If there is a DMA zone (usually 24bit = 16MB I believe), but no DMA32
zone, as is the case for some 32-bit kernels, then massage_gfp_flags()
will cause DMA memory allocated for devices with a 32..63-bit
coherent_dma_mask to fall back to using __GFP_DMA, even though there may
only be 32-bits of physical address available anyway.
Correct that case to compare against a mask the size of phys_addr_t
instead of always using a 64-bit mask.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Fixes: a2e715a86c ("MIPS: DMA: Fix computation of DMA flags from device's coherent_dma_mask.")
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: <stable@vger.kernel.org> # 2.6.36+
Patchwork: https://patchwork.linux-mips.org/patch/9610/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The calculation for the SMT scaling factor for a hardware thread
which has been partially idle needs to disregard the cycles spent
by the other threads of the core while the thread is idle.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
drm_kms_helper_poll_enable() is called from a context in
intel_hpd_irq_storm_disable() where the the mode_config mutex is
already locked.
When this function was converted to lock this mutex in
commit 8c4ccc4ab6 ("drm/probe-helper: Grab mode_config.mutex
in poll_init/enable") a deadlock occurred.
Call the newly implemented non-locking version of this function.
Changes since v1:
- use function name suffix '_locked' for the function that
is to be called from a locked context.
Signed-off-by: Egbert Eich <eich@suse.de>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
drm_kms_helper_poll_enable() was converted to lock the mode_config
mutex in commit 8c4ccc4ab6
("drm/probe-helper: Grab mode_config.mutex in poll_init/enable").
This disregarded the cases where this function is called from a context
where this mutex is already locked.
Add a non-locking version as well.
Changes since v1:
- use function name suffix '_locked' for the function that
is to be called from a locked context.
Signed-off-by: Egbert Eich <eich@suse.de>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
When get a CRC error, start the mmc_retune, it will issue CMD19/CMD21
to do tune, assume there were 10 clock phase need to try, phase 0 to
phase 6 is ok, phase 7 to phase 9 is NG, we try it from 0 to 9, so
the last CMD19/CMD21 will get CRC error, host->need_retune was set and
cause mmc_retune was called, then dead loop of mmc_retune
Signed-off-by: Chaotian Jing <chaotian.jing@mediatek.com>
Acked-by: Adrian Hunter <adrian.hunter@intel.com>
Fixes: bd11e8bd03 ("mmc: core: Flag re-tuning is needed on CRC errors")
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
When we're out of command buffer space, we turn on the command buffer
processed irq without re-checking for finished command buffers afterwards.
This might lead to a missed irq and the command submission process waiting
forever for space.
Fix this by rerunning the command buffer submission handler whenever we're
out of command space. This ensures both that we don't needlessly turn on
the irq, and that if we decide to turn on the irq, we recheck for finished
command buffers before going to sleep.
Reported-and-tested-by: Bryan Li <ldexin@vmware.com>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
In case we have less than maximum allowed channels (8) and autoconfiguration is
enabled the DWC_PARAMS read is wrong because it uses different arithmetic to
what is needed for channel priority setup.
Re-do the caclulations properly. This now works on AVR32 board well.
Fixes: fed2574b3c (dw_dmac: introduce software emulation of LLP transfers)
Cc: yitian.bu@tangramtek.com
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The only thing mode_fixup was doing was set the adjusted_mode->vrefresh to
60, but it already has the value of 60 when the decon_mode_fixup() is
called. That means this call is actually pointless and can be removed.
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The only thing mode_fixup was doing was set the adjusted_mode->vrefresh to
60, but it already has the value of 60 when the fimd_mode_fixup() is
called. That means this call is actually pointless and can be removed.
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
This patch fixes an over flow issue with the TX ring descriptor. Each
descriptor is 32B in size and an operation requires 2 of these
descriptors.
Signed-off-by: Rameshwar Prasad Sahu <rsahu@apm.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
dma_release_channel() decrements privatecnt counter and almost all dma_get*
function increments it with the exception of dma_get_slave_channel().
