]> err.no Git - linux-2.6/log
linux-2.6
18 years agoIB/mthca: Optimize large messages on Sinai HCAs
Eli Cohen [Thu, 2 Mar 2006 20:40:46 +0000 (12:40 -0800)]
IB/mthca: Optimize large messages on Sinai HCAs

Sinai (one-port PCI Express) HCAs get improved throughput for messages
bigger than 80 KB in DDR mode if memory keys are formatted in a
specific way.  The enhancement only works if the memory key table is
smaller than 2^24 entries.  For larger tables, the enhancement is off
and a warning is printed (to avoid silent performance loss).

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Michael Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Fix query QP return of sq_sig_all
Dotan Barak [Thu, 2 Mar 2006 19:25:27 +0000 (11:25 -0800)]
IB/uverbs: Fix query QP return of sq_sig_all

The old code didn't convert from the kernel's enum correctly.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Fix modify QP checking of "current QP state" attribute
Dotan Barak [Thu, 2 Mar 2006 19:22:28 +0000 (11:22 -0800)]
IB: Fix modify QP checking of "current QP state" attribute

According to the IB spec version 1.2, section 11.2.4.2, the current
table has a couple of mistakes where it allows the current QP state
(IB_QP_CUR_STATE) attribute.  For the transitions:

  RTS -> RTS: IB_QP_CUR_STATE should be allowed for all transports
  SQD -> SQD: IB_QP_CUR_STATE should never be allowed

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Fix multicast race between canceling and completing
Michael S. Tsirkin [Thu, 2 Mar 2006 19:07:47 +0000 (11:07 -0800)]
IPoIB: Fix multicast race between canceling and completing

ipoib_mcast_stop_thread currently tests mcast->query and if it is
NULL, does not perform wait_for_completion on the mcast and frees the
mcast object directly.

However, since both operations are done without locking, it is
possible that ipoib_mcast_join_complete is in progress on this mcast
object and has set mcast->query to NULL already.

Solve this by:
- taking priv->lock before we change mcast->query in ipoib_mcast_join_complete,
  and keeping it until we no longer need the mcast object
- taking priv->lock around mcast->query test in ipoib_mcast_stop_thread

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Clean up if posting receives fails
Eli Cohen [Thu, 2 Mar 2006 19:05:19 +0000 (11:05 -0800)]
IPoIB: Clean up if posting receives fails

If posting receives in ipoib_ib_dev_open() fails, call
ipoib_ib_dev_stop() to move the device's QP back to the RESET state so
that we can try again later.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Use an enum for HCA page size
Ishai Rabinovitz [Thu, 2 Mar 2006 06:33:11 +0000 (22:33 -0800)]
IB/mthca: Use an enum for HCA page size

Use a named enum for the HCA's internal page size, rather than having
magic values of 4096 and shifts by 12 all over the code.  Also, fix
one minor bug in EQ handling: only one HCA page is mapped to the HCA
during initialization, but a full kernel page is unmapped during
cleanup.  This might cause problems when PAGE_SIZE != 4096.

Signed-off-by: Ishai Rabinovitz <ishai@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Check alternate P_Key index when setting alternate path
Dotan Barak [Wed, 1 Mar 2006 22:28:12 +0000 (14:28 -0800)]
IB/mthca: Check alternate P_Key index when setting alternate path

Check that the alternate P_Key index is in range when setting the
alternate path for a QP.  Also make a cosmetic touch up to the debug
message printed when the main P_Key index is out of range.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Add support for send work request fence flag
Dotan Barak [Tue, 28 Feb 2006 05:02:00 +0000 (21:02 -0800)]
IB/mthca: Add support for send work request fence flag

Add support for IB_SEND_FENCE flag in post_send methods.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: Close race in setting mcast->ah
Eli Cohen [Tue, 28 Feb 2006 04:47:43 +0000 (20:47 -0800)]
IPoIB: Close race in setting mcast->ah

ipoib_mcast_send() tests mcast->ah twice.  If this value is changed
between these two points, we leak an skb.  However,
ipoib_mcast_join_finish() sets mcast->ah with no locking, so it could
race against ipoib_mcast_send().

As a solution, take priv->lock around assignment to mcast->ah thus
making sure ipoib_mcast_send() (which also takes priv->lock) is not in
flight.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Implement query_ah method
Jack Morgenstein [Mon, 27 Feb 2006 00:05:59 +0000 (16:05 -0800)]
IB/mthca: Implement query_ah method

Implement query_ah (except for AVs which are in HCA memory).  This is
needed to implement RMPP duplicate session detection on sending side
(extraction of DGID/DLID and GRH flag from address handle).

Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Write FW commands through doorbell page
Eli Cohen [Sun, 26 Feb 2006 22:36:06 +0000 (14:36 -0800)]
IB/mthca: Write FW commands through doorbell page

This patch is checks whether the HCA supports posting FW commands
through a doorbell page (user access region 0, or "UAR0").  If this is
supported, the driver maps UAR0 and uses it for FW commands. This can
be controlled by the value of a writable module parameter
fw_cmd_doorbell.  When the parameter is 0, the commands are posted
through HCR using the old method; otherwise if HCA is capable commands
go through UAR0.

This use of UAR0 to post commands eliminates the need for polling the
"go" bit prior to posting a new command. Since reading from a PCI
device is much more expensive then issuing a posted write, it is
expected that issuing FW commands this way will provide better CPU
utilization.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Return actual capacity from create SRQ operation
Dotan Barak [Thu, 23 Feb 2006 20:36:18 +0000 (12:36 -0800)]
IB/uverbs: Return actual capacity from create SRQ operation

Pass actual capacity of created SRQ back to userspace, so that
userspace can report accurate capacities.  This requires an ABI bump,
to change struct ib_uverbs_create_srq_resp.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Return actual capacity from create_srq
Dotan Barak [Thu, 23 Feb 2006 20:13:51 +0000 (12:13 -0800)]
IB/mthca: Return actual capacity from create_srq

Have mthca's create_srq method return the actual capacity of the SRQ
that gets created.  Also update comments in <rdma/ib_verbs.h> to
clarify that this is what is expected from ib_create_srq().

