Conflicts:
drivers/net/vxlan.c
drivers/vhost/net.c
include/linux/if_vlan.h
net/core/dev.c
The net/core/dev.c conflict was the overlap of one commit marking an
existing function static whilst another was adding a new function.
In the include/linux/if_vlan.h case, the type used for a local
variable was changed in 'net', whereas the function got rewritten
to fix a stacked vlan bug in 'net-next'.
In drivers/vhost/net.c, Al Viro's iov_iter conversions in 'net-next'
overlapped with an endainness fix for VHOST 1.0 in 'net'.
In drivers/net/vxlan.c, vxlan_find_vni() added a 'flags' parameter
in 'net-next' whereas in 'net' there was a bug fix to pass in the
correct network namespace pointer in calls to this function.
Signed-off-by: David S. Miller <davem@davemloft.net>
After each call to rdev_dec_pending() we should wakeup the
md thread if the device is found to be faulty.
Otherwise we'll incur heavy delays on failing devices.
Signed-off-by: Neil Brown <nfbrown@suse.de>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Rather than using mddev_lock() to take the reconfig_mutex
when writing to any md sysfs file, we only take mddev_lock()
in the particular _store() functions that require it.
Admittedly this is most, but it isn't all.
This also allows us to remove special-case handling for new_dev_store
(in md_attr_store).
Signed-off-by: NeilBrown <neilb@suse.de>
The one which is not inline (mddev_unlock) gets EXPORTed.
This makes the locking available to personality modules so that it
doesn't have to be imposed upon them.
Signed-off-by: NeilBrown <neilb@suse.de>
There are interdependencies between these two sysfs attributes
and whether a resync is currently running.
Rather than depending on reconfig_mutex to ensure no races when
testing these interdependencies are met, use the spinlock.
This will allow the mutex to be remove from protecting this
code in a subsequent patch.
Signed-off-by: NeilBrown <neilb@suse.de>
There isn't really much room for races with ->safemode_delay.
But as I am trying to clean up any racy code and will soon
be removing reconfig_mutex protection from most _store()
functions:
- only set mddev->safemode_delay once, to ensure no code
can see an intermediate value
- use safemode_timer to call md_safemode_timeout() rather than
calling it directly, to ensure it never races with itself.
Signed-off-by: NeilBrown <neilb@suse.de>
It makes more sense to report bitmap_info->file, rather than
bitmap->file (the later is only available once the array is
active).
With that change, use mddev->lock to protect bitmap_info being
set to NULL, and we can call get_bitmap_file() without taking
the mutex.
Signed-off-by: NeilBrown <neilb@suse.de>
1/ delay setting mddev->bitmap_info.file until 'f' looks
usable, so we don't have to unset it.
2/ Don't allow bitmap file to be set if bitmap_info.file
is already set.
Signed-off-by: NeilBrown <neilb@suse.de>
'buf' is only used because d_path fills from the end of the
buffer instead of from the start.
We don't need a separate buf to handle that, we just need to use
memmove() to move the string to the start.
Signed-off-by: NeilBrown <neilb@suse.de>
No rdev attributes need locking for 'show', though
state_show() might benefit from ensuring it sees a
consistent set of flags.
None even use rdev->mddev, so testing for it isn't really
needed and it certainly doesn't need to be held constant.
So improve state_show() and remove the locking.
Signed-off-by: NeilBrown <neilb@suse.de>
Most attributes can be read safely without any locking.
A race might lead to a slightly out-dated value, but nothing wrong.
We already have locking in some places where needed.
All that remains is can_clear_show(), behind_writes_used_show()
and action_show() which are easily fixed.
Signed-off-by: NeilBrown <neilb@suse.de>
It is important that mddev->private isn't freed while
a sysfs attribute function is accessing it.
So use mddev->lock to protect the setting of ->private to NULL, and
take that lock when checking ->private for NULL and de-referencing it
in the sysfs access functions.
This only applies to the read ('show') side of access. Write
access will be handled separately.
Signed-off-by: NeilBrown <neilb@suse.de>
The only access in md_seq_show that could suffer from races
not protected by ->lock is walking the rdev list.
This can receive sufficient protection from 'rcu'.
So use rdev_for_each_rcu() and get rid of mddev_lock().
Now reading /proc/mdstat will never block in md_seq_show.
Signed-off-by: NeilBrown <neilb@suse.de>
Allow bus clock specification as a common clock handle. This makes this
controller easier to use in a setup based on common clock framework.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Commit 9439eb3ab9 ("asm-generic: io: implement relaxed
accessor macros as conditional wrappers") has added
{read,write}{b,w,l,q}_relaxed to include/asm-generic/io.h.
