linux-edison.git
12 years agonet/irda: fix error return code in bfin_sir_open()
Wei Yongjun [Tue, 7 May 2013 02:14:33 +0000 (02:14 +0000)]
net/irda: fix error return code in bfin_sir_open()

Fix to return -ENOMEM in the irlap_open() error handling case instead
of 0(overwrite to 0 by bfin_sir_startup()), as done elsewhere in this
function.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agonet: of_mdio: fix behavior on missing phy device
Sebastian Hesselbarth [Mon, 6 May 2013 23:49:31 +0000 (23:49 +0000)]
net: of_mdio: fix behavior on missing phy device

of_mdiobus_register creates a phy_device even if get_phy_device failed
to create it previously. This causes indefinite polling on non-existent
PHYs. This fix makes of_mdio_register rely on get_phy_device to
properly create the device or fail otherwise.

Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agosierra_net: keep status interrupt URB active
Dan Williams [Mon, 6 May 2013 11:34:56 +0000 (11:34 +0000)]
sierra_net: keep status interrupt URB active

The driver and firmware sync up through SYNC messages, and the
firmware's affirmative reply to these SYNC messages appears to be the
"Reset" indication received via the status interrupt endpoint.  Thus the
driver needs the status interrupt endpoint always active so that the
Reset indication can be received even if the netdev is closed, which is
the case right after device insertion.

If the Reset indication is not received by the driver, it continues
sending SYNC messages to the firmware, which crashes about 10 seconds
later and the device stops responding.

Signed-off-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agousbnet: allow status interrupt URB to always be active
Dan Williams [Mon, 6 May 2013 11:29:23 +0000 (11:29 +0000)]
usbnet: allow status interrupt URB to always be active

Some drivers (sierra_net) need the status interrupt URB
active even when the device is closed, because they receive
custom indications from firmware.  Add functions to refcount
the status interrupt URB submit/kill operation so that
sub-drivers and the generic driver don't fight over whether
the status interrupt URB is active or not.

A sub-driver can call usbnet_status_start() at any time, but
the URB is only submitted the first time the function is
called.  Likewise, when the sub-driver is done with the URB,
it calls usbnet_status_stop() but the URB is only killed when
all users have stopped it.  The URB is still killed and
re-submitted for suspend/resume, as before, with the same
refcount it had at suspend.

Signed-off-by: Dan Williams <dcbw@redhat.com>
Acked-by: Oliver Neukum <oliver@neukum.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoqmi_wwan/cdc_ether: add device IDs for Dell 5804 (Novatel E371) WWAN card
Dan Williams [Mon, 6 May 2013 11:17:37 +0000 (11:17 +0000)]
qmi_wwan/cdc_ether: add device IDs for Dell 5804 (Novatel E371) WWAN card

A rebranded Novatel E371 for AT&T's LTE bands.  qmi_wwan should drive this
device, while cdc_ether should ignore it.  Even though the USB descriptors
are plain CDC-ETHER that USB interface is a QMI interface.

Cc: <stable@vger.kernel.org>
Signed-off-by: Dan Williams <dcbw@redhat.com>
Acked-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agobe2net: disable TX in be_close()
Sathya Perla [Wed, 8 May 2013 02:05:50 +0000 (02:05 +0000)]
be2net: disable TX in be_close()

be_close() followed by be_clear() is called as a part of cleanup in the
EEH/AER flow.  This patch stops TX in be_close() before cleaning/freeing
up the TX queues in be_clear(). This prevents be_xmit() from being called
while TX queues no longer exist.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agobe2net: fix EQ from getting full while cleaning RX CQ
Sathya Perla [Wed, 8 May 2013 02:05:49 +0000 (02:05 +0000)]
be2net: fix EQ from getting full while cleaning RX CQ

While cleaning RX queues, the CQ DB may be rung several times (with rearm)
while waiting for the flush compl. Each CQ-notify with rearm can result in
an event. The EQ may get full resulting in a HW error.

Fix this by not re-arming the CQ while notifying a valid completion.
Also, there's no need to wait for 1ms after destroying RXQ, as the code in
be_rx_cq_clean() waits for the flush compl to arrive.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agobe2net: fix payload_len value for GET_MAC_LIST cmd req
Sathya Perla [Wed, 8 May 2013 02:05:48 +0000 (02:05 +0000)]
be2net: fix payload_len value for GET_MAC_LIST cmd req

The buffer size for a FW cmd request must be big enough to fit the response,
else the cmd fails. For GET_MAC_LIST cmd, though the memory allocated for
the cmd is big enough to fit the response, the payload_len value in the
WRB hdr is being set to the request length only.
Fix this for GET_MAC_LIST cmd.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agobe2net: provision VF resources before enabling SR-IOV
Sathya Perla [Wed, 8 May 2013 02:05:47 +0000 (02:05 +0000)]
be2net: provision VF resources before enabling SR-IOV

When the PF driver calls pci_enable_sriov(), the VFs may be probed
inline before the call returns. So, the resources required for all VFs
must be provisioned by the PF driver *before* calling pci_enable_sriov();
else, VF probe will fail.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 years agoMerge branch 'for-3.10/drivers' of git://git.kernel.dk/linux-block
Linus Torvalds [Wed, 8 May 2013 18:51:05 +0000 (11:51 -0700)]
Merge branch 'for-3.10/drivers' of git://git.kernel.dk/linux-block

Pull block driver updates from Jens Axboe:
 "It might look big in volume, but when categorized, not a lot of
  drivers are touched.  The pull request contains:

   - mtip32xx fixes from Micron.

   - A slew of drbd updates, this time in a nicer series.

   - bcache, a flash/ssd caching framework from Kent.

   - Fixes for cciss"

* 'for-3.10/drivers' of git://git.kernel.dk/linux-block: (66 commits)
  bcache: Use bd_link_disk_holder()
  bcache: Allocator cleanup/fixes
  cciss: bug fix to prevent cciss from loading in kdump crash kernel
  cciss: add cciss_allow_hpsa module parameter
  drivers/block/mg_disk.c: add CONFIG_PM_SLEEP to suspend/resume functions
  mtip32xx: Workaround for unaligned writes
  bcache: Make sure blocksize isn't smaller than device blocksize
  bcache: Fix merge_bvec_fn usage for when it modifies the bvm
  bcache: Correctly check against BIO_MAX_PAGES
  bcache: Hack around stuff that clones up to bi_max_vecs
  bcache: Set ra_pages based on backing device's ra_pages
  bcache: Take data offset from the bdev superblock.
  mtip32xx: mtip32xx: Disable TRIM support
  mtip32xx: fix a smatch warning
  bcache: Disable broken btree fuzz tester
  bcache: Fix a format string overflow
  bcache: Fix a minor memory leak on device teardown
  bcache: Documentation updates
  bcache: Use WARN_ONCE() instead of __WARN()
  bcache: Add missing #include <linux/prefetch.h>
  ...

12 years agortc: rtc-tile: add missing platform_device_unregister() when module exit
Wei Yongjun [Sun, 5 May 2013 05:07:49 +0000 (01:07 -0400)]
rtc: rtc-tile: add missing platform_device_unregister() when module exit

We have registered platform device when module init, and
need unregister it when module exit.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
12 years agopstore/ram: fix error return code in ramoops_probe()
Wei Yongjun [Tue, 7 May 2013 11:39:20 +0000 (19:39 +0800)]
pstore/ram: fix error return code in ramoops_probe()

Fix to return a negative error code from the error handling
case instead of 0, as done elsewhere in this function.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
12 years agoMerge branch 'for-3.10/core' of git://git.kernel.dk/linux-block
Linus Torvalds [Wed, 8 May 2013 17:13:35 +0000 (10:13 -0700)]
Merge branch 'for-3.10/core' of git://git.kernel.dk/linux-block

Pull block core updates from Jens Axboe:

 - Major bit is Kents prep work for immutable bio vecs.