In most cases this does not cause issue since normally the channel is not
requested and released, but if a driver requests DMA channel via
dma_get_slave_channel() and releases the channel the privatecnt will be
unbalanced and this will prevent for example getting channel for memcpy.
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Protect the rotator_clk_crtl() function with an #ifdef CONFIG_PM guard
to avoid "defined but not used" warnings.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Protect the fimc_clk_ctrl() function with an #ifdef CONFIG_PM guard to
avoid "defined but not used" warnings.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Protect the suspend and resume callbacks with an #ifdef CONFIG_PM_SLEEP
guard to avoid "defined but not used" warnings.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Don't create a fake mmap offset in exynos_drm_gem_dumb_map_offset. If
not, it will call drm_gem_create_mmap_offset whenever user requests
DRM_IOCTL_MODE_MAP_DUMB ioctl.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The drm_gem_object_release() function already performs this cleanup,
so there is no reason to do it explicitly.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
All the user of these functions are inside exynos-drm driver and
you don't need to export the symbols for that case.
Signed-off-by: Daniel Kurtz <djkurtz@chromium.org>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The beginning of statement in function is next line of a brace.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
By if statment, some function callings are written twice. It needs
several line feed by indentation in if statment. Make to one function
calling from outside if statment.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
The exynos_drm_gem_init() is used only in exynos_drm_gem.c file. Make it
static and don't export it.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
They will be freed right or was freed already, so NULL assignment is
unnecessary.
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
When obj->import_attach is existed, code calling drm_prime_gem_destroy()
was removed from commit 67e93c808b ("drm/exynos: stop copying sg
table"), and it's a fault.
The drm_prime_gem_destroy() is cleanup function which GEM drivers need
to call when they use drm_gem_prime_import() to import dma-bufs, so
exynos-drm driver using drm_gem_prime_import() needs calling
drm_prime_gem_destroy().
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
There is no guarantee that DMA addresses are the same as physical
addresses, but dma_to_pfn() knows how to convert a dma_addr_t to a PFN
which can then be converted to a struct page.
Suggested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Joonyoung Shim <jy0922.shim@samsung.com>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Currently, sun4i_dma_free_contract iterates over lists and frees memory
as it goes through them, causing reads to recently freed memory to
be performed. Fix this by using the safe version of the iterator, so
freed memory is not referenced at all.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Emilio López <emilio@elopez.com.ar>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Recent changes in the Hyper-V driver:
b4370df2b1 ("Drivers: hv: vmbus: add special crash handler")
broke the build when CONFIG_KEXEC_CORE is not set:
arch/x86/built-in.o: In function `hv_machine_crash_shutdown':
arch/x86/kernel/cpu/mshyperv.c:112: undefined reference to `native_machine_crash_shutdown'
Decorate all kexec related code with #ifdef CONFIG_KEXEC_CORE.
Reported-by: Jim Davis <jim.epost@gmail.com>
Reported-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: devel@linuxdriverproject.org
Cc: K. Y. Srinivasan <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1443002577-25370-1-git-send-email-vkuznets@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The fib_table_lookup tracepoint found 2 places where the flowi4_flags is
not initialized.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Because of the default 0 value of ret in dsa_slave_port_attr_set, a
driver may return -EOPNOTSUPP from the commit phase of a STP state,
which triggers a WARN() from switchdev.
This happened on a 6185 switch which does not support hardware bridging.
