup!

the bug turned out to be an idle-wakeup bug introduced into the mainline
kernel this summer via an optimization in the x86_64 tree:

    commit 495ab9c045e1b0e5c82951b762257fe1c9d81564
    Author: Andi Kleen <ak@suse.de>
    Date:   Mon Jun 26 13:59:11 2006 +0200

    [PATCH] i386/x86-64/ia64: Move polling flag into thread_info_status

    During some profiling I noticed that default_idle causes a lot of
    memory traffic. I think that is caused by the atomic operations
    to clear/set the polling flag in thread_info. There is actually
    no reason to make this atomic - only the idle thread does it
    to itself, other CPUs only read it. So I moved it into ti->status.

the problem is this type of change:

        if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
-               clear_thread_flag(TIF_POLLING_NRFLAG);
+               current_thread_info()->status &= ~TS_POLLING;
                smp_mb__after_clear_bit();
                while (!need_resched()) {
                        local_irq_disable();

this changes clear_thread_flag() to an explicit clearing of TS_POLLING.
clear_thread_flag() is defined as:

        clear_bit(flag, &ti->flags);

and clear_bit() is a LOCK-ed atomic instruction on all x86 platforms:

  static inline void clear_bit(int nr, volatile unsigned long * addr)
  {
          __asm__ __volatile__( LOCK_PREFIX
                  "btrl %1,%0"

hence smp_mb__after_clear_bit() is defined as a simple compile barrier:

  #define smp_mb__after_clear_bit()       barrier()

but the explicit TS_POLLING clearing introduced by the patch:

+               current_thread_info()->status &= ~TS_POLLING;

is not an atomic op! So the clearing of the TS_POLLING bit is freely
reorderable with the reading of the NEED_RESCHED bit - and both now
reside in different memory addresses.

CPU idle wakeup very much depends on ordered memory ops, the clearing of
the TS_POLLING flag must always be done before we test need_resched()
and hit the idle instruction(s). [Symmetrically, the wakeup code needs
to set NEED_RESCHED before it tests the TS_POLLING flag, so memory
ordering is paramount.]

Fernando's dual-core Athlon64 system has a sufficiently advanced memory
ordering model so that it triggered this scenario very often.

( And it also turned out that the reason why these latencies never
  triggered on my testsystems is that i routinely use idle=poll, which
  was the only idle variant not affected by this bug. )

The fix is to change the smp_mb__after_clear_bit() to an smp_mb(), to
act as an absolute barrier between the TS_POLLING write and the
NEED_RESCHED read. This affects almost all idling methods (default,
ACPI, APM), on all 3 x86 architectures: i386, x86_64, ia64.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Tested-by: Fernando Lopez-Lezcano <nando@ccrma.Stanford.EDU>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] jbd: wait for already submitted t_sync_datalist buffer to complete
Hisashi Hifumi [Fri, 22 Dec 2006 09:11:50 +0000 (01:11 -0800)]
[PATCH] jbd: wait for already submitted t_sync_datalist buffer to complete

In the current jbd code, if a buffer on BJ_SyncData list is dirty and not
locked, the buffer is refiled to BJ_Locked list, submitted to the IO and
waited for IO completion.

But the fsstress test showed the case that when a buffer was already
submitted to the IO just before the buffer_dirty(bh) check, the buffer was
not waited for IO completion.

Following patch solves this problem.  If it is assumed that a buffer is
submitted to the IO before the buffer_dirty(bh) check and still being
written to disk, this buffer is refiled to BJ_Locked list.

Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Cc: Jan Kara <jack@ucw.cz>
Cc: "Stephen C. Tweedie" <sct@redhat.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix s3c24xx gpio driver (include linux/workqueue.h)
Ben Dooks [Fri, 22 Dec 2006 09:11:45 +0000 (01:11 -0800)]
[PATCH] fix s3c24xx gpio driver (include linux/workqueue.h)

The general gpio driver includes seem to now depend on having
<linux/workqueue.h> included before they are.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] md: fix a few problems with the interface (sysfs and ioctl) to md
NeilBrown [Fri, 22 Dec 2006 09:11:41 +0000 (01:11 -0800)]
[PATCH] md: fix a few problems with the interface (sysfs and ioctl) to md

While developing more functionality in mdadm I found some bugs in md...

- When we remove a device from an inactive array (write 'remove' to
  the 'state' sysfs file - see 'state_store') would should not
  update the superblock information - as we may not have
  read and processed it all properly yet.