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIPoIB: clarify to_ipoib_neigh()
Michael S. Tsirkin [Tue, 21 Feb 2006 05:50:39 +0000 (21:50 -0800)]
IPoIB: clarify to_ipoib_neigh()

Cosmetic change: make alignment explicit in to_ipoib_neigh.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Fix alignment of struct ib_uverbs_create_qp_resp
Roland Dreier [Thu, 16 Feb 2006 17:26:19 +0000 (09:26 -0800)]
IB/uverbs: Fix alignment of struct ib_uverbs_create_qp_resp

The size of struct ib_uverbs_create_qp_resp is not even multiple of 8
bytes.  This causes problems for low-level drivers that add private
data after the structure: 32-bit userspace will look in the wrong
place for a response from a 64-bit kernel.  Fix this by adding a
reserved field.  Also, bump the ABI version because this changes the
size of a structure.

Pointed out by Hoang-Nam Nguyen <HNGUYEN@de.ibm.com>.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Bump driver version and release date
Roland Dreier [Tue, 14 Feb 2006 01:21:09 +0000 (17:21 -0800)]
IB/mthca: Bump driver version and release date

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Support for query QP and SRQ
Eli Cohen [Tue, 14 Feb 2006 00:40:21 +0000 (16:40 -0800)]
IB/mthca: Support for query QP and SRQ

Implement the query_qp and query_srq methods in mthca.

Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Support for query SRQ from userspace
Dotan Barak [Tue, 14 Feb 2006 00:31:57 +0000 (16:31 -0800)]
IB/uverbs: Support for query SRQ from userspace

Add support to uverbs to handle querying userspace SRQs (shared
receive queues), including adding an ABI for marshalling requests and
responses.  The kernel midlayer already has the underlying
ib_query_srq() function.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/uverbs: Support for query QP from userspace
Dotan Barak [Tue, 14 Feb 2006 00:31:25 +0000 (16:31 -0800)]
IB/uverbs: Support for query QP from userspace

Add support to uverbs to handle querying userspace QPs (queue pairs),
including adding an ABI for marshalling requests and responses.  The
kernel midlayer already has the underlying ib_query_qp() function.

Signed-off-by: Dotan Barak <dotanb@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Whitespace cleanups
Roland Dreier [Tue, 14 Feb 2006 00:30:49 +0000 (16:30 -0800)]
IB: Whitespace cleanups

Remove trailing whitespace and fix indentation that with spaces
instead of tabs.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Convert to use ib_modify_qp_is_ok()
Roland Dreier [Tue, 14 Feb 2006 00:30:18 +0000 (16:30 -0800)]
IB/mthca: Convert to use ib_modify_qp_is_ok()

Use ib_modify_qp_is_ok() in mthca, and delete the big table of
attributes for queue pair state transitions.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Add ib_modify_qp_is_ok() library function
Roland Dreier [Mon, 13 Feb 2006 20:48:12 +0000 (12:48 -0800)]
IB: Add ib_modify_qp_is_ok() library function

The in-kernel mthca driver contains a table of which attributes are
valid for each queue pair state transition.  It turns out that both
other IB drivers -- ipath and ehca -- which are being prepared for
merging have copied this table, errors and all.

To forestall this code duplication, move this table and the code to
check parameters against it into a midlayer library function,
ib_modify_qp_is_ok().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Generate SQ drained events when requested
Roland Dreier [Fri, 3 Feb 2006 22:53:28 +0000 (14:53 -0800)]
IB/mthca: Generate SQ drained events when requested

Add low-level driver support to ib_mthca so that consumers can request
a "send queue drained" event be generated when a transiton to the SQD
state completes.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mad: Simplify SMI by eliminating smi_check_local_dr_smp()
Ralph Campbell [Fri, 3 Feb 2006 22:32:01 +0000 (14:32 -0800)]
IB/mad: Simplify SMI by eliminating smi_check_local_dr_smp()

The call to ib_get_agent_port() shouldn't be possible to fail when
smi_check_local_dr_smp() is called from ib_mad_recv_done_handler().
When it is called from handle_outgoing_dr_smp(), the device and
port_num come from mad_agent_priv so I assume the call to
ib_get_agent_port() shouldn't fail either.  In either case,
smi_check_local_smp() only uses the mad_agent pointer to check that
mad_agent->device->process_mad is not NULL.  The device pointer would
have to be the same as the one passed to smi_check_local_dr_smp()
since that pointer is used later instead of the one checked in
smi_check_local_smp().

Signed-off-by: Hal Rosenstock <halr@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mad: Remove redundant check from smi_check_local_dr_smp()
Ralph Campbell [Fri, 3 Feb 2006 22:30:24 +0000 (14:30 -0800)]
IB/mad: Remove redundant check from smi_check_local_dr_smp()

smi_check_local_dr_smp() is called only from two places in core/mad.c
It returns 0 or 1.  In smi_check_local_dr_smp(), it checks for
a directed route SMP but this function is only called when the SMP
is a directed route so this is a NOP.

Signed-off-by: Hal Rosenstock <halr@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Enable FMR pool user to set page size
Or Gerlitz [Thu, 2 Feb 2006 18:43:45 +0000 (10:43 -0800)]
IB: Enable FMR pool user to set page size

This patch allows the consumer to set the page size of "pages" mapped
by the pool FMRs, which is a feature already existing in the base
verbs API.  On the cosmetic side it changes ib_fmr_attr.page_size field
to be named page_shift.

