The following changes prevent the unrecoverable freezes and rcu_sched
stall warnings experienced in each of my attempts to take advantage of
lima.
Replace the COMPOSITE_NOGATE definition of aclk_gpu_pre with a
COMPOSITE that retains the selection of HDMIPHY as the PLL source, but
instead makes uses of the aclk_gpu PLL source gate and parent names
defined by mux_pll_src_4plls_p rather than mux_aclk_gpu_pre_p.
Remove the now unused mux_aclk_gpu_pre_p and the four named but also
unused definitions (cpll_gpu, gpll_gpu, hdmiphy_gpu and usb480m_gpu)
of the aclk_gpu PLL source gate.
Use the correct gate offset for aclk_gpu and aclk_gpu_noc.
Fixes: 307a2e9ac5 ("clk: rockchip: add clock controller for rk3228")
Cc: stable@vger.kernel.org
Signed-off-by: Justin Swartz <justin.swartz@risingedge.co.za>
[double-checked against SoC manual and added fixes tag]
Link: https://lore.kernel.org/r/20200114162503.7548-1-justin.swartz@risingedge.co.za
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
rbd_notify_op_lock() isn't interested in a notify reply. Instead of
accepting that page vector just to free it, have watch-notify code take
care of it.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
rbd_dev->opts is used to distinguish between the image that is being
mapped and a parent. However, because we no longer establish watch for
read-only mappings, this test is imprecise and results in unnecessary
rbd_unregister_watch() calls.
Make it consistent with need_watch in rbd_dev_image_probe().
Fixes: b9ef2b8858 ("rbd: don't establish watch for read-only mappings")
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
rbd_dev_unprobe() is supposed to undo most of rbd_dev_image_probe(),
including rbd_dev_header_info(), which means that rbd_dev_header_info()
isn't supposed to be called after rbd_dev_unprobe().
However, rbd_dev_image_release() calls rbd_dev_unprobe() before
rbd_unregister_watch(). This is racy because a header update notify
can sneak in:
"rbd unmap" thread ceph-watch-notify worker
rbd_dev_image_release()
rbd_dev_unprobe()
free and zero out header
rbd_watch_cb()
rbd_dev_refresh()
rbd_dev_header_info()
read in header
The same goes for "rbd map" because rbd_dev_image_probe() calls
rbd_dev_unprobe() on errors. In both cases this results in a memory
leak.
Fixes: fd22aef8b4 ("rbd: move rbd_unregister_watch() call into rbd_dev_image_release()")
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
rbd_unregister_watch() flushes notifies and therefore cannot be called
under header_rwsem because a header update notify takes header_rwsem to
synchronize with "rbd map". If mapping an image fails after the watch
is established and a header update notify sneaks in, we deadlock when
erroring out from rbd_dev_image_probe().
Move watch registration and unregistration out of the critical section.
The only reason they were put there was to make header_rwsem management
slightly more obvious.
Fixes: 811c668877 ("rbd: fix rbd map vs notify races")
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
wfx_tx_flush() wait there are no more frame in device buffer. However,
this event may never happens since wfx_tx_flush() don't forbid to
enqueue new frames.
Note that wfx_tx_flush() should only ensure that all frames currently in
hardware queues are sent. So the current code is more restrictive that
it should.
Note that wfx_tx_flush() release the lock before to return while
wfx_tx_lock_flush() keep the lock.
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-31-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Until now, wfx_flush() flushed queue for while device instead of only
the queue of the intended vif. It sometime failed with a timeout, but
this error was not reported.
Moreover, if the device was frozen, wfx_flush didn't do anything and it
results a potential warning (and maybe a resource leak) when the frozen
device was unregistered.
We can also notice that wfx_tx_queues_wait_empty_vif() did only exist to
work around the broken feature of wfx_flush().
This patch repair wfx_flush() and therefore drop
wfx_tx_queues_wait_empty_vif().
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-29-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, wfx_pending_remove() (from queue.c) call wfx_skb_dtor()
(from data_tx.c) that forward the tx status to mac80211.
Moreover, there no purpose to retrieve a frame from the pending queue
without dequeuing it. So, the main purpose of wfx_pending_remove() is to
forward the tx status to mac80211.
Let's make the architecture cleaner:
- merge wfx_pending_remove() into wfx_pending_get()
- call wfx_skb_dtor() from data_tx.c
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-27-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current code computes itself the QoS policy to choose which frame
should be sent. However, firmware already do that job. Firmware would
prefer to have packets in every queues and be able to choose itself
which queue to use.
So, this patch sort the queues from the emptiest to the fulliest (thanks
to the pending frames counter introduced a few commits earlier). It send
frame to the least full queue.
However, we continue to be careful with frames that have to be sent
after a dtim ("cab": Content After (DTIM) Beacon).
So, this patch splits AC queues in two skb_queues: one for normal frames
and another for cab frames. It cares to send frames from CAB skb_queue
if appropriate.
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-23-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, the driver tracks power save state of the stations with the
variable sta_asleep_mask. Then, it takes care to not sent data to asleep
stations.
However, this work is already done by mac80211. Normally, there are no
frames for asleep stations in our queues. So, driver do not have to
filter frames in its queues (apart the frames marked "AFTER_DTIM").
Notice that there is a risk of race between state of the station and
data send to the firmware. However, this risk is limited since the
number of frame in queues are small. In add, this race also exists with
the current code. Anyway, the firmware is able to detect the problem and
driver will receive a 'REQUEUE' status (translated in
TX_STAT_TX_FILTERED for mac80211).
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-13-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of maintaining stop status for each queue, we can just maintain
a global status for all queues.
In add, wfx_tx_queues_{lock/unlock} are only used when no more
tx_policies are available. Therefore, the counter of recursive locks
("tx_locked_cnt") is useless.
So, wfx_tx_queues_{lock/unlock} can be replaced by
ieee80211_{stop/start}_queues.
Reviewed-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jérôme Pouiller <jerome.pouiller@silabs.com>
Link: https://lore.kernel.org/r/20200401110405.80282-4-Jerome.Pouiller@silabs.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