 - Stable candidate fix for a scheduling-while-atomic in the queue
   bypass operation.

 - Fix for the hang on exceeded rq->datalen 32-bit unsigned when merging
   discard bios.

 - Tejuns changes to convert the writeback thread pool to the generic
   workqueue mechanism.

 - Runtime PM framework, SCSI patches exists on top of these in James'
   tree.

 - A few random fixes.

* 'for-3.10/core' of git://git.kernel.dk/linux-block: (40 commits)
  relay: move remove_buf_file inside relay_close_buf
  partitions/efi.c: replace useless kzalloc's by kmalloc's
  fs/block_dev.c: fix iov_shorten() criteria in blkdev_aio_read()
  block: fix max discard sectors limit
  blkcg: fix "scheduling while atomic" in blk_queue_bypass_start
  Documentation: cfq-iosched: update documentation help for cfq tunables
  writeback: expose the bdi_wq workqueue
  writeback: replace custom worker pool implementation with unbound workqueue
  writeback: remove unused bdi_pending_list
  aoe: Fix unitialized var usage
  bio-integrity: Add explicit field for owner of bip_buf
  block: Add an explicit bio flag for bios that own their bvec
  block: Add bio_alloc_pages()
  block: Convert some code to bio_for_each_segment_all()
  block: Add bio_for_each_segment_all()
  bounce: Refactor __blk_queue_bounce to not use bi_io_vec
  raid1: use bio_copy_data()
  pktcdvd: Use bio_reset() in disabled code to kill bi_idx usage
  pktcdvd: use bio_copy_data()
  block: Add bio_copy_data()
  ...

12 years agoNVMe: Simplify Firmware Activate code slightly
Matthew Wilcox [Mon, 6 May 2013 12:22:18 +0000 (08:22 -0400)]
NVMe: Simplify Firmware Activate code slightly

Add definitions for the three Firmware Activate actions, and change the
SCSI translation code to construct the command into a temporary variable
instead of translating the endianness back-and-forth.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@linux.intel.com>
12 years agoNVMe: Only clear the enable bit when disabling controller
Matthew Wilcox [Sat, 4 May 2013 10:43:17 +0000 (06:43 -0400)]
NVMe: Only clear the enable bit when disabling controller

Many of the bits in the Controller Configuration register may only be
modified when the Enable bit is clear.  Clearing them at the same time
as the Enable bit might be OK, but let's play it safe and only touch the
Enable bit.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
12 years agoNVMe: Wait for device to acknowledge shutdown
Matthew Wilcox [Sat, 4 May 2013 10:43:16 +0000 (06:43 -0400)]
NVMe: Wait for device to acknowledge shutdown

A recent update to the specification makes it clear that the host
is expected to wait for the device to acknowledge the Enable bit
transitioning to 0 as well as waiting for the device to acknowledge a
transition to 1.

Reported-by: Khosrow Panah <Khosrow.Panah@idt.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
12 years agoALSA: Add comment for control TLV API
David Henningsson [Wed, 8 May 2013 13:33:11 +0000 (15:33 +0200)]
ALSA: Add comment for control TLV API

Userspace is not meant to have to handle all strange dB ranges,
so add a specification comment.

Signed-off-by: David Henningsson <david.henningsson@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
12 years agoACPICA: ACPICA: Fix for _INI regression
Tomasz Nowicki [Wed, 8 May 2013 04:02:06 +0000 (04:02 +0000)]
ACPICA: ACPICA: Fix for _INI regression

This change fixes a problem introduced by recent commit c34c82b
(ACPICA: Predefine names: Add allowed argument types to master info
table) in 20130328 where _INI methods are no longer executed properly
because of a memory block that is not initialized properly.  ACPICA
BZ1016. Tomasz Nowicki <tomasz.nowicki@linaro.org>

References: https://bugs.acpica.org/show_bug.cgi?id=1016
Signed-off-by: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
12 years agoACPICA: _OSI support: Fix possible memory leak
Jung-uk Kim [Wed, 8 May 2013 04:01:49 +0000 (04:01 +0000)]
ACPICA: _OSI support: Fix possible memory leak

Fixes a possible memory leak in the error exit path introduced by
recent commit 388a990 ("ACPICA: _OSI Support: handle any errors from
acpi_os_acquire_mutex()").

[rjw: Changelog]
Signed-off-by: Jung-uk Kim <jkim@FreeBSD.org>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
12 years agoACPICA: Fix possible buffer overflow during a field unit read operation
Bob Moore [Wed, 8 May 2013 04:01:15 +0000 (04:01 +0000)]
ACPICA: Fix possible buffer overflow during a field unit read operation

Can only happen under these conditions: 1) The DSDT version is 1,
meaning integers are 32-bits.  2) The field is between 33 and 64
bits long.

It applies cleanly back to ACPICA 20100806+ (Linux v2.6.37+).

Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Cc: 2.6.37+ <stable@vger.kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
12 years agorbd: always set read-only flag in rbd_add()
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: always set read-only flag in rbd_add()

Hold off setting the read-only flag in rbd_add() for an image being
mapped until we have successfully probed the image.  At that point
we know whether it's a snapshot mapping or not, so we can set the
read-only flag in that one place rather than doing so (for
snapshots) in rbd_dev_mapping_set().  To do this, pass a flag to the
image probe routine indicating whether we want a read-only mapping.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: kill rbd_dev_clear_mapping()
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: kill rbd_dev_clear_mapping()

This function is a duplicate of rbd_dev_mapping_clear(), and was
added by mistake.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: don't look up snapshot id in rbd_dev_mapping_set()
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: don't look up snapshot id in rbd_dev_mapping_set()

Currently rbd_dev_mapping_set() looks up the snapshot id for the
snapshot whose name is found in the rbd device's spec structure.

That function gets called by rbd_dev_device_setup(), which is
called by rbd_add() *after* rbd_dev_image_probe().  If the
image probe succeeds, the rbd device's spec will already have
been updated to include names and ids for all fields.

Therefore there's no need to look up the snapshot id in
rbd_dev_mapping_set().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: don't print warning if not mapping a parent
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: don't print warning if not mapping a parent

The presence of the LAYERING bit in an rbd image's feature mask does
not guarantee the image actually has a parent image.  Currently that
bit is set only when a clone (i.e., image with a parent) is created,
but it is (currently) not cleared if that clone gets flattened back
into a "normal" image.  A "parent_id" query will leave the
parent_spec for the image being mapped a null pointer, but will not
return an error.

Currently, whenever an image with the LAYERED feature gets mapped, a
warning about the use of layered images gets printed.  But we don't
want to do this for a flattened image, so print the warning only
if we find there is a parent spec after the probe.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: kill rbd_update_mapping_size()
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: kill rbd_update_mapping_size()

Since rbd_update_mapping_size() is now a trivial wrapper, just open
code it in its two callers.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: update capacity in rbd_dev_refresh()
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: update capacity in rbd_dev_refresh()

When a mapped image changes size, we change the capacity recorded
for the Linux disk associated with it, in rbd_update_mapping_size().
That function is called in two places--the format 1 and format 2
refresh routines.

There is no need to set the capacity while holding the header
semaphore.  Instead, do it in the common rbd_dev_refresh(), using
the logic that's already there to initiate disk revalidation.

Add handling in the request function, just in case a request
that exceeds the capacity of the device comes in (perhaps one
that was started before a refresh shrunk the device).

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agoregulator: dbx500: Make local symbol static
Sachin Kamat [Wed, 8 May 2013 11:05:10 +0000 (16:35 +0530)]
regulator: dbx500: Make local symbol static

power_state_active_get is used only in this file. Make it static.
While at it also move this function definition inside the
CONFIG_REGULATOR_DEBUG macro as it is called only from within it.
This also avoids further build warning related to unused definition.