Signed-off-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Add modify_device method to set node description
Roland Dreier [Thu, 2 Feb 2006 18:04:19 +0000 (10:04 -0800)]
IB/mthca: Add modify_device method to set node description

Add a modify_device method to mthca, which implements setting the node
description.  This makes the writable "node_desc" sysfs attribute work
for Mellanox HCAs.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Allow userspace to set node description
Roland Dreier [Thu, 2 Feb 2006 17:47:14 +0000 (09:47 -0800)]
IB: Allow userspace to set node description

Expose a writable "node_desc" sysfs attribute for InfiniBand devices.
This allows userspace to update the node description with information
such as the node's hostname, so that IB network management software
can tie its view to the real world.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Whitespace cleanups
Roland Dreier [Wed, 1 Feb 2006 21:38:24 +0000 (13:38 -0800)]
IB/mthca: Whitespace cleanups

Remove trailing whitespace and fix indentation that with spaces
instead of tabs.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Add device-specific support for resizing CQs
Roland Dreier [Mon, 30 Jan 2006 22:31:33 +0000 (14:31 -0800)]
IB/mthca: Add device-specific support for resizing CQs

Add low-level driver support for resizing CQs (both kernel and
userspace) to mthca.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB: Add userspace support for resizing CQs
Roland Dreier [Mon, 30 Jan 2006 22:29:21 +0000 (14:29 -0800)]
IB: Add userspace support for resizing CQs

Add support to uverbs to handle resizing userspace CQs (completion
queues), including adding an ABI for marshalling requests and
responses.  The kernel midlayer already has ib_resize_cq().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Get rid of might_sleep() annotations
Roland Dreier [Wed, 1 Feb 2006 05:14:18 +0000 (21:14 -0800)]
IB/mthca: Get rid of might_sleep() annotations

The might_sleep() annotations in mthca are silly -- they all occur
shortly before calls that will end up in core functions like kmalloc()
that will print the same warning in an unsafe context anyway.  In
fact, beyond cluttering the source, we're actually bloating text with
CONFIG_DEBUG_SPINLOCK_SLEEP and/or CONFIG_PREEMPT_VOLUNTARY set.

With both options set, getting rid of the might_sleep()s saves a lot:
add/remove: 0/0 grow/shrink: 0/7 up/down: 0/-171 (-171)
function                                     old     new   delta
mthca_pd_alloc                               132     109     -23
mthca_init_cq                                969     946     -23
mthca_mr_alloc                               592     568     -24
mthca_pd_free                                 67      42     -25
mthca_free_mr                                219     194     -25
mthca_free_cq                                570     545     -25
mthca_fmr_alloc                              742     716     -26

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoIB/mthca: Make functions that never fail return void
Roland Dreier [Wed, 1 Feb 2006 04:45:51 +0000 (20:45 -0800)]
IB/mthca: Make functions that never fail return void

The function mthca_free_err_wqe() can never fail, so get rid of its
return value.  That means handle_error_cqe() doesn't have to check
what mthca_free_err_wqe() returns, which means it can't fail either
and doesn't have to return anything either.  All this results in
simpler source code and a slight object code improvement:

add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-10 (-10)
function                                     old     new   delta
mthca_free_err_wqe                            83      81      -2
mthca_poll_cq                               1758    1750      -8

Signed-off-by: Roland Dreier <rolandd@cisco.com>
18 years agoLinux 2.6.16 v2.6.16
Linus Torvalds [Mon, 20 Mar 2006 05:53:29 +0000 (21:53 -0800)]
Linux 2.6.16

18 years ago[PATCH] Remove obsolete CREDITS address
Andrea Arcangeli [Sun, 19 Mar 2006 18:04:17 +0000 (19:04 +0100)]
[PATCH] Remove obsolete CREDITS address

This address is going to be obsolete, so I should update it.

18 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Mon, 20 Mar 2006 05:12:00 +0000 (21:12 -0800)]
Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] SB1: Check for -mno-sched-prolog if building corelis debug kernel.
  [MIPS] Sibyte: Fix race in sb1250_gettimeoffset().
  [MIPS] Sibyte: Fix interrupt timer off by one bug.
  [MIPS] Sibyte: Fix M_SCD_TIMER_INIT and M_SCD_TIMER_CNT wrong field width.
  [MIPS] Protect more of timer_interrupt() by xtime_lock.
  [MIPS] Work around bad code generation for <asm/io.h>.
  [MIPS] Simple patch to power off DBAU1200
  [MIPS] Fix DBAu1550 software power off.
  [MIPS] local_r4k_flush_cache_page fix
  [MIPS] SB1: Fix interrupt disable hazard.
  [MIPS] Get rid of the IP22-specific code in arclib.
  Update MAINTAINERS entry for MIPS.

18 years ago[TG3]: 40-bit DMA workaround part 2
Michael Chan [Sun, 19 Mar 2006 21:21:12 +0000 (13:21 -0800)]
[TG3]: 40-bit DMA workaround part 2

The 40-bit DMA workaround recently implemented for 5714, 5715, and
5780 needs to be expanded because there may be other tg3 devices
behind the EPB Express to PCIX bridge in the 5780 class device.

For example, some 4-port card or mother board designs have 5704 behind
the 5714.

All devices behind the EPB require the 40-bit DMA workaround.

Thanks to Chris Elmquist again for reporting the problem and testing
the patch.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[AX.25]: Fix potencial memory hole.
Ralf Baechle DL5RB [Sun, 19 Mar 2006 21:20:06 +0000 (13:20 -0800)]
[AX.25]: Fix potencial memory hole.

If the AX.25 dialect chosen by the sysadmin is set to DAMA master / 3
(or DAMA slave / 2, if CONFIG_AX25_DAMA_SLAVE=n) ax25_kick() will fall
through the switch statement without calling ax25_send_iframe() or any
other function that would eventually free skbn thus leaking the packet.

Fix by restricting the sysctl inferface to allow only actually supported
AX.25 dialects.

The system administration mistake needed for this to happen is rather
unlikely, so this is an uncritical hole.

Coverity #651.

Signed-off-by: Ralf Baechle DL5RB <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] Kconfig: swap VIDEO_CX88_ALSA and VIDEO_CX88_DVB
Michael Krufky [Wed, 15 Mar 2006 05:36:13 +0000 (02:36 -0300)]
[PATCH] Kconfig: swap VIDEO_CX88_ALSA and VIDEO_CX88_DVB

VIDEO_CX88_ALSA should not be between VIDEO_CX88_DVB and
VIDEO_CX88_DVB_ALL_FRONTENDS

When cx88-alsa was added to cx88/Kconfig, it was added in between
VIDEO_CX88_DVB and VIDEO_CX88_DVB_ALL_FRONTENDS.  This caused
undesireable effects to the appearance of the menu options in
menuconfig.

This fix reorders cx88-alsa and cx88-dvb in Kconfig, to match saa7134,
and restore the correct menuconfig appearance.

Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fixed em28xx based system lockup
Markus Rechberger [Tue, 7 Feb 2006 10:49:13 +0000 (08:49 -0200)]
[PATCH] Fixed em28xx based system lockup

Fixed em28xx based system lockup, device needs to be initialized before
starting the isoc transfer otherwise the system will completly lock up.