Fixes: 3563606258 ("switchdev: convert STP update to switchdev attr set")
Reported-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When compiling Documentation/ptp/testptp.c the following compiler
warnings are printed out:
Documentation/ptp/testptp.c: In function ‘main’:
Documentation/ptp/testptp.c:367:11: warning: format ‘%lld’ expects argument
of type ‘long long int’, but argument 3 has type ‘__s64’ [-Wformat=]
event.t.sec, event.t.nsec);
^
Documentation/ptp/testptp.c:505:5: warning: format ‘%lld’ expects argument
of type ‘long long int’, but argument 2 has type ‘__s64’ [-Wformat=]
(pct+2*i)->sec, (pct+2*i)->nsec);
^
Documentation/ptp/testptp.c:507:5: warning: format ‘%lld’ expects argument
of type ‘long long int’, but argument 2 has type ‘__s64’ [-Wformat=]
(pct+2*i+1)->sec, (pct+2*i+1)->nsec);
^
Documentation/ptp/testptp.c:509:5: warning: format ‘%lld’ expects argument
of type ‘long long int’, but argument 2 has type ‘__s64’ [-Wformat=]
(pct+2*i+2)->sec, (pct+2*i+2)->nsec);
This happens because __s64 is by default defined as "long" on ppc64,
not as "long long". However, to fix these warnings, it's possible to
define the __SANE_USERSPACE_TYPES__ so that __s64 gets defined to
"long long" on ppc64, too.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both new_steering_entry() and existing_steering_entry() return values
based on their success or failure, but currently they fall through
silently. This can make troubleshooting difficult, as we were unable
to tell which one of these two functions returned errors or
specifically what code was returned. This patch remedies that
situation by passing the return codes to err, which is returned by
mlx4_qp_attach_common() itself.
This also addresses a leak in the call to mlx4_bitmap_free() as well.
Signed-off-by: Robb Manes <rmanes@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Frames destined to an unknown address must be forwarded to the CPU
port. Otherwise incoming ARP, dhcp leases, etc, do not work.
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Earlier patch 6ae459bda tried to detect void ckecksum partial
skb by comparing pull length to checksum offset. But it does
not work for all cases since checksum-offset depends on
updates to skb->data.
Following patch fixes it by validating checksum start offset
after skb-data pointer is updated. Negative value of checksum
offset start means there is no need to checksum.
Fixes: 6ae459bda ("skbuff: Fix skb checksum flag on skb pull")
Reported-by: Andrew Vagin <avagin@odin.com>
Signed-off-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The data race happens on ps2dev->cmdcnt and ps2dev->cmdbuf contents.
__ps2_command reads that data concurrently with the interrupt handler. As
the result, for example, if a response arrives just after the timeout,
__ps2_command can copy out garbage from ps2dev->cmdbuf but then see that
ps2dev->cmdcnt is 0 and return success.
Stop the interrupt handler with serio_pause_rx() before reading the
results.
The data race was found with KernelThreadSanitizer (KTSAN).
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Wolfgang reported that IPv6 stack is ignoring oif in output route lookups:
With ipv6, ip -6 route get always returns the specific route.
$ ip -6 r
2001:db8:e2::1 dev enp2s0 proto kernel metric 256
2001:db8:e2::/64 dev enp2s0 metric 1024
2001:db8:e3::1 dev enp3s0 proto kernel metric 256
2001:db8:e3::/64 dev enp3s0 metric 1024
fe80::/64 dev enp3s0 proto kernel metric 256
default via 2001:db8:e3::255 dev enp3s0 metric 1024
$ ip -6 r get 2001:db8:e2::100
2001:db8:e2::100 from :: dev enp2s0 src 2001:db8:e3::1 metric 0
cache
$ ip -6 r get 2001:db8:e2::100 oif enp3s0
2001:db8:e2::100 from :: dev enp2s0 src 2001:db8:e3::1 metric 0
cache
The stack does consider the oif but a mismatch in rt6_device_match is not
considered fatal because RT6_LOOKUP_F_IFACE is not set in the flags.
Cc: Wolfgang Nothdurft <netdev@linux-dude.de>
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Otherwise 4294967295 (MBit/s) (-1) will be printed when there is no link.
Documentation/ABI/testing/sysfs-class-net does not state if this shall be
signed or unsigned.
Also remove the now unused variable fmt_udec.
Signed-off-by: Alexander Stein <alexander.stein@systec-electronic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Several functions can return negative value in case of error,
so their return type should be fixed as well as type of variables
to which this value is assigned.