- initialise all raid_disk entries to '-1' else the 'slot sysfs file
  will claim '0' for all devices in an array before the array is
  started.

- all '\n' not to be present at the end of words written to
  sysfs files

- when we use SET_ARRAY_INFO to set the md metadata version,
  set the flag to say that there is persistant metadata.

- allow GET_BITMAP_FILE to be called on an array that hasn't
  been started yet.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] increase CARDBUS_MEM_SIZE
Andrew Morton [Fri, 22 Dec 2006 09:11:36 +0000 (01:11 -0800)]
[PATCH] increase CARDBUS_MEM_SIZE

Linus sayeth:

Google knows everything, and finds, on MS own site no less:

  "Windows 2000 default resources:

   One 4K memory window

   One 2 MB memory window

   Two 256-byte I/O windows"

which is clearly utterly bogus and insufficient. But Microsoft apparently
realized this, and:

  "Windows XP default resources:

   Because one memory window of 4K and one window of 2 MB are not
   sufficient for CardBus controllers in many configurations, Windows XP
   allocates larger memory windows to CardBus controllers where possible.
   However, resource windows are static (that is, the operating system
   does not dynamically allocate larger memory windows if new devices
   appear.) Under Windows XP, CardBus controllers will be assigned the
   following resources:

   One 4K memory window, as in Windows 2000

   64 MB memory, if that amount of memory is available. If 64 MB is not
   available the controller will receive 32 MB; if 32 MB is not available,
   the controller will receive 16 MB; if 16 MB is not available, the
   bridge will receive 8 MB; and so on down to a minimum assignment of 1
   MB in configurations where memory is too constrained for the operating
   system to provide a larger window.

   Two 256-byte I/O windows"

So I think we have our answer. Windows uses one 4k window, and one 64MB
window. And they are no more dynamic than we are (we _could_ try to do it
dynamically, but let's face it, it's fairly painful to dynamically expand
PCI bus resources - you may need to reprogram everything up to the root,
so it would be absolutely crazy to do that unless you have some serious
masochistic tendencies).

So let's just increase our default value to 64M too.

Cc: Markus Rechberger <mrechberger@gmail.com>
Cc: Daniel Ritz <daniel.ritz@gmx.ch>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] relay: remove inlining
Andrew Morton [Fri, 22 Dec 2006 09:11:30 +0000 (01:11 -0800)]
[PATCH] relay: remove inlining

          text    data     bss     dec     hex filename
before:   4036      44       0    4080     ff0 kernel/relay.o
after:    3727      44       0    3771     ebb kernel/relay.o

Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Tom Zanussi <zanussi@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ptrace: Fix EFL_OFFSET value according to i386 pda changes
Jeremy Fitzhardinge [Fri, 22 Dec 2006 09:11:21 +0000 (01:11 -0800)]
[PATCH] ptrace: Fix EFL_OFFSET value according to i386 pda changes

The PDA patches introduced a bug in ptrace: it reads eflags from the wrong
place on the target's stack, but writes it back to the correct place.  The
result is a corrupted eflags, which is most visible when it turns interrupts
off unexpectedly.

This patch fixes this by making the ptrace code a little less fragile.  It
changes [gs]et_stack_long to take a straightforward byte offset into struct
pt_regs, rather than requiring all callers to do a sizeof(struct pt_regs)
offset adjustment.  This means that the eflag's offset (EFL_OFFSET) on the
target stack can be simply computed with offsetof().

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Frederik Deweerdt <deweerdt@free.fr>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] memory hotplug: fix compile error for i386 with NUMA config
Yasunori Goto [Fri, 22 Dec 2006 09:11:13 +0000 (01:11 -0800)]
[PATCH] memory hotplug: fix compile error for i386 with NUMA config

Fix compile error when config memory hotplug with numa on i386.

The cause of compile error was missing of arch_add_memory(),
remove_memory(), and memory_add_physaddr_to_nid().

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@cs.washington.edu>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] mips: if_fddi.h: Add a missing inclusion
Maciej W. Rozycki [Fri, 22 Dec 2006 09:11:04 +0000 (01:11 -0800)]
[PATCH] mips: if_fddi.h: Add a missing inclusion

This is a change to include <linux/netdevice.h> in <linux/if_fddi.h> which is
needed for "struct fddi_statistics".