Signed-off-by: Sachin Kamat <sachin.kamat@linaro.org>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
12 years agorbd: revalidate only for mapping size changes
Alex Elder [Mon, 6 May 2013 12:40:30 +0000 (07:40 -0500)]
rbd: revalidate only for mapping size changes

This commit:
    d98df63e rbd: revalidate_disk upon rbd resize
instituted a call to revalidate_disk() to notify interested parties
that a mapped image has changed size.  This works well, as long as
the the rbd device doesn't map a snapshot.

A snapshot will never change size.  However, the base image the
snapshot is associated with can, and it can do so while the snapshot
is mapped.

The problem is that the test for the size is looking at the size of
the base image, not the size of the mapped snapshot.  This patch
corrects that.

Update the warning message shown in the event of error, and move
it into the callers.

This resolves:
    http://tracker.ceph.com/issues/4911

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agorbd: fix leak of format 2 snapshot context
Alex Elder [Mon, 6 May 2013 13:37:00 +0000 (08:37 -0500)]
rbd: fix leak of format 2 snapshot context

When rbd_dev_v2_refresh() is called, the rbd device already has a
snapshot context associated with it.  But that never gets freed,
the pointer just gets overwritten.

Fix this by dropping the rbd device's reference to the snapshot
context before overwriting the pointer.

Because ceph_put_snap_context() already handles for a null pointer
we don't need to check for that (for the probe case, where no
context has yet been assigned).

This resolves:
    http://tracker.ceph.com/issues/4912

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
12 years agoxen: clear IRQ_NOAUTOEN and IRQ_NOREQUEST
Julien Grall [Tue, 30 Apr 2013 17:29:13 +0000 (18:29 +0100)]
xen: clear IRQ_NOAUTOEN and IRQ_NOREQUEST

Reset the IRQ_NOAUTOEN and IRQ_NOREQUEST flags that are enabled by
default on ARM. If IRQ_NOAUTOEN is set, __setup_irq doesn't call
irq_startup, that is responsible for calling irq_unmask at startup time.
As a result event channels remain masked.

The clear is already made in bind_evtchn_to_irq with commit a8636c0 but was
missing on all others bind_*_to_irq. Move the clear in xen_irq_info_common_init.

On x86, IRQ_NOAUTOEN and IRQ_NOREQUEST are cleared by default, so this commit
doesn't impact this architecture.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Julien Grall <julien.grall@linaro.org>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen: mask x2APIC feature in PV
Zhenzhong Duan [Fri, 29 Mar 2013 01:14:00 +0000 (09:14 +0800)]
xen: mask x2APIC feature in PV

On x2apic enabled pvm, doing sysrq+l, got NULL pointer dereference as below.

    SysRq : Show backtrace of all active CPUs
    BUG: unable to handle kernel NULL pointer dereference at           (null)
    IP: [<ffffffff8125e3cb>] memcpy+0xb/0x120
    Call Trace:
     [<ffffffff81039633>] ? __x2apic_send_IPI_mask+0x73/0x160
     [<ffffffff8103973e>] x2apic_send_IPI_all+0x1e/0x20
     [<ffffffff8103498c>] arch_trigger_all_cpu_backtrace+0x6c/0xb0
     [<ffffffff81501be4>] ? _raw_spin_lock_irqsave+0x34/0x50
     [<ffffffff8131654e>] sysrq_handle_showallcpus+0xe/0x10
     [<ffffffff8131616d>] __handle_sysrq+0x7d/0x140
     [<ffffffff81316230>] ? __handle_sysrq+0x140/0x140
     [<ffffffff81316287>] write_sysrq_trigger+0x57/0x60
     [<ffffffff811ca996>] proc_reg_write+0x86/0xc0
     [<ffffffff8116dd8e>] vfs_write+0xce/0x190
     [<ffffffff8116e3e5>] sys_write+0x55/0x90
     [<ffffffff8150a242>] system_call_fastpath+0x16/0x1b

That's because apic points to apic_x2apic_cluster or apic_x2apic_phys
but the basic element like cpumask isn't initialized.

Mask x2APIC feature in pvm to avoid overwrite of apic pointer,
update commit message per Konrad's suggestion.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Tested-by: Tamon Shiose <tamon.shiose@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen: SWIOTLB is only used on x86
Arnd Bergmann [Thu, 2 May 2013 15:16:22 +0000 (17:16 +0200)]
xen: SWIOTLB is only used on x86

Enabling SWIOTLB_XEN on ARM results in build errors because the
underlying SWIOTLB is only available on X86:

drivers/xen/swiotlb-xen.c: In function 'is_xen_swiotlb_buffer':
drivers/xen/swiotlb-xen.c:105:2: error: implicit declaration of function 'mfn_to_local_pfn

Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen/spinlock: Fix check from greater than to be also be greater or equal to.
Konrad Rzeszutek Wilk [Mon, 6 May 2013 12:33:15 +0000 (08:33 -0400)]
xen/spinlock: Fix check from greater than to be also be greater or equal to.

During review of git commit cb9c6f15f318aa3aeb62fe525aa5c6dcf6eee159
("xen/spinlock:  Check against default value of -1 for IRQ line.")
Stefano pointed out a bug in the patch. Unfortunatly due to vacation
timing the fix was not applied and this patch fixes it up.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoxen/smp/pvhvm: Don't point per_cpu(xen_vpcu, 33 and larger) to shared_info
Konrad Rzeszutek Wilk [Sun, 5 May 2013 13:24:07 +0000 (09:24 -0400)]
xen/smp/pvhvm: Don't point per_cpu(xen_vpcu, 33 and larger) to shared_info

As it will point to some data, but not event channel data (the
shared_info has an array limited to 32).

This means that for PVHVM guests with more than 32 VCPUs without
the usage of VCPUOP_register_info any interrupts to VCPUs
larger than 32 would have gone unnoticed during early bootup.

That is OK, as during early bootup, in smp_init we end up calling
the hotplug mechanism (xen_hvm_cpu_notify) which makes the
VCPUOP_register_vcpu_info call for all VCPUs and we can receive
interrupts on VCPUs 33 and further.

This is just a cleanup.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
12 years agoASoC: ux500: Swap even/odd AD slot definitions
Fabio Baltieri [Wed, 8 May 2013 07:14:20 +0000 (09:14 +0200)]
ASoC: ux500: Swap even/odd AD slot definitions

AD slots definitions for ab8500 codec were erroneously swapped between
even and odd channels.  Fix this by swapping the definitions to be
coherent with the channel number.

Signed-off-by: Fabio Baltieri <fabio.baltieri@linaro.org>
Acked-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
12 years agof2fs: cover free_nid management with spin_lock
Jaegeuk Kim [Tue, 7 May 2013 11:47:40 +0000 (20:47 +0900)]
f2fs: cover free_nid management with spin_lock

After build_free_nids() searches free nid candidates from nat pages and
current journal blocks, it checks all the candidates if they are allocated
so that the nat cache has its nid with an allocated block address.

In this procedure, previously we used
    list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list).
But, this is not covered by free_nid_list_lock, resulting in null pointer bug.

This patch moves this checking routine inside add_free_nid() in order not to use
the spin_lock.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: optimize scan_nat_page()
Haicheng Li [Mon, 6 May 2013 15:15:43 +0000 (23:15 +0800)]
f2fs: optimize scan_nat_page()

When nm_i->fcnt > 2 * MAX_FREE_NIDS, stop scanning other NAT entries.

Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
[Jaegeuk Kim: fix handling the return value of add_free_nid()]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: code cleanup for scan_nat_page() and build_free_nids()
Haicheng Li [Mon, 6 May 2013 15:15:42 +0000 (23:15 +0800)]
f2fs: code cleanup for scan_nat_page() and build_free_nids()

This patch does two cleanups:
1. remove unused variable "fcnt" in build_free_nids().
2. make scan_nat_page() as void type and remove useless variable "fcnt".

Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: bugfix for alloc_nid_failed()
Haicheng Li [Mon, 6 May 2013 15:15:41 +0000 (23:15 +0800)]
f2fs: bugfix for alloc_nid_failed()

Directly drop the free_nid cache when nm_i->fcnt > 2 * MAX_FREE_NIDS

Since there is NOT nmi->free_nid_list_lock spinlock protection between
a sequential calling of alloc_nid() and alloc_nid_failed(), some other
threads may already add new free_nid to the free_nid_list during this
period.

We need to make sure nmi->fcnt is never > 2 * MAX_FREE_NIDS.

Signed-off-by: Haicheng Li <haicheng.li@linux.intel.com>
[Jaegeuk Kim: fit the coding style]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: recover when journal contains deleted files
Chris Fries [Thu, 2 May 2013 21:09:05 +0000 (16:09 -0500)]
f2fs: recover when journal contains deleted files

When recovering a journal file with fsync data for files that have
been deleted, don't bail out on recovery.

Signed-off-by: Chris Fries <C.Fries@motorola.com>
Reviewed-by: Russell Knize <rknize2@motorola.com>
Reviewed-by: Jason Hrycay <jason.hrycay@motorola.com>
[Jaegeuk Kim: fit the coding style]
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: continue to mount after failing recovery
Chris Fries [Thu, 2 May 2013 21:07:34 +0000 (16:07 -0500)]
f2fs: continue to mount after failing recovery

When unable to roll forward the journal, we shouldn't bail out and
not mount, we should continue to attempt the mount.  Bad recovery data
is likely unrecoverable at this point, and requiring the user to try
to mount again doesn't solve any issues.

Signed-off-by: Chris Fries <C.Fries@motorola.com>
Reviewed-by: Russell Knize <rknize2@motorola.com>
Reviewed-by: Jason Hrycay <jason.hrycay@motorola.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agof2fs: avoid deadlock during evict after f2fs_gc
Jaegeuk Kim [Tue, 30 Apr 2013 02:33:27 +0000 (11:33 +0900)]
f2fs: avoid deadlock during evict after f2fs_gc

o Deadlock case #1

Thread 1:
- writeback_sb_inodes
 - do_writepages
  - f2fs_write_data_pages
   - write_cache_pages
    - f2fs_write_data_page
     - f2fs_balance_fs
      - wait mutex_lock(gc_mutex)

Thread 2:
- f2fs_balance_fs
 - mutex_lock(gc_mutex)
 - f2fs_gc
  - f2fs_iget
   - wait iget_locked(inode->i_lock)

Thread 3:
- do_unlinkat
 - iput
  - lock(inode->i_lock)
   - evict
    - inode_wait_for_writeback

o Deadlock case #2

Thread 1:
- __writeback_single_inode
 : set I_SYNC
  - do_writepages
   - f2fs_write_data_page
    - f2fs_balance_fs
     - f2fs_gc
      - iput
       - evict
        - inode_wait_for_writeback(I_SYNC)

In order to avoid this, even though iput is called with the zero-reference
count, we need to stop the eviction procedure if the inode is on writeback.
So this patch links f2fs_drop_inode which checks the I_SYNC flag.

Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
12 years agoMIPS: microMIPS: Support dynamic ASID sizing.
Steven J. Hill [Mon, 25 Mar 2013 17:01:00 +0000 (12:01 -0500)]
MIPS: microMIPS: Support dynamic ASID sizing.

Changes for pure microMIPS cores to dynamically determine the ASID
size at boot time.

Includes bug fix https://patchwork.linux-mips.org/patch/5230/

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
Signed-off-by: Jonas Gorski <jogo@openwrt.org>
12 years agoMIPS: Allow ASID size to be determined at boot time.
Steven J. Hill [Mon, 25 Mar 2013 16:58:57 +0000 (11:58 -0500)]
MIPS: Allow ASID size to be determined at boot time.

Original patch by Ralf Baechle and removed by Harold Koerfgen
with commit f67e4ffc79905482c3b9b8c8dd65197bac7eb508. This
allows for more generic kernels since the size of the ASID
and corresponding masks can be determined at run-time. This
patch is also required for the new Aptiv cores and has been
tested on Malta and Malta Aptiv platforms.

[ralf@linux-mips.org: Added relevant part of fix
https://patchwork.linux-mips.org/patch/5213/]

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: FW: malta: Code formatting clean-ups.
Steven J. Hill [Mon, 25 Mar 2013 20:05:40 +0000 (15:05 -0500)]
MIPS: FW: malta: Code formatting clean-ups.

Clean-up code according to the 'checkpatch.pl' script.

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: FW: Remove obsolete header file for MTI platforms.
Steven J. Hill [Mon, 25 Mar 2013 19:55:32 +0000 (14:55 -0500)]
MIPS: FW: Remove obsolete header file for MTI platforms.

Remove 'arch/mips/include/asm/mips-boards/prom.h' and get rid of
all inclusions of it by Malta and SEAD-3 platforms.

[ralf@linux-mips.org: Fold in John Crispin <blogic@openwrt.org>'s "MIPS:
ar7 powertv build"].

[ralf@linux-mips.org: Fold in John Crispin <blogic@openwrt.org>'s "MIPS:
unbreak powertv build"].

[ralf@linux-mips.org: Test. Build. Your. Fscking. Code. Or...]

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: FW: malta: Use new common FW library variable processing.
Steven J. Hill [Mon, 25 Mar 2013 19:47:05 +0000 (14:47 -0500)]
MIPS: FW: malta: Use new common FW library variable processing.

Remove old YAMON prom code and use common firmware library code.

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: FW: sead3: Use new common FW library variable processing.
Steven J. Hill [Mon, 25 Mar 2013 19:35:30 +0000 (14:35 -0500)]
MIPS: FW: sead3: Use new common FW library variable processing.

Remove old YAMON prom code and use common firmware library code.

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: FW: Add environment variable processing.
Steven J. Hill [Mon, 25 Mar 2013 19:32:15 +0000 (14:32 -0500)]
MIPS: FW: Add environment variable processing.

Add parsing of the environment and command line variables passed to
the kernel to the firmware library.

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: Add declarations to MIPS Technologies Inc. generic header.
Steven J. Hill [Mon, 25 Mar 2013 19:24:42 +0000 (14:24 -0500)]
MIPS: Add declarations to MIPS Technologies Inc. generic header.

Add declaration of 'mips_scroll_message' and 'mips_display_message'
to the common generic header file for the MIPS Technologies Inc.
development boards.

Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoMIPS: sead3: Use generic suspend/resume for LEDs.
Lars-Peter Clausen [Mon, 25 Mar 2013 16:54:17 +0000 (11:54 -0500)]
MIPS: sead3: Use generic suspend/resume for LEDs.

Setting the LED_CORE_SUSPENDRESUME flag causes the LED driver core to call
led_classdev_suspend/led_classdev_resume during suspend/resume. Since this is
exactly what the driver's custom suspend/resume callbacks do we can replace them
by setting the LED_CORE_SUSPENDRESUME flag.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
12 years agoKVM: Fix kvm_irqfd_init initialization
Asias He [Wed, 8 May 2013 02:57:29 +0000 (10:57 +0800)]
KVM: Fix kvm_irqfd_init initialization

In commit a0f155e96 'KVM: Initialize irqfd from kvm_init()', when
kvm_init() is called the second time (e.g kvm-amd.ko and kvm-intel.ko),
kvm_arch_init() will fail with -EEXIST, then kvm_irqfd_exit() will be
called on the error handling path. This way, the kvm_irqfd system will
not be ready.