The problem has been detected using proposed semantic patch
scripts/coccinelle/tests/assign_signed_to_unsigned.cocci [1].
[1]: http://permalink.gmane.org/gmane.linux.kernel/2046107
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Aaron Conole says:
====================
af_unix: return data from multiple SKBs on recv() with MSG_PEEK flag
This patch set implements a bugfix for kernel.org bugzilla #12323, allowing
MSG_PEEK to return all queued data on the unix domain socket, not just the
data contained in a single SKB.
This is the v3 version of this patch, which includes a suggested modification
by Eric Dumazet to convert the unix_sk() conversion macro to a static inline
function. These patches are independent and can be applied separately.
This set was tested over a 24-hour period, utilizing a loop continually
executing the bugzilla issue attached python code. It was instrumented with
a pr_err_once() ([ 13.798683] unix: went there at least one time).
v2->v3:
- Added Eric Dumazet's suggestion for #define to static inline
- Fixed an issue calling unix_state_lock() with an invalid argument
v3->v4:
- Eliminated an XXX comment
- Changed from goto unlock to explicit unix_state_unlock() and break
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
AF_UNIX sockets now return multiple skbs from recv() when MSG_PEEK flag
is set.
This is referenced in kernel bugzilla #12323 @
https://bugzilla.kernel.org/show_bug.cgi?id=12323
As described both in the BZ and lkml thread @
http://lkml.org/lkml/2008/1/8/444 calling recv() with MSG_PEEK on an
AF_UNIX socket only reads a single skb, where the desired effect is
to return as much skb data has been queued, until hitting the recv
buffer size (whichever comes first).
The modified MSG_PEEK path will now move to the next skb in the tree
and jump to the again: label, rather than following the natural loop
structure. This requires duplicating some of the loop head actions.
This was tested using the python socketpair python code attached to
the bugzilla issue.
Signed-off-by: Aaron Conole <aconole@bytheb.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
As suggested by Eric Dumazet this change replaces the
#define with a static inline function to enjoy
complaints by the compiler when misusing the API.
Signed-off-by: Aaron Conole <aconole@bytheb.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of only enabling the backlight (which seems to set it to max
brightness), just re-set the current backlight level, which also takes
care of enabling the backlight if necessary.
Port of radeon commit:
drm/radeon: Restore LCD backlight level on resume (>= R5xx)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Instead of only enabling the backlight (which seems to set it to max
brightness), just re-set the current backlight level, which also takes
care of enabling the backlight if necessary.
Only the radeon_atom_encoder_dpms_dig part tested on a Kaveri laptop,
the radeon_atom_encoder_dpms_avivo part is only compile tested.
Cc: stable@vger.kernel.org
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The "i" variable should be signed or it leads to a crash in the error
handling code.
Fixes: 1d263474c4 ('drm/amdgpu: unwind properly in amdgpu_cs_parser_init()')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
CPU hotplug handling for blk-mq (blk_mq_queue_reinit) acquires
all_q_mutex in blk_mq_queue_reinit_notify() and then removes sysfs
entries by blk_mq_sysfs_unregister(). Removing sysfs entry needs to
be blocked until the active reference of the kernfs_node to be zero.
On the other hand, reading blk_mq_hw_sysfs_cpu sysfs entry (e.g.
/sys/block/nullb0/mq/0/cpu_list) acquires all_q_mutex in
blk_mq_hw_sysfs_cpus_show().
If these happen at the same time, a deadlock can happen. Because one
can wait for the active reference to be zero with holding all_q_mutex,
and the other tries to acquire all_q_mutex with holding the active
reference.
The reason that all_q_mutex is acquired in blk_mq_hw_sysfs_cpus_show()
is to avoid reading an imcomplete hctx->cpumask. Since reading sysfs
entry for blk-mq needs to acquire q->sysfs_lock, we can avoid deadlock
and reading an imcomplete hctx->cpumask by protecting q->sysfs_lock
while hctx->cpumask is being updated.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Cc: Wanpeng Li <wanpeng.li@hotmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Notifier callbacks for CPU_ONLINE action can be run on the other CPU
than the CPU which was just onlined. So it is possible for the
process running on the just onlined CPU to insert request and run
hw queue before establishing new mapping which is done by
blk_mq_queue_reinit_notify().