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] kernel-doc: remove Martin from MAINTAINERS
Martin Waitz [Fri, 22 Dec 2006 09:10:56 +0000 (01:10 -0800)]
[PATCH] kernel-doc: remove Martin from MAINTAINERS

I don't have the time to work on Linux Documentation, so I really should
document that in MAINTAINERS.  With Randy, kernel-doc is in good hands
anyway.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] kernel-doc: allow unnamed structs/unions
Randy Dunlap [Fri, 22 Dec 2006 09:10:50 +0000 (01:10 -0800)]
[PATCH] kernel-doc: allow unnamed structs/unions

Make kernel-doc support unnamed (anonymous) structs and unions.  There is
one (union) in include/linux/skbuff.h (inside struct sk_buff) that is
currently generating a kernel-doc warning, so this fixes that warning.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fdtable: Provide free_fdtable() wrapper
Vadim Lobanov [Fri, 22 Dec 2006 09:10:43 +0000 (01:10 -0800)]
[PATCH] fdtable: Provide free_fdtable() wrapper

Christoph Hellwig has expressed concerns that the recent fdtable changes
expose the details of the RCU methodology used to release no-longer-used
fdtable structures to the rest of the kernel.  The trivial patch below
addresses these concerns by introducing the appropriate free_fdtable()
calls, which simply wrap the release RCU usage.  Since free_fdtable() is a
one-liner, it makes sense to promote it to an inline helper.

Signed-off-by: Vadim Lobanov <vlobanov@speakeasy.net>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] gxt4500: Fix colormap and PLL setting, support GXT6000P
Paul Mackerras [Fri, 22 Dec 2006 09:10:36 +0000 (01:10 -0800)]
[PATCH] gxt4500: Fix colormap and PLL setting, support GXT6000P

This fixes some bugs in the gxt4500 framebuffer driver, and adds support
for GXT6000P cards.

First, I had the red and blue channels swapped in the colormap update code,
resulting in penguins' noses and feet turning blue (though the penguins
weren't actually shivering :).

Secondly, the code that calculated the values to put in the PLL that
generates the pixel clock wasn't observing some constraints that I wasn't
originally aware of, but am now that I have some documentation on the chip.

The GXT6000P is essentially identical from software's point of view, except
for a different reference clock for the PLL, and the addition of a geometry
engine (which this driver doesn't use).

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: James Simmons <jsimmons@infradead.org>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] tlclk: delete unnecessary sysfs_remove_group
Akinobu Mita [Fri, 22 Dec 2006 09:10:28 +0000 (01:10 -0800)]
[PATCH] tlclk: delete unnecessary sysfs_remove_group

It is unnecessary and invalid to call sysfs_remove_group() after
sysfs_create_group() failure.

Cc: Sebastien Bouchard <sebastien.bouchard@ca.kontron.com>
Cc: Mark Gross <mark.gross@intel.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] MAINTAINERS: fix email for S3C2410 and S3C2440
Ben Dooks [Fri, 22 Dec 2006 09:10:23 +0000 (01:10 -0800)]
[PATCH] MAINTAINERS: fix email for S3C2410 and S3C2440

Change the email address for the S3C2410 and S3C2440 maintainer.  The old
addresses have been deleted due to spam issues.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] microcode: fix mc_cpu_notifier section warning
Jean Delvare [Fri, 22 Dec 2006 09:10:19 +0000 (01:10 -0800)]
[PATCH] microcode: fix mc_cpu_notifier section warning

Structure mc_cpu_notifier references a __cpuinit function, but isn't
declared __cpuinitdata itself:

WARNING: arch/i386/kernel/microcode.o - Section mismatch: reference
to .init.text: from .data after 'mc_cpu_notifier' (at offset 0x118)

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Tigran Aivazian <tigran@aivazian.fsnet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] schedule_timeout(): improve warning message
Andrew Morton [Fri, 22 Dec 2006 09:10:14 +0000 (01:10 -0800)]
[PATCH] schedule_timeout(): improve warning message

Kyle is hitting this warning, and we don't have a clue what it's caused by.
Add the obligatory dump_stack().

Cc: kyle <kylewong@southa.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] gss_spkm3: fix error handling in module init
Akinobu Mita [Fri, 22 Dec 2006 09:10:09 +0000 (01:10 -0800)]
[PATCH] gss_spkm3: fix error handling in module init

Return error and prevent from loading module when gss_mech_register()
failed.