This patch fix the following:

BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff81c0721e>] _raw_spin_lock+0xe/0x30
PGD 0
Oops: 0002 [#1] SMP
Modules linked in: vhost_net
CPU 6
Pid: 4257, comm: qemu-system-x86 Not tainted 3.9.0-rc3+ #757 Dell Inc. OptiPlex 790/0V5HMK
RIP: 0010:[<ffffffff81c0721e>]  [<ffffffff81c0721e>] _raw_spin_lock+0xe/0x30
RSP: 0018:ffff880221721cc8  EFLAGS: 00010046
RAX: 0000000000000100 RBX: ffff88022dcc003f RCX: ffff880221734950
RDX: ffff8802208f6ca8 RSI: 000000007fffffff RDI: 0000000000000000
RBP: ffff880221721cc8 R08: 0000000000000002 R09: 0000000000000002
R10: 00007f7fd01087e0 R11: 0000000000000246 R12: ffff8802208f6ca8
R13: 0000000000000080 R14: ffff880223e2a900 R15: 0000000000000000
FS:  00007f7fd38488e0(0000) GS:ffff88022dcc0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000022309f000 CR4: 00000000000427e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process qemu-system-x86 (pid: 4257, threadinfo ffff880221720000, task ffff880222bd5640)
Stack:
 ffff880221721d08 ffffffff810ac5c5 ffff88022431dc00 0000000000000086
 0000000000000080 ffff880223e2a900 ffff8802208f6ca8 0000000000000000
 ffff880221721d48 ffffffff810ac8fe 0000000000000000 ffff880221734000
Call Trace:
 [<ffffffff810ac5c5>] __queue_work+0x45/0x2d0
 [<ffffffff810ac8fe>] queue_work_on+0x8e/0xa0
 [<ffffffff810ac949>] queue_work+0x19/0x20
 [<ffffffff81009b6b>] irqfd_deactivate+0x4b/0x60
 [<ffffffff8100a69d>] kvm_irqfd+0x39d/0x580
 [<ffffffff81007a27>] kvm_vm_ioctl+0x207/0x5b0
 [<ffffffff810c9545>] ? update_curr+0xf5/0x180
 [<ffffffff811b66e8>] do_vfs_ioctl+0x98/0x550
 [<ffffffff810c1f5e>] ? finish_task_switch+0x4e/0xe0
 [<ffffffff81c054aa>] ? __schedule+0x2ea/0x710
 [<ffffffff811b6bf7>] sys_ioctl+0x57/0x90
 [<ffffffff8140ae9e>] ? trace_hardirqs_on_thunk+0x3a/0x3c
 [<ffffffff81c0f602>] system_call_fastpath+0x16/0x1b
Code: c1 ea 08 38 c2 74 0f 66 0f 1f 44 00 00 f3 90 0f b6 03 38 c2 75 f7 48 83 c4 08 5b c9 c3 55 48 89 e5 66 66 66 66 90 b8 00 01 00 00 <f0> 66 0f c1 07 89 c2 66 c1 ea 08 38 c2 74 0c 0f 1f 00 f3 90 0f
RIP  [<ffffffff81c0721e>] _raw_spin_lock+0xe/0x30
RSP <ffff880221721cc8>
CR2: 0000000000000000
---[ end trace 13fb1e4b6e5ab21f ]---

Signed-off-by: Asias He <asias@redhat.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
12 years agoKVM: x86: fix maintenance of guest/host xcr0 state
Marcelo Tosatti [Tue, 16 Apr 2013 02:30:13 +0000 (23:30 -0300)]
KVM: x86: fix maintenance of guest/host xcr0 state

Emulation of xcr0 writes zero guest_xcr0_loaded variable so that
subsequent VM-entry reloads CPU's xcr0 with guests xcr0 value.

However, this is incorrect because guest_xcr0_loaded variable is
read to decide whether to reload hosts xcr0.

In case the vcpu thread is scheduled out after the guest_xcr0_loaded = 0
assignment, and scheduler decides to preload FPU:

switch_to
{
  __switch_to
    __math_state_restore
      restore_fpu_checking
        fpu_restore_checking
          if (use_xsave())
              fpu_xrstor_checking
xrstor64 with CPU's xcr0 == guests xcr0

Fix by properly restoring hosts xcr0 during emulation of xcr0 writes.

Analyzed-by: Ulrich Obergfell <uobergfe@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
12 years agoarm64: Treat the bitops index argument as an 'int'
Catalin Marinas [Tue, 7 May 2013 17:02:58 +0000 (18:02 +0100)]
arm64: Treat the bitops index argument as an 'int'

The bitops prototype use an 'int' as the bit index type but the asm
implementation assume it to be a 'long'. Since the compiler does not
guarantee zeroing the upper 32-bits in a register when used as 'int',
change the bitops implementation accordingly.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agoarm64: Ignore the 'write' ESR flag on cache maintenance faults
Catalin Marinas [Tue, 7 May 2013 15:57:06 +0000 (16:57 +0100)]
arm64: Ignore the 'write' ESR flag on cache maintenance faults

ESR.WnR bit is always set on data cache maintenance faults even though
the page is not required to have write permission. If a translation
fault (page not yet mapped) happens for read-only user address range,
Linux incorrectly assumes a permission fault. This patch adds the check
of the ESR.CM bit during the page fault handling to ignore the 'write'
flag.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: Tim Northover <Tim.Northover@arm.com>
Cc: stable@vger.kernel.org
12 years agoarm64: dts: fix #address-cells for foundation-v8
Mark Rutland [Tue, 7 May 2013 13:04:03 +0000 (14:04 +0100)]
arm64: dts: fix #address-cells for foundation-v8

Commit 90556ca1 ("arm64: vexpress: Add dts files for the ARMv8 RTSM
models") added foundation-v8.dts, but erroneously set
/cpus/#address-cells = <1> while providing two cells in each cpus/cpu@N
node's reg property.

As of commit ea393a2e ("arm64: smp: honour #address-size when parsing
CPU reg property") we read in as many address cells as specified rather
than always reading two. This means that for foundation-v8.dts, we only
read the first reg cell (zero) for each cpu node, and receive a lot of
warnings at boot of the form "/cpus/cpu@1: duplicate cpu reg properties
in the DT".

This patch corrects foundation-v8.dts to have the correct value for
/cpus/#address-cells.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Pawel Moll <pawel.moll@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Tested-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agoarm64: vexpress: Add support for poweroff/restart
Catalin Marinas [Thu, 28 Feb 2013 18:14:37 +0000 (18:14 +0000)]
arm64: vexpress: Add support for poweroff/restart

This patch adds the arm_pm_poweroff definition expected by the
vexpress-poweroff.c driver and enables the latter for arm64.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Pawel Moll <pawel.moll@arm.com>
12 years agoarm64: Enable support for the ARM GIC interrupt controller
Catalin Marinas [Mon, 14 Jan 2013 12:39:31 +0000 (12:39 +0000)]
arm64: Enable support for the ARM GIC interrupt controller

This patch enables ARM_GIC on the arm64 kernel.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
12 years agoMerge branch 'gic' into HEAD
Catalin Marinas [Wed, 8 May 2013 09:22:07 +0000 (10:22 +0100)]
Merge branch 'gic' into HEAD

* arm64-prep-gic:
  irqchip: gic: Perform the gic_secondary_init() call via CPU notifier
  irqchip: gic: Call handle_bad_irq() directly
  arm: Move chained_irq_(enter|exit) to a generic file
  arm: Move the set_handle_irq and handle_arch_irq declarations to asm/irq.h

12 years agoALSA: hda - Apply pin-enablement workaround to all Haswell HDMI codecs
Takashi Iwai [Wed, 8 May 2013 06:09:34 +0000 (08:09 +0200)]
ALSA: hda - Apply pin-enablement workaround to all Haswell HDMI codecs

This is a revised patch based on Mengdong Lin's fix patch, which is a
supplement to a previous patch [1611a9c9: ALSA: hda - Add fixup for
Haswell to enable all pin and convertor widgets].