Signed-off-by: Markus Rechberger <mrechberger@gmail.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] disable unshare(CLONE_VM) for now
Oleg Nesterov [Sat, 18 Mar 2006 17:41:10 +0000 (20:41 +0300)]
[PATCH] disable unshare(CLONE_VM) for now

sys_unshare() does mmput(new_mm).  This is not enough if we have
mm->core_waiters.

This patch is a temporary fix for soon to be released 2.6.16.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
[ Checked with Uli: "I'm not planning to use unshare(CLONE_VM).  It's
  not needed for any functionality planned so far.  What we (as in Red
  Hat) need unshare() for now is the filesystem side." ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[MIPS] SB1: Check for -mno-sched-prolog if building corelis debug kernel.
Ralf Baechle [Sat, 18 Mar 2006 16:59:31 +0000 (16:59 +0000)]
[MIPS] SB1: Check for -mno-sched-prolog if building corelis debug kernel.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Sibyte: Fix race in sb1250_gettimeoffset().
Ralf Baechle [Wed, 15 Mar 2006 00:03:29 +0000 (00:03 +0000)]
[MIPS] Sibyte: Fix race in sb1250_gettimeoffset().

From Dave Johnson <djohnson+linuxmips@sw.starentnetworks.com>:

sb1250_gettimeoffset() simply reads the current cpu 0 timer remaining
value, however once this counter reaches 0 and the interrupt is raised,
it immediately resets and begins to count down again.

If sb1250_gettimeoffset() is called on cpu 1 via do_gettimeofday() after
the timer has reset but prior to cpu 0 processing the interrupt and
taking write_seqlock() in timer_interrupt() it will return a full value
(or close to it) causing time to jump backwards 1ms. Once cpu 0 handles
the interrupt and timer_interrupt() gets far enough along it will jump
forward 1ms.

Fix this problem by implementing mips_hpt_*() on sb1250 using a spare
timer unrelated to the existing periodic interrupt timers. It runs at
1Mhz with a full 23bit counter.  This eliminated the custom
do_gettimeoffset() for sb1250 and allowed use of the generic
fixed_rate_gettimeoffset() using mips_hpt_*() and timerhi/timerlo.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Sibyte: Fix interrupt timer off by one bug.
Ralf Baechle [Tue, 14 Mar 2006 23:52:47 +0000 (23:52 +0000)]
[MIPS] Sibyte: Fix interrupt timer off by one bug.

From Dave Johnson <djohnson+linuxmips@sw.starentnetworks.com>:

The timers need to be loaded with 1 less than the desired interval not
the interval itself.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Sibyte: Fix M_SCD_TIMER_INIT and M_SCD_TIMER_CNT wrong field width.
Ralf Baechle [Tue, 14 Mar 2006 23:47:35 +0000 (23:47 +0000)]
[MIPS] Sibyte: Fix M_SCD_TIMER_INIT and M_SCD_TIMER_CNT wrong field width.

From Dave Johnson <djohnson+linuxmips@sw.starentnetworks.com>:

Field width should be 23 bits not 20 bits.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Protect more of timer_interrupt() by xtime_lock.
Ralf Baechle [Tue, 14 Mar 2006 23:46:58 +0000 (23:46 +0000)]
[MIPS] Protect more of timer_interrupt() by xtime_lock.

From Dave Johnson <djohnson+linuxmips@sw.starentnetworks.com>:

* do_timer() expects the arch-specific handler to take the lock as it
  modifies jiffies[_64] and xtime.
* writing timerhi/lo in timer_interrupt() will mess up
  fixed_rate_gettimeoffset() which reads timerhi/lo.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Work around bad code generation for <asm/io.h>.
Ralf Baechle [Wed, 15 Mar 2006 11:36:31 +0000 (11:36 +0000)]
[MIPS] Work around bad code generation for <asm/io.h>.

If a call to set_io_port_base() was being followed by usage of
mips_io_port_base in the same function gcc was possibly using the old
value due to some clever abuse of const.  Adding a barrier will keep
the optimization and result in correct code with latest gcc.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Simple patch to power off DBAU1200
Matej Kupljen [Wed, 30 Nov 2005 09:20:01 +0000 (10:20 +0100)]
[MIPS] Simple patch to power off DBAU1200

Signed-off-by: Matej Kupljen <matej.kupljen@ultra.si>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix DBAu1550 software power off.
Sergei Shtylylov [Tue, 14 Mar 2006 04:20:00 +0000 (07:20 +0300)]
[MIPS] Fix DBAu1550 software power off.

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] local_r4k_flush_cache_page fix
Atsushi Nemoto [Mon, 13 Mar 2006 09:23:03 +0000 (18:23 +0900)]
[MIPS] local_r4k_flush_cache_page fix

If dcache_size != icache_size or dcache_size != scache_size, or
set-associative cache, icache/scache does not flushed properly.  Make
blast_?cache_page_indexed() masks its index value correctly.  Also,
use physical address for physically indexed pcache/scache.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] SB1: Fix interrupt disable hazard.
Ralf Baechle [Mon, 13 Mar 2006 16:16:29 +0000 (16:16 +0000)]
[MIPS] SB1: Fix interrupt disable hazard.

The SB1 core has a three cycle interrupt disable hazard but we were
wrongly treating it as fully interlocked.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Get rid of the IP22-specific code in arclib.
Ralf Baechle [Fri, 10 Mar 2006 19:47:17 +0000 (19:47 +0000)]
[MIPS] Get rid of the IP22-specific code in arclib.

This breaks the kernel build if sgiwd93 was configured as a module.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years agoUpdate MAINTAINERS entry for MIPS.
Ralf Baechle [Fri, 10 Mar 2006 13:47:21 +0000 (13:47 +0000)]
Update MAINTAINERS entry for MIPS.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[NET]: Fix race condition in sk_wait_event().
Alexey Kuznetsov [Sat, 18 Mar 2006 00:05:43 +0000 (16:05 -0800)]
[NET]: Fix race condition in sk_wait_event().

It is broken, the condition is checked out of socket lock. It is
wonderful the bug survived for so long time.