Cc: Andy Adamson <andros@citi.umich.edu>
Cc: J. Bruce Fields <bfields@citi.umich.edu>
Acked-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] audit: fix kstrdup() error check
Akinobu Mita [Fri, 22 Dec 2006 09:10:02 +0000 (01:10 -0800)]
[PATCH] audit: fix kstrdup() error check

kstrdup() returns NULL on error.

Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] compile error of register_memory()
Yasunori Goto [Fri, 22 Dec 2006 09:09:54 +0000 (01:09 -0800)]
[PATCH] compile error of register_memory()

register_memory() becomes double definition in 2.6.20-rc1.  It is defined
in arch/i386/kernel/setup.c as static definition in 2.6.19.  But it is
moved to arch/i386/kernel/e820.c in 2.6.20-rc1.  And same name function is
defined in driver/base/memory.c too.  So, it becomes cause of compile error
of duplicate definition if memory hotplug option is on.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] handle SLOB with sparsemen
Yasunori Goto [Fri, 22 Dec 2006 09:09:44 +0000 (01:09 -0800)]
[PATCH] handle SLOB with sparsemen

This is to disallow to make SLOB with SMP or SPARSEMEM.  This avoids latent
troubles of SLOB with SLAB_DESTROY_BY_RCU.  And fix compile error.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] mm: more rmap debugging
Nick Piggin [Fri, 22 Dec 2006 09:09:33 +0000 (01:09 -0800)]
[PATCH] mm: more rmap debugging

Add more debugging in the rmap code in an attempt to locate to source of
the occasional "mapcount went negative" assertions.

Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix aoe without scatter-gather [Bug 7662]
Ed L. Cashin [Fri, 22 Dec 2006 09:09:21 +0000 (01:09 -0800)]
[PATCH] fix aoe without scatter-gather [Bug 7662]

Fix a bug that only appears when AoE goes over a network card that does not
support scatter-gather.  The headers in the linear part of the skb appeared
to be larger than they really were, resulting in data that was offset by 24
bytes.

This patch eliminates the offset data on cards that don't support
scatter-gather or have had scatter-gather turned off.  There remains an
unrelated issue that I'll address in a separate email.

Fixes bugzilla #7662

Signed-off-by: "Ed L. Cashin" <ecashin@coraid.com>
Cc: <stable@kernel.org>
Cc: Greg KH <greg@kroah.com>
Cc: <boddingt@optusnet.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Add a new section to CodingStyle, promoting include/linux/kernel.h
Robert P. J. Day [Fri, 22 Dec 2006 09:09:11 +0000 (01:09 -0800)]
[PATCH] Add a new section to CodingStyle, promoting include/linux/kernel.h

Add a new section to the CodingStyle file, encouraging people not to
re-invent available kernel macros such as ARRAY_SIZE(), FIELD_SIZEOF(),
min() and max(), among others.

Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Jan Engelhardt <jengelh@gmx.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Make JFFS depend on CONFIG_BROKEN
Josh Boyer [Fri, 22 Dec 2006 09:09:03 +0000 (01:09 -0800)]
[PATCH] Make JFFS depend on CONFIG_BROKEN

Mark JFFS as broken and provide a warning to users that it is deprecated
and scheduled for removal in 2.6.21

Signed-off-by: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] lock debugging: fix DEBUG_LOCKS_WARN_ON() & debug_locks_silent
Ingo Molnar [Fri, 22 Dec 2006 09:08:52 +0000 (01:08 -0800)]
[PATCH] lock debugging: fix DEBUG_LOCKS_WARN_ON() & debug_locks_silent

Matthew Wilcox noticed that the debug_locks_silent use should be inverted
in DEBUG_LOCKS_WARN_ON().  This bug was causing spurious stacktraces and
incorrect failures in the locking self-test on the parisc kernel.

Bug-found-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fsstack: Remove inode copy
Michael Halcrow [Fri, 22 Dec 2006 09:08:43 +0000 (01:08 -0800)]
[PATCH] fsstack: Remove inode copy

Trevor found a file size problem in eCryptfs in recent kernels, and he
tracked it down to an fsstack change.