This can cause a problem when the CPU has just been onlined first time
since the request queue was initialized. At this time ctx->index_hw
for the CPU, which is the index in hctx->ctxs[] for this ctx, is still
zero before blk_mq_queue_reinit_notify() is called by notifier
callbacks for CPU_ONLINE action.
For example, there is a single hw queue (hctx) and two CPU queues
(ctx0 for CPU0, and ctx1 for CPU1). Now CPU1 is just onlined and
a request is inserted into ctx1->rq_list and set bit0 in pending
bitmap as ctx1->index_hw is still zero.
And then while running hw queue, flush_busy_ctxs() finds bit0 is set
in pending bitmap and tries to retrieve requests in
hctx->ctxs[0]->rq_list. But htx->ctxs[0] is a pointer to ctx0, so the
request in ctx1->rq_list is ignored.
Fix it by ensuring that new mapping is established before onlined cpu
starts running.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
CPU hotplug handling for blk-mq (blk_mq_queue_reinit) accesses
q->mq_usage_counter while freezing all request queues in all_q_list.
On the other hand, q->mq_usage_counter is deinitialized in
blk_mq_free_queue() before deleting the queue from all_q_list.
So if CPU hotplug event occurs in the window, percpu_ref_kill() is
called with q->mq_usage_counter which has already been marked dead,
and it triggers warning. Fix it by deleting the queue from all_q_list
earlier than destroying q->mq_usage_counter.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
CPU hotplug handling for blk-mq (blk_mq_queue_reinit) updates
q->mq_map by blk_mq_update_queue_map() for all request queues in
all_q_list. On the other hand, q->mq_map is released before deleting
the queue from all_q_list.
So if CPU hotplug event occurs in the window, invalid memory access
can happen. Fix it by releasing q->mq_map in blk_mq_release() to make
it happen latter than removal from all_q_list.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Suggested-by: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
There is a race between cpu hotplug handling and adding/deleting
gendisk for blk-mq, where both are trying to register and unregister
the same sysfs entries.
null_add_dev
--> blk_mq_init_queue
--> blk_mq_init_allocated_queue
--> add to 'all_q_list' (*)
--> add_disk
--> blk_register_queue
--> blk_mq_register_disk (++)
null_del_dev
--> del_gendisk
--> blk_unregister_queue
--> blk_mq_unregister_disk (--)
--> blk_cleanup_queue
--> blk_mq_free_queue
--> del from 'all_q_list' (*)
blk_mq_queue_reinit
--> blk_mq_sysfs_unregister (-)
--> blk_mq_sysfs_register (+)
While the request queue is added to 'all_q_list' (*),
blk_mq_queue_reinit() can be called for the queue anytime by CPU
hotplug callback. But blk_mq_sysfs_unregister (-) and
blk_mq_sysfs_register (+) in blk_mq_queue_reinit must not be called
before blk_mq_register_disk (++) and after blk_mq_unregister_disk (--)
is finished. Because '/sys/block/*/mq/' is not exists.
There has already been BLK_MQ_F_SYSFS_UP flag in hctx->flags which can
be used to track these sysfs stuff, but it is only fixing this issue
partially.
In order to fix it completely, we just need per-queue flag instead of
per-hctx flag with appropriate locking. So this introduces
q->mq_sysfs_init_done which is properly protected with all_q_mutex.
Also, we need to ensure that blk_mq_map_swqueue() is called with
all_q_mutex is held. Since hctx->nr_ctx is reset temporarily and
updated in blk_mq_map_swqueue(), so we should avoid
blk_mq_register_hctx() seeing the temporary hctx->nr_ctx value
in CPU hotplug handling or adding/deleting gendisk .
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Cc: Ming Lei <tom.leiming@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