[ This fixes bugzilla #6233:
  race condition in tcp_sendmsg when connection became established ]

Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] fix free swap cache latency
Hugh Dickins [Fri, 17 Mar 2006 07:04:09 +0000 (23:04 -0800)]
[PATCH] fix free swap cache latency

Lee Revell reported 28ms latency when process with lots of swapped memory
exits.

2.6.15 introduced a latency regression when unmapping: in accounting the
zap_work latency breaker, pte_none counted 1, pte_present PAGE_SIZE, but a
swap entry counted nothing at all.  We think of pages present as the slow
case, but Lee's trace shows that free_swap_and_cache's radix tree lookup
can make a lot of work - and we could have been doing it many thousands of
times without a latency break.

Move the zap_work update up to account swap entries like pages present.
This does account non-linear pte_file entries, and unmap_mapping_range
skipping over swap entries, by the same amount even though they're quick:
but neither of those cases deserves complicating the code (and they're
treated no worse than they were in 2.6.14).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kbuild: fix buffer overflow in modpost
Sam Ravnborg [Fri, 17 Mar 2006 07:04:08 +0000 (23:04 -0800)]
[PATCH] kbuild: fix buffer overflow in modpost

Jiri Benc <jbenc@suse.cz> reported that modpost would stop with SIGABRT if
used with long filepaths.
The error looked like:
>   Building modules, stage 2.
>   MODPOST
> *** glibc detected *** scripts/mod/modpost: realloc(): invalid next size:
+0x0809f588 ***
> [...]

Fix this by allocating at least the required memory + SZ bytes each time.
Before we sometimes ended up allocating too little memory resuting in the
glibc detected bug above.  Based on patch originally submitted by: Jiri
Benc <jbenc@suse.cz>

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] nfsservctl(): remove user-triggerable printk
Peter Staubach [Fri, 17 Mar 2006 07:04:02 +0000 (23:04 -0800)]
[PATCH] nfsservctl(): remove user-triggerable printk

A user can use nfsservctl() to spam the logs.

This can happen because the arguments to the nfsservctl() system call are
versioned.  This is a good thing.  However, when a bad version is detected,
the kernel prints a message and then returns an error.

Signed-off-by: Peter Staubach <staubach@redhat.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix race in pagevec_strip?
Christoph Lameter [Fri, 17 Mar 2006 07:04:07 +0000 (23:04 -0800)]
[PATCH] fix race in pagevec_strip?

We can call try_to_release_page() with PagePrivate off and a valid
page->mapping This may cause all sorts of trouble for the filesystem
*_releasepage() handlers.  XFS bombs out in that case.

Lock the page before checking for page private.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] dm stripe: Fix bounds
Kevin Corry [Fri, 17 Mar 2006 07:04:03 +0000 (23:04 -0800)]
[PATCH] dm stripe: Fix bounds

The dm-stripe target currently does not enforce that the size of a stripe
device be a multiple of the chunk-size.  Under certain conditions, this can
lead to I/O requests going off the end of an underlying device.  This
test-case shows one example.

echo "0 100 linear /dev/hdb1 0" | dmsetup create linear0
echo "0 100 linear /dev/hdb1 100" | dmsetup create linear1
echo "0 200 striped 2 32 /dev/mapper/linear0 0 /dev/mapper/linear1 0" | \
   dmsetup create stripe0
dd if=/dev/zero of=/dev/mapper/stripe0 bs=1k

This will produce the output:
dd: writing '/dev/mapper/stripe0': Input/output error
97+0 records in
96+0 records out

And in the kernel log will be:
attempt to access beyond end of device
dm-0: rw=0, want=104, limit=100

The patch will check that the table size is a multiple of the stripe
chunk-size when the table is created, which will prevent the above striped
device from being created.

This should not affect tools like LVM or EVMS, since in all the cases I can
think of, striped devices are always created with the sizes being a
multiple of the chunk-size.

The size of a stripe device must be a multiple of its chunk-size.

(akpm: that typecast is quite gratuitous)

Signed-off-by: Kevin Corry <kevcorry@us.ibm.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86: check for online cpus before bringing them up
Srivatsa Vaddagiri [Fri, 17 Mar 2006 07:04:06 +0000 (23:04 -0800)]
[PATCH] x86: check for online cpus before bringing them up

Bryce reported a bug wherein offlining CPU0 (on x86 box) and then
subsequently onlining it resulted in a lockup.

On x86, CPU0 is never offlined.  The subsequent attempt to online CPU0
doesn't take that into account.  It actually tries to bootup the already
booted CPU.  Following patch fixes the problem (as acknowledged by Bryce).
Please consider for inclusion in 2.6.16.

Check if cpu is already online.

Signed-off-by: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] v9fs: fix overzealous dropping of dentry which breaks dcache
Eric Van Hensbergen [Fri, 17 Mar 2006 07:04:04 +0000 (23:04 -0800)]
[PATCH] v9fs: fix overzealous dropping of dentry which breaks dcache

There is a d_drop in dir_release which caused problems as it invalidates
dcache entries too soon.  This was likely a part of the wierd cwd behavior
folks were seeing.

Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] posix-timers: fix requeue accounting when signal is ignored
Roman Zippel [Fri, 17 Mar 2006 07:04:01 +0000 (23:04 -0800)]
[PATCH] posix-timers: fix requeue accounting when signal is ignored

When the posix-timer signal is ignored then the timer is rearmed by the
callback function.  The requeue pending accounting has to be fixed up else
the state might be wrong.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] time_interpolator: add __read_mostly
Christoph Lameter [Fri, 17 Mar 2006 07:04:00 +0000 (23:04 -0800)]
[PATCH] time_interpolator: add __read_mostly

The pointer to the current time interpolator and the current list of time
interpolators are typically only changed during bootup.  Adding
__read_mostly takes them away from possibly hot cachelines.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: Fail with error if swap not setup
Christoph Lameter [Fri, 17 Mar 2006 07:03:59 +0000 (23:03 -0800)]
[PATCH] page migration: Fail with error if swap not setup

Currently the migration of anonymous pages will silently fail if no swap is
setup.  This patch makes page migration functions check for available swap
and fail with -ENODEV if no swap space is available.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] unshare: Use rcu_assign_pointer when setting sighand
Eric W. Biederman [Thu, 16 Mar 2006 17:31:38 +0000 (10:31 -0700)]
[PATCH] unshare: Use rcu_assign_pointer when setting sighand

The sighand pointer only needs the rcu_read_lock on the
read side.  So only depending on task_lock protection
when setting this pointer is not enough.  We also need
a memory barrier to ensure the initialization is seen first.