This was the eCryptfs copy_attr_all:

> -void ecryptfs_copy_attr_all(struct inode *dest, const struct inode *src)
> -{
> -       dest->i_mode = src->i_mode;
> -       dest->i_nlink = src->i_nlink;
> -       dest->i_uid = src->i_uid;
> -       dest->i_gid = src->i_gid;
> -       dest->i_rdev = src->i_rdev;
> -       dest->i_atime = src->i_atime;
> -       dest->i_mtime = src->i_mtime;
> -       dest->i_ctime = src->i_ctime;
> -       dest->i_blkbits = src->i_blkbits;
> -       dest->i_flags = src->i_flags;
> -}

This is the fsstack copy_attr_all:

> +void fsstack_copy_attr_all(struct inode *dest, const struct inode *src,
> +                               int (*get_nlinks)(struct inode *))
> +{
> +       if (!get_nlinks)
> +               dest->i_nlink = src->i_nlink;
> +       else
> +               dest->i_nlink = (*get_nlinks)(dest);
> +
> +       dest->i_mode = src->i_mode;
> +       dest->i_uid = src->i_uid;
> +       dest->i_gid = src->i_gid;
> +       dest->i_rdev = src->i_rdev;
> +       dest->i_atime = src->i_atime;
> +       dest->i_mtime = src->i_mtime;
> +       dest->i_ctime = src->i_ctime;
> +       dest->i_blkbits = src->i_blkbits;
> +       dest->i_flags = src->i_flags;
> +
> +       fsstack_copy_inode_size(dest, src);
> +}

The addition of copy_inode_size breaks eCryptfs, since eCryptfs needs to
interpolate the file sizes (eCryptfs has extra space in the lower file for
the header).  The setting of the upper inode size occurs elsewhere in
eCryptfs, and the new copy_attr_all now undoes what eCryptfs was doing
right beforehand.

I see three ways of going forward from here.  (1) Something like this patch
needs to go in (assuming it jives with Unionfs), (2) we need to make a
change to the fsstack API for more fine-grained control over copying
attributes (e.g., by also including a callback function for calculating the
right file size, which will require some more work on both eCryptfs and
Unionfs), or (3) the fsstack patch on eCryptfs (commit
0cc72dc7f050188d8d7344b1dd688cbc68d3cd30 made on Fri Dec 8 02:36:31 2006
-0800) needs to be yanked in 2.6.20.

I think the simplest solution, from eCryptfs' perspective, is to just
remove the inode size copy.

Remove inode size copy in general fsstack attr copy code. Stacked
filesystems may need to interpolate the inode size, since the file
size in the lower file may be different than the file size in the
stacked layer.

Signed-off-by: Michael Halcrow <mhalcrow@us.ibm.com>
Acked-by: Josef "Jeff" Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] smc911 workqueue fixes
Andrew Morton [Fri, 22 Dec 2006 09:08:33 +0000 (01:08 -0800)]
[PATCH] smc911 workqueue fixes

Teach this driver about the workqueue changes.

Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] smc911x: fix netpoll compilation faliure
Vitaly Wool [Fri, 22 Dec 2006 09:08:24 +0000 (01:08 -0800)]
[PATCH] smc911x: fix netpoll compilation faliure

Fix the compilation failure for smc911x.c when NET_POLL_CONTROLLER is set.

Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] genirq: fix irq flow handler uninstall
Thomas Gleixner [Fri, 22 Dec 2006 09:08:14 +0000 (01:08 -0800)]
[PATCH] genirq: fix irq flow handler uninstall

The sanity check for no_irq_chip in __set_irq_hander() is unconditional on
both install and uninstall of an handler.  This triggers false warnings and
replaces no_irq_chip by dummy_irq_chip in the uninstall case.

Check only, when a real handler is installed.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Sylvain Munaut <tnt@246tNt.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix vm_events_fold_cpu() build breakage
Magnus Damm [Fri, 22 Dec 2006 09:08:01 +0000 (01:08 -0800)]
[PATCH] fix vm_events_fold_cpu() build breakage

fix vm_events_fold_cpu() build breakage

2.6.20-rc1 does not build properly if CONFIG_VM_EVENT_COUNTERS is set
and CONFIG_HOTPLUG is unset:

  CC      init/version.o
  LD      init/built-in.o
  LD      .tmp_vmlinux1
mm/built-in.o: In function `page_alloc_cpu_notify':
page_alloc.c:(.text+0x56eb): undefined reference to `vm_events_fold_cpu'
make: *** [.tmp_vmlinux1] Error 1

[akpm@osdl.org: cleanup]
Signed-off-by: Magnus Damm <magnus@valinux.co.jp>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: remove __cpuinitdata anotation to cpu_isolated_map
Tim Chen [Fri, 22 Dec 2006 09:07:50 +0000 (01:07 -0800)]
[PATCH] sched: remove __cpuinitdata anotation to cpu_isolated_map

The structure cpu_isolated_map is used not only during initialization.
Multi-core scheduler configuration changes and exclusive cpusets
use this during run time.  During setting of sched_mc_power_savings
 policy, this structure is accessed to update sched_domains.