Some Haswell BIOS will disable the 2nd and 3rd pin/covertor widgets
when the HD-A controller changes state from D3 to D0.  So when the
controller resumes after a system or runtime suspend, these widgets
are disabled and programming these widgets to D0 will cause H/W error
and codec will not respond.

In addition, we found out that some BIOS disables the pins at S3
although it shows up at boot.  This confuses the driver utterly, and
the hardware falls into the fatal communication error like the above.

So in this patch, we apply intel_haswell_enable_all_pins() not only as
a fixup to a certain device (with 8086:2010) but to all Haswell
machines.  The codec driver basically assumes that all pins are
exposed, so it's anyway better to see them from the beginning.  Even
if all pins and converters are shown by this call, there should be no
regression in practice: the pin default configurations are still kept,
thus the disabled pins are handled as disabled by the driver
properly.

Signed-off-by: Mengdong Lin <mengdong.lin@intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
12 years agopowerpc/powernv: Properly drop characters if console is closed
Benjamin Herrenschmidt [Wed, 8 May 2013 04:14:26 +0000 (14:14 +1000)]
powerpc/powernv: Properly drop characters if console is closed

If the firmware returns an error such as "closed" (or hardware
error), we should drop characters.

Currently we only do that when a firmware compatible with OPAL v2
APIs is detected, in the code that calls opal_console_write_buffer_space(),
which didn't exist with OPAL v1 (or didn't work).

However, when enabling early debug consoles, the flag indicating
that v2 is supported isn't set yet, causing us, in case of errors
or closed console, to spin forever.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
12 years agoaudit: fix message spacing printing auid
Eric Paris [Wed, 8 May 2013 04:01:07 +0000 (00:01 -0400)]
audit: fix message spacing printing auid

The helper function didn't include a leading space, so it was jammed
against the previous text in the audit record.

Signed-off-by: Eric Paris <eparis@redhat.com>
12 years agoMerge branch 'akpm' (incoming from Andrew)
Linus Torvalds [Wed, 8 May 2013 03:49:51 +0000 (20:49 -0700)]
Merge branch 'akpm' (incoming from Andrew)

Merge more incoming from Andrew Morton:

 - Various fixes which were stalled or which I picked up recently

 - A large rotorooting of the AIO code.  Allegedly to improve
   performance but I don't really have good performance numbers (I might
   have lost the email) and I can't raise Kent today.  I held this out
   of 3.9 and we could give it another cycle if it's all too late/scary.

I ended up taking only the first two thirds of the AIO rotorooting.  I
left the percpu parts and the batch completion for later.  - Linus

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (33 commits)
  aio: don't include aio.h in sched.h
  aio: kill ki_retry
  aio: kill ki_key
  aio: give shared kioctx fields their own cachelines
  aio: kill struct aio_ring_info
  aio: kill batch allocation
  aio: change reqs_active to include unreaped completions
  aio: use cancellation list lazily
  aio: use flush_dcache_page()
  aio: make aio_read_evt() more efficient, convert to hrtimers
  wait: add wait_event_hrtimeout()
  aio: refcounting cleanup
  aio: make aio_put_req() lockless
  aio: do fget() after aio_get_req()
  aio: dprintk() -> pr_debug()
  aio: move private stuff out of aio.h
  aio: add kiocb_cancel()
  aio: kill return value of aio_complete()
  char: add aio_{read,write} to /dev/{null,zero}
  aio: remove retry-based AIO
  ...

12 years agoaio: don't include aio.h in sched.h
Kent Overstreet [Tue, 7 May 2013 23:19:08 +0000 (16:19 -0700)]
aio: don't include aio.h in sched.h

Faster kernel compiles by way of fewer unnecessary includes.

[akpm@linux-foundation.org: fix fallout]
[akpm@linux-foundation.org: fix build]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: kill ki_retry
Kent Overstreet [Tue, 7 May 2013 23:19:11 +0000 (16:19 -0700)]
aio: kill ki_retry

Thanks to Zach Brown's work to rip out the retry infrastructure, we don't
need this anymore - ki_retry was only called right after the kiocb was
initialized.

This also refactors and trims some duplicated code, as well as cleaning up
the refcounting/error handling a bit.

[akpm@linux-foundation.org: use fmode_t in aio_run_iocb()]
[akpm@linux-foundation.org: fix file_start_write/file_end_write tests]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: kill ki_key
Kent Overstreet [Tue, 7 May 2013 23:19:10 +0000 (16:19 -0700)]
aio: kill ki_key

ki_key wasn't actually used for anything previously - it was always 0.
Drop it to trim struct kiocb a bit.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoRevert "audit: move kaudit thread start from auditd registration to kaudit init"
Eric Paris [Wed, 8 May 2013 01:24:02 +0000 (21:24 -0400)]
Revert "audit: move kaudit thread start from auditd registration to kaudit init"

This reverts commit 6ff5e45985c2fcb97947818f66d1eeaf9d6600b2.

Conflicts:
kernel/audit.c

This patch was starting a kthread for all the time.  Since the follow on
patches that required it didn't get finished in 3.10 time, we shouldn't
ship this change in 3.10.

Signed-off-by: Eric Paris <eparis@redhat.com>
12 years agoaudit: vfs: fix audit_inode call in O_CREAT case of do_last
Jeff Layton [Fri, 12 Apr 2013 19:16:32 +0000 (15:16 -0400)]
audit: vfs: fix audit_inode call in O_CREAT case of do_last

Jiri reported a regression in auditing of open(..., O_CREAT) syscalls.
In older kernels, creating a file with open(..., O_CREAT) created
audit_name records that looked like this:

type=PATH msg=audit(1360255720.628:64): item=1 name="/abc/foo" inode=138810 dev=fd:00 mode=0100640 ouid=0 ogid=0 rdev=00:00 obj=unconfined_u:object_r:default_t:s0
type=PATH msg=audit(1360255720.628:64): item=0 name="/abc/" inode=138635 dev=fd:00 mode=040750 ouid=0 ogid=0 rdev=00:00 obj=unconfined_u:object_r:default_t:s0

...in recent kernels though, they look like this:

type=PATH msg=audit(1360255402.886:12574): item=2 name=(null) inode=264599 dev=fd:00 mode=0100640 ouid=0 ogid=0 rdev=00:00 obj=unconfined_u:object_r:default_t:s0
type=PATH msg=audit(1360255402.886:12574): item=1 name=(null) inode=264598 dev=fd:00 mode=040750 ouid=0 ogid=0 rdev=00:00 obj=unconfined_u:object_r:default_t:s0
type=PATH msg=audit(1360255402.886:12574): item=0 name="/abc/foo" inode=264598 dev=fd:00 mode=040750 ouid=0 ogid=0 rdev=00:00 obj=unconfined_u:object_r:default_t:s0

Richard bisected to determine that the problems started with commit
bfcec708, but the log messages have changed with some later
audit-related patches.

The problem is that this audit_inode call is passing in the parent of
the dentry being opened, but audit_inode is being called with the parent
flag false. This causes later audit_inode and audit_inode_child calls to
match the wrong entry in the audit_names list.

This patch simply sets the flag to properly indicate that this inode
represents the parent. With this, the audit_names entries are back to
looking like they did before.