Use rcu_assign_pointer as it does this for us, and clearly
documents that we are setting an rcu readable pointer.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[netdrvr] fix array overflows in Chelsio driver
Scott Bardone [Fri, 17 Mar 2006 00:20:40 +0000 (19:20 -0500)]
[netdrvr] fix array overflows in Chelsio driver

Adrian Bunk wrote:
> The Coverity checker spotted the following two array overflows in
> drivers/net/chelsio/sge.c (in both cases, the arrays contain 3
> elements):
[snip]

This is a bug. The array should contain 2 elements.  Here is the fix.

Signed-off-by: Scott Bardone <sbardone@chelsio.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
18 years ago[PATCH] e1000 endianness bugs
David S. Miller [Wed, 15 Mar 2006 22:26:28 +0000 (14:26 -0800)]
[PATCH] e1000 endianness bugs

return -E_NO_BIG_ENDIAN_TESTING;

[E1000]: Fix 4 missed endianness conversions on RX descriptor fields.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
18 years agoMerge branch 'upstream-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/linvil...
Jeff Garzik [Fri, 17 Mar 2006 00:16:59 +0000 (19:16 -0500)]
Merge branch 'upstream-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6

18 years agoMerge branch 'e100-fixes' of git://198.78.49.142/~jbrandeb/linux-2.6
Jeff Garzik [Fri, 17 Mar 2006 00:15:26 +0000 (19:15 -0500)]
Merge branch 'e100-fixes' of git://198.78.49.142/~jbrandeb/linux-2.6

18 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc-merge
Linus Torvalds [Thu, 16 Mar 2006 17:13:34 +0000 (09:13 -0800)]
Merge git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc-merge

* git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc-merge:
  powerpc: update defconfigs
  [PATCH] powerpc: properly configure DDR/P5IOC children devs
  [PATCH] powerpc: remove duplicate EXPORT_SYMBOLS
  [PATCH] powerpc: RTC memory corruption
  [PATCH] powerpc: enable NAP only on cpus who support it to avoid memory corruption
  [PATCH] powerpc: Clarify wording for CRASH_DUMP Kconfig option
  [PATCH] powerpc/64: enable CONFIG_BLK_DEV_SL82C105
  [PATCH] powerpc: correct cacheflush loop in zImage
  powerpc: Fix problem with time going backwards
  powerpc: Disallow lparcfg being a module

18 years agopowerpc: update defconfigs
Paul Mackerras [Wed, 15 Mar 2006 05:22:54 +0000 (16:22 +1100)]
powerpc: update defconfigs

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: properly configure DDR/P5IOC children devs
John Rose [Tue, 14 Mar 2006 23:46:45 +0000 (17:46 -0600)]
[PATCH] powerpc: properly configure DDR/P5IOC children devs

The dynamic add path for PCI Host Bridges can fail to configure children
adapters under P5IOC controllers.  It fails to properly fixup bus/device
resources, and it fails to properly enable EEH.  Both of these steps
need to occur before any children devices are enabled in
pci_bus_add_devices().

Signed-off-by: John Rose <johnrose@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: remove duplicate EXPORT_SYMBOLS
Olaf Hering [Tue, 14 Mar 2006 20:21:11 +0000 (21:21 +0100)]
[PATCH] powerpc: remove duplicate EXPORT_SYMBOLS

remove warnings when building a 64bit kernel.
smp_call_function triggers also with 32bit kernel.

WARNING: vmlinux: duplicate symbol 'smp_call_function' previous definition was in vmlinux
arch/powerpc/kernel/ppc_ksyms.c:164:EXPORT_SYMBOL(smp_call_function);
arch/powerpc/kernel/smp.c:300:EXPORT_SYMBOL(smp_call_function);

WARNING: vmlinux: duplicate symbol 'ioremap' previous definition was in vmlinux
arch/powerpc/kernel/ppc_ksyms.c:113:EXPORT_SYMBOL(ioremap);
arch/powerpc/mm/pgtable_64.c:321:EXPORT_SYMBOL(ioremap);

WARNING: vmlinux: duplicate symbol '__ioremap' previous definition was in vmlinux
arch/powerpc/kernel/ppc_ksyms.c:117:EXPORT_SYMBOL(__ioremap);
arch/powerpc/mm/pgtable_64.c:322:EXPORT_SYMBOL(__ioremap);

WARNING: vmlinux: duplicate symbol 'iounmap' previous definition was in vmlinux
arch/powerpc/kernel/ppc_ksyms.c:118:EXPORT_SYMBOL(iounmap);
arch/powerpc/mm/pgtable_64.c:323:EXPORT_SYMBOL(iounmap);

Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: RTC memory corruption
Michael Neuling [Tue, 14 Mar 2006 06:11:51 +0000 (17:11 +1100)]
[PATCH] powerpc: RTC memory corruption

We should be memset'ing the data we are pointing to, not the pointer
itself.  This is in an error path so we probably don't hit it much.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: enable NAP only on cpus who support it to avoid memory corruption
Benjamin Herrenschmidt [Sat, 11 Mar 2006 23:55:01 +0000 (10:55 +1100)]
[PATCH] powerpc: enable NAP only on cpus who support it to avoid memory corruption

This patch fixes incorrect setting of powersave_nap to 1 on all
PowerMacs, potentially causing memory corruption on some models. This
bug was introuced by me during the 32/64 bits merge.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: Clarify wording for CRASH_DUMP Kconfig option
Michael Ellerman [Fri, 10 Mar 2006 04:01:08 +0000 (15:01 +1100)]
[PATCH] powerpc: Clarify wording for CRASH_DUMP Kconfig option

The wording of the CRASH_DUMP Kconfig option is not very clear. It gives you a
kernel that can be used _as_ the kdump kernel, not a kernel that can boot into
a kdump kernel.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc/64: enable CONFIG_BLK_DEV_SL82C105
Olaf Hering [Mon, 6 Mar 2006 15:39:25 +0000 (16:39 +0100)]
[PATCH] powerpc/64: enable CONFIG_BLK_DEV_SL82C105

Enable the onboard IDE driver for p610, p615 and p630.
They have the CD connected to this card. All other RS/6000 systems with this
controller have no connectors and dont need this option.

Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years ago[PATCH] powerpc: correct cacheflush loop in zImage
Olaf Hering [Sat, 4 Mar 2006 12:15:40 +0000 (13:15 +0100)]
[PATCH] powerpc: correct cacheflush loop in zImage

Correct the loop for cacheflush. No idea where I copied the code from,
but the original does not work correct. Maybe the flush is not needed.

Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agopowerpc: Fix problem with time going backwards
Paul Mackerras [Wed, 15 Mar 2006 02:47:15 +0000 (13:47 +1100)]
powerpc: Fix problem with time going backwards

The recent changes to keep gettimeofday in sync with xtime had the side
effect that it was occasionally possible for the time reported by
gettimeofday to go back by a microsecond.  There were two reasons:
(1) when we recalculated the offsets used by gettimeofday every 2^31
timebase ticks, we lost an accumulated fractional microsecond, and
(2) because the update is done some time after the notional start of
jiffy, if ntp is slowing the clock, it is possible to see time go backwards
when the timebase factor gets reduced.

This fixes it by (a) slowing the gettimeofday clock by about 1us in
2^31 timebase ticks (a factor of less than 1 in 3.7 million), and (b)
adjusting the timebase offsets in the rare case that the gettimeofday
result could possibly go backwards (i.e. when ntp is slowing the clock
and the timer interrupt is late).  In this case the adjustment will
reduce to zero eventually because of (a).

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Thu, 16 Mar 2006 02:17:11 +0000 (18:17 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 3362/1: [cleanup] - duplicate decleration of mem_fclk_21285
  [ARM] 3365/1: [cleanup] header for compat.c exported functions
  [ARM] 3364/1: [cleanup] warning fix - definitions for enable_hlt and disable_hlt
  [ARM] 3363/1: [cleanup] process.c - fix warnings
  [ARM] 3358/1: [S3C2410] add missing SPI DMA resources
  [ARM] 3357/1: enable frontlight on collie
  [ARM] Fix "thead" typo

18 years ago[PATCH] Fix ext2 readdir f_pos re-validation logic
Al Viro [Wed, 15 Mar 2006 21:41:59 +0000 (21:41 +0000)]
[PATCH] Fix ext2 readdir f_pos re-validation logic

This fixes not one, but _two_, silly (but admittedly hard to hit) bugs
in the ext2 filesystem "readdir()" function.  It also cleans up the code
to avoid the unnecessary goto mess.

The bugs were related to re-valiating the f_pos value after somebody had
either done an "lseek()" on the directory to an invalid offset, or when
the offset had become invalid due to a file being unlinked in the
directory.  The code would not only set the f_version too eagerly, it
would also not update f_pos appropriately for when the offset fixup took
place.

When that happened, we'd occasionally subsequently fail the readdir()
even when we shouldn't (no real harm done, but an ugly printk, and
obviously you would end up not necessarily seeing all entries).

Thanks to Masoud Sharbiani <masouds@google.com> who noticed the problem
and had a test-case for it, and also fixed up a thinko in the first
version of this patch.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Masoud Sharbiani <masouds@google.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[ARM] 3362/1: [cleanup] - duplicate decleration of mem_fclk_21285
Ben Dooks [Wed, 15 Mar 2006 23:18:08 +0000 (23:18 +0000)]
[ARM] 3362/1: [cleanup] - duplicate decleration of mem_fclk_21285

Patch from Ben Dooks

arch/arm/kernel/setup.c declares mem_fclk_21285 when
this is already declared in include/asm-arm/system.h

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] 3365/1: [cleanup] header for compat.c exported functions
Ben Dooks [Wed, 15 Mar 2006 23:17:30 +0000 (23:17 +0000)]
[ARM] 3365/1: [cleanup] header for compat.c exported functions

Patch from Ben Dooks

arch/arm/kernel/compat.c exports two functions,
convert_to_tag_list and squash_mem_tags which
are not defined in any header files, and not
used outside arch/arm/kernel.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] 3364/1: [cleanup] warning fix - definitions for enable_hlt and disable_hlt
Ben Dooks [Wed, 15 Mar 2006 23:17:26 +0000 (23:17 +0000)]
[ARM] 3364/1: [cleanup] warning fix - definitions for enable_hlt and disable_hlt

Patch from Ben Dooks

The enable_hlt and disable_hlt should be declared in
include/asm/setup.h. This fixes sparse errors from
arch/arm/kernel/process.c

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] 3363/1: [cleanup] process.c - fix warnings
Ben Dooks [Wed, 15 Mar 2006 23:17:23 +0000 (23:17 +0000)]
[ARM] 3363/1: [cleanup] process.c - fix warnings

Patch from Ben Dooks

Fix the following warnings from sparse:

arch/arm/kernel/process.c:86:6: warning: symbol 'default_idle' was not declared. Should it be static?
arch/arm/kernel/process.c:378:5: warning: symbol 'dump_fpu' was not declared.   Should it be static?

Include <linux/elfcore.h> for dump_fpu() decleration, and
make default_idle() static as it is not used outside the file.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[PATCH] ieee80211: Fix QoS is not active problem
Hong Liu [Wed, 8 Mar 2006 02:50:20 +0000 (10:50 +0800)]
[PATCH] ieee80211: Fix QoS is not active problem

Fix QoS is not active even the network and the card is QOS enabled.
The problem is we pass the wrong ieee80211_network address to
ipw_handle_beacon/ipw_handle_probe_response, thus the
ieee80211_network->qos_data.active will not be set, causing the driver
not sending QoS frames at all.

Signed-off-by: Hong Liu <hong.liu@intel.com>
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
18 years ago[PATCH] ieee80211: Fix CCMP decryption problem when QoS is enabled
Zhu Yi [Tue, 28 Feb 2006 07:10:36 +0000 (15:10 +0800)]
[PATCH] ieee80211: Fix CCMP decryption problem when QoS is enabled

Use the correct STYPE for Qos data.

Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
18 years agoe100: fix eeh on pseries during ethtool -t
Jesse Brandeburg [Wed, 15 Mar 2006 18:55:24 +0000 (10:55 -0800)]
e100: fix eeh on pseries during ethtool -t

Olaf Hering reported a problem on pseries with e100 where ethtool -t would
cause a bus error, and the e100 driver would stop working.  Due to the new
load ucode command the cb list must be allocated before calling
e100_init_hw, so remove the call and just let e100_up take care of it.

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
18 years ago[PATCH] fs/namespace.c:dup_namespace(): fix a use after free
Adrian Bunk [Wed, 15 Mar 2006 16:37:32 +0000 (17:37 +0100)]
[PATCH] fs/namespace.c:dup_namespace(): fix a use after free

The Coverity checker spotted the following bug in dup_namespace():

<--  snip  -->

        if (!new_ns->root) {
                up_write(&namespace_sem);
                kfree(new_ns);
                goto out;
        }
...
out:
        return new_ns;

<--  snip  -->

Callers expect a non-NULL result to not be freed.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[ARM] 3358/1: [S3C2410] add missing SPI DMA resources
Albrecht Dreß [Wed, 15 Mar 2006 16:03:05 +0000 (16:03 +0000)]
[ARM] 3358/1: [S3C2410] add missing SPI DMA resources

Patch from Albrecht Dreß

Add DMA resources to s3c2410 spi platform devices - dma_(alloc|free)_coherent should now work as expected.

Signed-off-by: Albrecht Dreß <albrecht.dress@lios-tech.com>
Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] 3357/1: enable frontlight on collie
Pavel Machek [Wed, 15 Mar 2006 16:03:03 +0000 (16:03 +0000)]
[ARM] 3357/1: enable frontlight on collie

Patch from Pavel Machek

Enable frontlight during collie bootup, so that display is actually
readable in anything other than bright sunlight.

Signed-off-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] Fix "thead" typo
Russell King [Wed, 15 Mar 2006 14:57:13 +0000 (14:57 +0000)]
[ARM] Fix "thead" typo

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[PATCH] Consistent capabilites associated with MPOL_MOVE_ALL
Christoph Lameter [Wed, 15 Mar 2006 03:50:21 +0000 (19:50 -0800)]
[PATCH] Consistent capabilites associated with MPOL_MOVE_ALL

It seems that setting scheduling policy and priorities is also the kind of
thing that might be performed in apps that also use the NUMA API, so it
would seem consistent to use CAP_SYS_NICE for NUMA also.

So use CAP_SYS_NICE for controlling migration permissions.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Page migration documentation update
Christoph Lameter [Wed, 15 Mar 2006 03:50:20 +0000 (19:50 -0800)]
[PATCH] Page migration documentation update

Update the documentation for page migration.

- Fix up bits and pieces in cpusets.txt

- Rework text in vm/page-migration to be clearer and reflect the final
  version of page migration in 2.6.16. Mention Andi Kleen's numactl
  package that contains user space tools for page migration via
  libnuma. Add reference to numa_maps and to the manpage in numactl.

- Add todo list for outstanding issues

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] page migration: fail if page is in a vma flagged VM_LOCKED
Christoph Lameter [Wed, 15 Mar 2006 03:50:19 +0000 (19:50 -0800)]
[PATCH] page migration: fail if page is in a vma flagged VM_LOCKED

page migration currently simply retries a couple of times if try_to_unmap()
fails without inspecting the return code.

However, SWAP_FAIL indicates that the page is in a vma that has the
VM_LOCKED flag set (if ignore_refs ==1).  We can check for that return code
and avoid retrying the migration.

migrate_page_remove_references() now needs to return a reason why the
failure occured.  So switch migrate_page_remove_references to use -Exx
style error messages.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] "s390: multiple subchannel sets support" fix
Greg Smith [Wed, 15 Mar 2006 03:50:17 +0000 (19:50 -0800)]
[PATCH] "s390: multiple subchannel sets support" fix

It seems this patch got dropped (it was in addition to the `s390:
improve response code handling in chsc_enable_facility()' patch).

Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge git://oss.sgi.com:8090/oss/git/rc-fixes
Linus Torvalds [Wed, 15 Mar 2006 04:50:45 +0000 (20:50 -0800)]
Merge git://oss.sgi.com:8090/oss/git/rc-fixes

* git://oss.sgi.com:8090/oss/git/rc-fixes:
  Fix a direct I/O locking issue revealed by the new mutex code.

18 years agoFix a direct I/O locking issue revealed by the new mutex code.
Nathan Scott [Wed, 15 Mar 2006 04:14:45 +0000 (15:14 +1100)]
Fix a direct I/O locking issue revealed by the new mutex code.
Affects only XFS (i.e. DIO_OWN_LOCKING case) - currently it is
not possible to get i_mutex locking correct when using DIO_OWN
direct I/O locking in a filesystem due to indeterminism in the
possible return code/lock/unlock combinations.  This can cause
a direct read to attempt a double i_mutex unlock inside XFS.

We're now ensuring __blockdev_direct_IO always exits with the
inode i_mutex (still) held for a direct reader.

Tested with the three different locking modes (via direct block
device access, ext3 and XFS) - both reading and writing; cannot
find any regressions resulting from this change, and it clearly
fixes the mutex_unlock warning originally reported here:
http://marc.theaimsgroup.com/?l=linux-kernel&m=114189068126253&w=2

Signed-off-by: Nathan Scott <nathans@sgi.com>
Acked-by: Christoph Hellwig <hch@lst.de>
18 years ago[PATCH] JFS: Take logsync lock before testing mp->lsn
Dave Kleikamp [Tue, 14 Mar 2006 19:44:00 +0000 (13:44 -0600)]
[PATCH] JFS: Take logsync lock before testing mp->lsn

This fixes a race where lsn could be cleared before taking the lock

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
Linus Torvalds [Tue, 14 Mar 2006 19:03:50 +0000 (11:03 -0800)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
  [SCSI] zfcp: fix device registration issues
  [SCSI] scsi_transport_fc: fix FC_HOST_NUM_ATTRS
  [SCSI] scsi: aha152x pcmcia driver needs spi transport
  [SCSI] zfcp: correctly set this_id for hosts
  [SCSI] Add Brownie to blacklist