Signed-off-by: Tim Chen <tim.c.chen@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Add cscope generated files to .gitignore
Tobias Klauser [Fri, 22 Dec 2006 09:07:32 +0000 (01:07 -0800)]
[PATCH] Add cscope generated files to .gitignore

Ignore files generated by 'make cscope'

Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix swapped parameters in mm/vmscan.c
Nigel Cunningham [Fri, 22 Dec 2006 09:07:21 +0000 (01:07 -0800)]
[PATCH] Fix swapped parameters in mm/vmscan.c

The version of mm/vmscan.c in Linus' current tree has swapped parameters in
the shrink_all_zones declaration and call, used by the various
suspend-to-disk implementations.  This doesn't seem to have any great
adverse effect, but it's clearly wrong.

Signed-off-by: Nigel Cunningham <nigel@suspend2.net>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fs/sysv/: proper prototypes for 2 functions
Adrian Bunk [Fri, 22 Dec 2006 09:07:09 +0000 (01:07 -0800)]
[PATCH] fs/sysv/: proper prototypes for 2 functions

Add proper prototypes for sysv_{init,destroy}_icache() in sysv.h

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] make kernel/printk.c:ignore_loglevel_setup() static
Adrian Bunk [Fri, 22 Dec 2006 09:07:00 +0000 (01:07 -0800)]
[PATCH] make kernel/printk.c:ignore_loglevel_setup() static

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix kernel-doc warnings in 2.6.20-rc1
Randy Dunlap [Fri, 22 Dec 2006 09:06:52 +0000 (01:06 -0800)]
[PATCH] fix kernel-doc warnings in 2.6.20-rc1

Fix kernel-doc warnings in 2.6.20-rc1.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] slab: fix kmem_ptr_validate definition
Christoph Lameter [Fri, 22 Dec 2006 09:06:44 +0000 (01:06 -0800)]
[PATCH] slab: fix kmem_ptr_validate definition

The declaration of kmem_ptr_validate in slab.h does not match the
one in slab.c. Remove the fastcall attribute (this is the only use in
slab.c).

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] rtc warning fix
Andrew Morton [Fri, 22 Dec 2006 09:06:36 +0000 (01:06 -0800)]
[PATCH] rtc warning fix

drivers/char/rtc.c:116: warning: 'hpet_rtc_interrupt' defined but not used

Cc: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix for shmem_truncate_range() BUG_ON()
Badari Pulavarty [Fri, 22 Dec 2006 09:06:23 +0000 (01:06 -0800)]
[PATCH] Fix for shmem_truncate_range() BUG_ON()

Ran into BUG() while doing madvise(REMOVE) testing.  If we are punching a
hole into shared memory segment using madvise(REMOVE) and the entire hole
is below the indirect blocks, we hit following assert.

        BUG_ON(limit <= SHMEM_NR_DIRECT);

Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Conditionally check expected_preempt_count in __resched_legal()
Mark Fasheh [Fri, 22 Dec 2006 09:06:15 +0000 (01:06 -0800)]
[PATCH] Conditionally check expected_preempt_count in __resched_legal()

Commit 2d7d253548cffdce80f4e03664686e9ccb1b0ed7 ("fix cond_resched() fix")
introduced an 'expected_preempt_count' parameter to __resched_legal() to
fix a bug where it was returning a false negative when called from
cond_resched_lock() and preemption was enabled.

Unfortunately this broke things for when preemption is disabled.
preempt_count() will always return zero, thus failing the check against any
value of expected_preempt_count not equal to zero.  cond_resched_lock() for
example, passes an expected_preempt_count value of 1.

So fix the fix for the cond_resched() fix by skipping the check of
preempt_count() against expected_preempt_count when preemption is disabled.

Credit should go to Sunil Mushran for spotting the bug during testing.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] CONFIG_VM_EVENT_COUNTER comment decrustify
Paul Jackson [Fri, 22 Dec 2006 09:06:10 +0000 (01:06 -0800)]
[PATCH] CONFIG_VM_EVENT_COUNTER comment decrustify

The VM event counters, enabled by CONFIG_VM_EVENT_COUNTERS, which provides
VM event counters in /proc/vmstat, has become more essential to
non-EMBEDDED kernel configurations than they were in the past.  Comments in
the code and the Kconfig configuration explanation were stale, downplaying
their role excessively.