Cc: <stable@vger.kernel.org> # v3.7+
Reported-by: Jiri Jaburek <jjaburek@redhat.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Test By: Richard Guy Briggs <rbriggs@redhat.com>
Signed-off-by: Eric Paris <eparis@redhat.com>
12 years agoaudit: Make testing for a valid loginuid explicit.
Eric W. Biederman [Tue, 9 Apr 2013 09:22:10 +0000 (02:22 -0700)]
audit: Make testing for a valid loginuid explicit.

audit rule additions containing "-F auid!=4294967295" were failing
with EINVAL because of a regression caused by e1760bd.

Apparently some userland audit rule sets want to know if loginuid uid
has been set and are using a test for auid != 4294967295 to determine
that.

In practice that is a horrible way to ask if a value has been set,
because it relies on subtle implementation details and will break
every time the uid implementation in the kernel changes.

So add a clean way to test if the audit loginuid has been set, and
silently convert the old idiom to the cleaner and more comprehensible
new idiom.

Cc: <stable@vger.kernel.org> # 3.7
Reported-By: Richard Guy Briggs <rgb@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Tested-by: Richard Guy Briggs <rgb@redhat.com>
Signed-off-by: Eric Paris <eparis@redhat.com>
12 years agoMIPS: Export symbols used by KVM/MIPS module
Sanjay Lal [Thu, 22 Nov 2012 02:34:14 +0000 (18:34 -0800)]
MIPS: Export symbols used by KVM/MIPS module

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: ASM offsets for VCPU arch specific fields.
Sanjay Lal [Thu, 22 Nov 2012 02:34:12 +0000 (18:34 -0800)]
MIPS: ASM offsets for VCPU arch specific fields.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: If KVM is enabled then use the KVM specific routine to flush the TLBs on a...
Sanjay Lal [Thu, 22 Nov 2012 02:34:11 +0000 (18:34 -0800)]
MIPS: If KVM is enabled then use the KVM specific routine to flush the TLBs on a ASID wrap.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: Export routines needed by the KVM module.
Sanjay Lal [Thu, 22 Nov 2012 02:34:10 +0000 (18:34 -0800)]
MIPS: Export routines needed by the KVM module.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Routines to handle specific traps/exceptions while executing the guest.
Sanjay Lal [Thu, 22 Nov 2012 02:34:09 +0000 (18:34 -0800)]
KVM/MIPS32: Routines to handle specific traps/exceptions while executing the guest.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Guest interrupt delivery.
Sanjay Lal [Thu, 22 Nov 2012 02:34:08 +0000 (18:34 -0800)]
KVM/MIPS32: Guest interrupt delivery.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: COP0 accesses profiling.
Sanjay Lal [Thu, 22 Nov 2012 02:34:07 +0000 (18:34 -0800)]
KVM/MIPS32: COP0 accesses profiling.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Release notes and KVM module Makefile
Sanjay Lal [Thu, 22 Nov 2012 02:34:06 +0000 (18:34 -0800)]
KVM/MIPS32: Release notes and KVM module Makefile

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: MMU/TLB operations for the Guest.
Sanjay Lal [Thu, 22 Nov 2012 02:34:05 +0000 (18:34 -0800)]
KVM/MIPS32: MMU/TLB operations for the Guest.

- Note that this file is statically linked with the rest of the host kernel (KSEG0). This is because kernel modules are
loaded into mapped space on MIPS and we want to make sure that we don't get any host kernel TLB faults while
manipulating TLBs.
- Virtual Guest TLBs are implemented as 64 entry array regardless of the number of host TLB entries.
- Shadow TLBs map Guest virtual addresses to Host physical addresses.

    - TLB miss handling details:
        Guest KSEG0 TLBMISS (0x40000000 – 0x60000000): Transparent to the Guest.
        Guest KSEG2/3 (0x60000000 – 0x80000000) & Guest UM TLBMISS (0x00000000 – 0x40000000)
            Lookup in Guest/Virtual TLB
            If an entry doesn’t match
                deliver appropriate TLBMISS LD/ST exception to the guest
            If entry does exist in the Guest TLB and is NOT Valid
                Deliver TLB invalid exception to the guest
            If entry does exist in the Guest TLB and is VALID
                Inject the TLB entry into the Shadow TLB

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Privileged instruction/target branch emulation.
Sanjay Lal [Thu, 22 Nov 2012 02:34:04 +0000 (18:34 -0800)]
KVM/MIPS32: Privileged instruction/target branch emulation.

- The Guest kernel is run in UM and privileged instructions cause a trap.
- If the instruction causing the trap is in a branch delay slot, the branch
  needs to be emulated to figure out the PC @ which the guest will resume
  execution.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: KVM Guest kernel support.
Sanjay Lal [Thu, 22 Nov 2012 02:34:03 +0000 (18:34 -0800)]
KVM/MIPS32: KVM Guest kernel support.

Both Guest kernel and Guest Userspace execute in UM. The memory map is as follows:
Guest User address space:   0x00000000 -> 0x40000000
Guest Kernel Unmapped:      0x40000000 -> 0x60000000
Guest Kernel Mapped:        0x60000000 -> 0x80000000
- Guest Usermode virtual memory is limited to 1GB.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: MIPS arch specific APIs for KVM
Sanjay Lal [Thu, 22 Nov 2012 02:34:02 +0000 (18:34 -0800)]
KVM/MIPS32: MIPS arch specific APIs for KVM

- Implements the arch specific APIs for KVM, some are stubs for MIPS
- kvm_mips_handle_exit(): Main 'C' distpatch routine for handling exceptions while in "Guest" mode.
- Also implements in-kernel timer interrupt support for the guest.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Entry point for trampolining to the guest and trap handlers.
Sanjay Lal [Thu, 22 Nov 2012 02:34:01 +0000 (18:34 -0800)]
KVM/MIPS32: Entry point for trampolining to the guest and trap handlers.

- __kvm_mips_vcpu_run: main entry point to enter guest, we save kernel context, load
  up guest context from and ERET to guest context.
- mips32_exception: L1 exception handler(s), save k0/k1 and jump to main handlers.
- mips32_GuestException: Generic exception handlers for exceptions/interrupts while in
  guest context.  Save guest context, restore some kernel context and jump to
  main 'C' handler: kvm_mips_handle_exit()

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Arch specific KVM data structures.
Sanjay Lal [Thu, 22 Nov 2012 02:34:00 +0000 (18:34 -0800)]
KVM/MIPS32: Arch specific KVM data structures.

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoKVM/MIPS32: Infrastructure/build files.
Sanjay Lal [Thu, 22 Nov 2012 02:33:59 +0000 (18:33 -0800)]
KVM/MIPS32: Infrastructure/build files.

- Add the KVM option to MIPS build files.
- Add default config files for KVM host/guest kernels.
- Change the link address for the Malta KVM Guest kernel to UM (0x40100000).
- Add KVM Kconfig file with KVM/MIPS specific options

Signed-off-by: Sanjay Lal <sanjayl@kymasys.com>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoMIPS: IP27: Remove pfn_t.
Ralf Baechle [Wed, 8 May 2013 01:51:58 +0000 (03:51 +0200)]
MIPS: IP27: Remove pfn_t.

In the Linux kernel traditionally pfns are represented by an unsigned long.
However a few bits of the SGI IP27 platform code that were ported from
IRIX are using pfn_t for historic reasons.  This is conflicting with
KVM's use of pfn_t.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
12 years agoaio: give shared kioctx fields their own cachelines
Kent Overstreet [Tue, 7 May 2013 23:18:56 +0000 (16:18 -0700)]
aio: give shared kioctx fields their own cachelines

[akpm@linux-foundation.org: make reqs_active __cacheline_aligned_in_smp]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: kill struct aio_ring_info
Kent Overstreet [Tue, 7 May 2013 23:18:55 +0000 (16:18 -0700)]
aio: kill struct aio_ring_info

struct aio_ring_info was kind of odd, the only place it's used is where
it's embedded in struct kioctx - there's no real need for it.