Refresh those comments to correctly reflect the current role of VM event
counters.

Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: API versioning
Avi Kivity [Fri, 22 Dec 2006 09:06:02 +0000 (01:06 -0800)]
[PATCH] KVM: API versioning

Add compile-time and run-time API versioning.

Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: Handle p5 mce msrs
Michael Riepe [Fri, 22 Dec 2006 09:05:53 +0000 (01:05 -0800)]
[PATCH] KVM: Handle p5 mce msrs

This allows plan9 to get a little further booting.

Signed-off-by: Michael Riepe <michael@mr511.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: Force real-mode cs limit to 64K
Michael Riepe [Fri, 22 Dec 2006 09:05:45 +0000 (01:05 -0800)]
[PATCH] KVM: Force real-mode cs limit to 64K

This allows opensolaris to boot on kvm/intel.

Signed-off-by: Michael Riepe <michael@mr511.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: Do not export unsupported msrs to userspace
Michael Riepe [Fri, 22 Dec 2006 09:05:36 +0000 (01:05 -0800)]
[PATCH] KVM: Do not export unsupported msrs to userspace

Some msrs, such as MSR_STAR, are not available on all processors.  Exporting
them causes qemu to try to fetch them, which will fail.

So, check all msrs for validity at module load time.

Signed-off-by: Michael Riepe <michael@mr511.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: Use more traditional error handling in kvm_mmu_init()
Avi Kivity [Fri, 22 Dec 2006 09:05:28 +0000 (01:05 -0800)]
[PATCH] KVM: Use more traditional error handling in kvm_mmu_init()

Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: AMD SVM: Save and restore the floating point unit state
Avi Kivity [Fri, 22 Dec 2006 09:05:20 +0000 (01:05 -0800)]
[PATCH] KVM: AMD SVM: Save and restore the floating point unit state

Fixes sf bug 1614113 (segfaults in nbench).

Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: AMD SVM: handle MSR_STAR in 32-bit mode
Avi Kivity [Fri, 22 Dec 2006 09:05:08 +0000 (01:05 -0800)]
[PATCH] KVM: AMD SVM: handle MSR_STAR in 32-bit mode

This is necessary for linux guests.

Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] KVM: add valid_vcpu() helper
James Morris [Fri, 22 Dec 2006 09:04:55 +0000 (01:04 -0800)]
[PATCH] KVM: add valid_vcpu() helper

Consolidate the logic for checking whether a vcpu index is valid.  Also, use
likely(), as a valid value should be the overwhelmingly common case.

Signed-off-by: James Morris <jmorris@namei.org>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] truncate: dirty memory accounting fix
Andrew Morton [Fri, 22 Dec 2006 09:04:48 +0000 (01:04 -0800)]
[PATCH] truncate: dirty memory accounting fix

Only (un)account for IO and page-dirtying for devices which have real backing
store (ie: not tmpfs or ramdisks).

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] elevator: fixup typo in merge logic
Jens Axboe [Thu, 21 Dec 2006 20:20:01 +0000 (21:20 +0100)]
[PATCH] elevator: fixup typo in merge logic

The recent io scheduler allow_merge commit left the block layer with
no merging, oops. This patch fixes that up.

That means the CFQ change needs to be verified again, it might not fix
the original bug now.  But that's a seperate thing, I'll double check
that tomorrow.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] truncate: clear page dirtiness before running try_to_free_buffers()
Andrew Morton [Thu, 21 Dec 2006 19:00:33 +0000 (11:00 -0800)]
[PATCH] truncate: clear page dirtiness before running try_to_free_buffers()

truncate presently invalidates the dirty page's buffer_heads then shoots down
the page.  But try_to_free_buffers() will now bale out because the page is
dirty.

Net effect: the LRU gets filled with dirty pages which have invalidated
buffer_heads attached.  They have no ->mapping and hence cannot be cleaned.
The machine leaks memory at an enormous rate.

Fix this by cleaning the page before running try_to_free_buffers(), so
try_to_free_buffers() can do its work.

Also, remember to do dirty-page-acoounting in cancel_dirty_page() so the
machine won't wedge up trying to write non-existent dirty pages.

Probably still wrong, but now less so.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix XFS after clear_page_dirty() removal
David Chinner [Wed, 20 Dec 2006 23:24:01 +0000 (10:24 +1100)]
[PATCH] Fix XFS after clear_page_dirty() removal

XFS appears to call clear_page_dirty to get the mapping tree dirty tag
set correctly at the same time the page dirty flag is cleared.  I note
that this can be done by set_page_writeback() if we clear the dirty flag
on the page first when we are writing back the entire page.

Hence it seems to me that the XFS call to clear_page_dirty() could
easily be substituted by clear_page_dirty_for_io() followed by a call to
set_page_writeback() to get the mapping tree tags set correctly after
the page has been marked clean.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fuse: remove clear_page_dirty() call
Miklos Szeredi [Thu, 21 Dec 2006 14:18:23 +0000 (15:18 +0100)]
[PATCH] fuse: remove clear_page_dirty() call

The use by FUSE was just a remnant of an optimization from the time
when writable mappings were supported.

Now FUSE never actually allows the creation of dirty pages, so this
invocation of clear_page_dirty() is effectively a no-op.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Fix JFS after clear_page_dirty() removal
Dave Kleikamp [Wed, 20 Dec 2006 21:55:35 +0000 (15:55 -0600)]
[PATCH] Fix JFS after clear_page_dirty() removal

This patch removes some questionable code that attempted to make a
no-longer-used page easier to reclaim.

Calling metapage_writepage against such a page will not result in any
I/O being performed, so removing this code shouldn't be a big deal.

[ It's likely that we could have just replaced the "clear_page_dirty()"
  call with a call to "cancel_dirty_page()" instead, but in the
  meantime this is cleaner and simpler anyway, so unless there is some
  overriding reason (and Dave implies there isn't) I'll just use this
  patch as-is. - Linus ]

Signed-off-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoVM: Remove "clear_page_dirty()" and "test_clear_page_dirty()" functions
Linus Torvalds [Wed, 20 Dec 2006 21:46:42 +0000 (13:46 -0800)]
VM: Remove "clear_page_dirty()" and "test_clear_page_dirty()" functions

They were horribly easy to mis-use because of their tempting naming, and
they also did way more than any users of them generally wanted them to
do.

A dirty page can become clean under two circumstances:

 (a) when we write it out.  We have "clear_page_dirty_for_io()" for
     this, and that function remains unchanged.

     In the "for IO" case it is not sufficient to just clear the dirty
     bit, you also have to mark the page as being under writeback etc.

 (b) when we actually remove a page due to it becoming inaccessible to
     users, notably because it was truncate()'d away or the file (or
     metadata) no longer exists, and we thus want to cancel any
     outstanding dirty state.

For the (b) case, we now introduce "cancel_dirty_page()", which only
touches the page state itself, and verifies that the page is not mapped
(since cancelling writes on a mapped page would be actively wrong as it
is still accessible to users).

Some filesystems need to be fixed up for this: CIFS, FUSE, JFS,
ReiserFS, XFS all use the old confusing functions, and will be fixed
separately in subsequent commits (with some of them just removing the
offending logic, and others using clear_page_dirty_for_io()).

This was confirmed by Martin Michlmayr to fix the apt database
corruption on ARM.

Cc: Martin Michlmayr <tbm@cyrius.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Andrei Popa <andrei.popa@i-neo.ro>
Cc: Andrew Morton <akpm@osdl.org>
Cc: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Cc: Gordon Farquharson <gordonfarquharson@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoClean up and make try_to_free_buffers() not race with dirty pages
Linus Torvalds [Tue, 19 Dec 2006 23:21:59 +0000 (15:21 -0800)]
Clean up and make try_to_free_buffers() not race with dirty pages

This is preparatory work in our continuing saga on some hard-to-trigger
file corruption with shared writable mmap() after the dirty page
tracking changes (commit d08b3851da41d0ee60851f2c75b118e1f7a5fc89 etc)
were merged.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] workqueue: fix schedule_on_each_cpu()
Ingo Molnar [Mon, 18 Dec 2006 19:05:09 +0000 (20:05 +0100)]
[PATCH] workqueue: fix schedule_on_each_cpu()

fix the schedule_on_each_cpu() implementation: __queue_work() is now
stricter, hence set the work-pending bit before passing in the new work.

(found in the -rt tree, using Peter Zijlstra's files-lock scalability
patchset)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix leaks on pipe(2) failure exits
Al Viro [Mon, 18 Dec 2006 13:31:18 +0000 (13:31 +0000)]
[PATCH] fix leaks on pipe(2) failure exits

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>