The next patch rearranges struct kioctx and puts various things on their
own cachelines - getting rid of struct aio_ring_info now makes that
reordering a bit clearer.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: kill batch allocation
Kent Overstreet [Tue, 7 May 2013 23:18:53 +0000 (16:18 -0700)]
aio: kill batch allocation

Previously, allocating a kiocb required touching quite a few global
(well, per kioctx) cachelines...  so batching up allocation to amortize
those was worthwhile.  But we've gotten rid of some of those, and in
another couple of patches kiocb allocation won't require writing to any
shared cachelines, so that means we can just rip this code out.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: change reqs_active to include unreaped completions
Kent Overstreet [Tue, 7 May 2013 23:18:51 +0000 (16:18 -0700)]
aio: change reqs_active to include unreaped completions

The aio code tries really hard to avoid having to deal with the
completion ringbuffer overflowing.  To do that, it has to keep track of
the number of outstanding kiocbs, and the number of completions
currently in the ringbuffer - and it's got to check that every time we
allocate a kiocb.  Ouch.

But - we can improve this quite a bit if we just change reqs_active to
mean "number of outstanding requests and unreaped completions" - that
means kiocb allocation doesn't have to look at the ringbuffer, which is
a fairly significant win.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: use cancellation list lazily
Kent Overstreet [Tue, 7 May 2013 23:18:49 +0000 (16:18 -0700)]
aio: use cancellation list lazily

Cancelling kiocbs requires adding them to a per kioctx linked list,
which is one of the few things we need to take the kioctx lock for in
the fast path.  But most kiocbs can't be cancelled - so if we just do
this lazily, we can avoid quite a bit of locking overhead.

While we're at it, instead of using a flag bit switch to using ki_cancel
itself to indicate that a kiocb has been cancelled/completed.  This lets
us get rid of ki_flags entirely.

[akpm@linux-foundation.org: remove buggy BUG()]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: use flush_dcache_page()
Kent Overstreet [Tue, 7 May 2013 23:18:47 +0000 (16:18 -0700)]
aio: use flush_dcache_page()

This wasn't causing problems before because it's not needed on x86, but
it is needed on other architectures.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: make aio_read_evt() more efficient, convert to hrtimers
Kent Overstreet [Tue, 7 May 2013 23:18:45 +0000 (16:18 -0700)]
aio: make aio_read_evt() more efficient, convert to hrtimers

Previously, aio_read_event() pulled a single completion off the
ringbuffer at a time, locking and unlocking each time.  Change it to
pull off as many events as it can at a time, and copy them directly to
userspace.

This also fixes a bug where if copying the event to userspace failed,
we'd lose the event.

Also convert it to wait_event_interruptible_hrtimeout(), which
simplifies it quite a bit.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agowait: add wait_event_hrtimeout()
Kent Overstreet [Tue, 7 May 2013 23:18:43 +0000 (16:18 -0700)]
wait: add wait_event_hrtimeout()

Analagous to wait_event_timeout() and friends, this adds
wait_event_hrtimeout() and wait_event_interruptible_hrtimeout().

Note that unlike the versions that use regular timers, these don't
return the amount of time remaining when they return - instead, they
return 0 or -ETIME if they timed out.  because I was uncomfortable with
the semantics of doing it the other way (that I could get it right,
anyways).

If the timer expires, there's no real guarantee that expire_time -
current_time would be <= 0 - due to timer slack certainly, and I'm not
sure I want to know the implications of the different clock bases in
hrtimers.

If the timer does expire and the code calculates that the time remaining
is nonnegative, that could be even worse if the calling code then reuses
that timeout.  Probably safer to just return 0 then, but I could imagine
weird bugs or at least unintended behaviour arising from that too.

I came to the conclusion that if other users end up actually needing the
amount of time remaining, the sanest thing to do would be to create a
version that uses absolute timeouts instead of relative.

[akpm@linux-foundation.org: fix description of `timeout' arg]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: refcounting cleanup
Kent Overstreet [Tue, 7 May 2013 23:18:41 +0000 (16:18 -0700)]
aio: refcounting cleanup

The usage of ctx->dead was fubar - it makes no sense to explicitly check
it all over the place, especially when we're already using RCU.

Now, ctx->dead only indicates whether we've dropped the initial
refcount. The new teardown sequence is:

  set ctx->dead
  hlist_del_rcu();
  synchronize_rcu();

Now we know no system calls can take a new ref, and it's safe to drop
the initial ref:

  put_ioctx();

We also need to ensure there are no more outstanding kiocbs.  This was
done incorrectly - it was being done in kill_ctx(), and before dropping
the initial refcount.  At this point, other syscalls may still be
submitting kiocbs!

Now, we cancel and wait for outstanding kiocbs in free_ioctx(), after
kioctx->users has dropped to 0 and we know no more iocbs could be
submitted.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: make aio_put_req() lockless
Kent Overstreet [Tue, 7 May 2013 23:18:39 +0000 (16:18 -0700)]
aio: make aio_put_req() lockless

Freeing a kiocb needed to touch the kioctx for three things:

 * Pull it off the reqs_active list
 * Decrementing reqs_active
 * Issuing a wakeup, if the kioctx was in the process of being freed.

This patch moves these to aio_complete(), for a couple reasons:

 * aio_complete() already has to issue the wakeup, so if we drop the
   kioctx refcount before aio_complete does its wakeup we don't have to
   do it twice.
 * aio_complete currently has to take the kioctx lock, so it makes sense
   for it to pull the kiocb off the reqs_active list too.
 * A later patch is going to change reqs_active to include unreaped
   completions - this will mean allocating a kiocb doesn't have to look
   at the ringbuffer. So taking the decrement of reqs_active out of
   kiocb_free() is useful prep work for that patch.

This doesn't really affect cancellation, since existing (usb) code that
implements a cancel function still calls aio_complete() - we just have
to make sure that aio_complete does the necessary teardown for cancelled
kiocbs.

It does affect code paths where we free kiocbs that were never
submitted; they need to decrement reqs_active and pull the kiocb off the
reqs_active list.  This occurs in two places: kiocb_batch_free(), which
is going away in a later patch, and the error path in io_submit_one.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: do fget() after aio_get_req()
Kent Overstreet [Tue, 7 May 2013 23:18:37 +0000 (16:18 -0700)]
aio: do fget() after aio_get_req()

aio_get_req() will fail if we have the maximum number of requests
outstanding, which depending on the application may not be uncommon.  So
avoid doing an unnecessary fget().

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: dprintk() -> pr_debug()
Kent Overstreet [Tue, 7 May 2013 23:18:35 +0000 (16:18 -0700)]
aio: dprintk() -> pr_debug()

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: move private stuff out of aio.h
Kent Overstreet [Tue, 7 May 2013 23:18:33 +0000 (16:18 -0700)]
aio: move private stuff out of aio.h

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: add kiocb_cancel()
Kent Overstreet [Tue, 7 May 2013 23:18:31 +0000 (16:18 -0700)]
aio: add kiocb_cancel()

Minor refactoring, to get rid of some duplicated code

[akpm@linux-foundation.org: fix warning]
Signed-off-by: Kent Overstreet <koverstreet@google.com>
Cc: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoaio: kill return value of aio_complete()
Kent Overstreet [Tue, 7 May 2013 23:18:29 +0000 (16:18 -0700)]
aio: kill return value of aio_complete()

Nothing used the return value, and it probably wasn't possible to use it
safely for the locked versions (aio_complete(), aio_put_req()).  Just
kill it.

Signed-off-by: Kent Overstreet <koverstreet@google.com>
Acked-by: Zach Brown <zab@redhat.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Asai Thambi S P <asamymuthupa@micron.com>
Cc: Selvan Mani <smani@micron.com>
Cc: Sam Bradshaw <sbradshaw@micron.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Reviewed-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>