So COMPILE_TEST can be added.
Signed-off-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
There was quite some confusion why this completion is there and if it is
still necessary. Sadly, it is. However, let's improve the comments and
share what we rediscovered.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
clock-frequency property is meant to control the bus frequency for i2c bus
drivers, but it was incorrectly used to specify i2c controller input clock
frequency.
Introduce new attribute, opencores,ip-clock-frequency, that specifies i2c
controller clock frequency and make clock-frequency attribute compatible
with other i2c drivers. Maintain backwards compatibility in case
opencores,ip-clock-frequency attribute is missing.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Pull networking fixes from David Miller:
1) Stretch ACKs can kill performance with Reno and CUBIC congestion
control, largely due to LRO and GRO. Fix from Neal Cardwell.
2) Fix userland breakage because we accidently emit zero length netlink
messages from the bridging code. From Roopa Prabhu.
3) Carry handling in generic csum_tcpudp_nofold is broken, fix from
Karl Beldan.
4) Remove bogus dev_set_net() calls from CAIF driver, from Nicolas
Dichtel.
5) Make sure PPP deflation never returns a length greater then the
output buffer, otherwise we overflow and trigger skb_over_panic().
Fix from Florian Westphal.
6) COSA driver needs VIRT_TO_BUS Kconfig dependencies, from Arnd
Bergmann.
7) Don't increase route cached MTU on datagram too big ICMPs. From Li
Wei.
8) Fix error path leaks in nf_tables, from Pablo Neira Ayuso.
9) Fix bitmask handling regression in netlink that broke things like
acpi userland tools. From Pablo Neira Ayuso.
10) Wrong header pointer passed to param_type2af() in SCTP code, from
Saran Maruti Ramanara.
11) Stacked vlans not handled correctly by vlan_get_protocol(), from
Toshiaki Makita.
12) Add missing DMA memory barrier to xgene driver, from Iyappan
Subramanian.
13) Fix crash in rate estimators, from Eric Dumazet.
14) We've been adding various workarounds, one after another, for the
change which added the per-net tcp_sock. It was meant to reduce
socket contention but added lots of problems.
Reduce this instead to a proper per-cpu socket and that rids us of
all the daemons.
From Eric Dumazet.
15) Fix memory corruption and OOPS in mlx4 driver, from Jack
Morgenstein.
16) When we disabled UFO in the virtio_net device, it introduces some
serious performance regressions. The orignal problem was IPV6
fragment ID generation, so fix that properly instead. From Vlad
Yasevich.
17) sr9700 driver build breaks on xtensa because it defines macros with
the same name as those used by the arch code. Use more unique
names. From Chen Gang.
18) Fix endianness in new virio 1.0 mode of the vhost net driver, from
Michael S Tsirkin.
19) Several sysctls were setting the maxlen attribute incorrectly, from
Sasha Levin.
20) Don't accept an FQ scheduler quantum of zero, that leads to crashes.
From Kenneth Klette Jonassen.
21) Fix dumping of non-existing actions in the packet scheduler
classifier. From Ignacy Gawędzki.
22) Return the write work_done value when doing TX work in the qlcnic
driver.
23) ip6gre_err accesses the info field with the wrong endianness, from
Sabrina Dubroca.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (54 commits)
sit: fix some __be16/u16 mismatches
ipv6: fix sparse errors in ip6_make_flowlabel()
net: remove some sparse warnings
flow_keys: n_proto type should be __be16
ip6_gre: fix endianness errors in ip6gre_err
qlcnic: Fix NAPI poll routine for Tx completion
amd-xgbe: Set RSS enablement based on hardware features
amd-xgbe: Adjust for zero-based traffic class count
cls_api.c: Fix dumping of non-existing actions' stats.
pkt_sched: fq: avoid hang when quantum 0
net: rds: use correct size for max unacked packets and bytes
vhost/net: fix up num_buffers endian-ness
gianfar: correct the bad expression while writing bit-pattern
net: usb: sr9700: Use 'SR_' prefix for the common register macros
Revert "drivers/net: Disable UFO through virtio"
Revert "drivers/net, ipv6: Select IPv6 fragment idents for virtio UFO packets"
ipv6: Select fragment id during UFO segmentation if not set.
xen-netback: stop the guest rx thread after a fatal error
net/mlx4_core: Fix kernel Oops (mem corruption) when working with more than 80 VFs
isdn: off by one in connect_res()
...
Pull SCSI fixes from James Bottomley:
"This patch set is fixing two serious problems which have turned up
late in the release cycle.
The first fixes a problem with 4k sector disks where the transfer
length (amount of data sent to the disk) was getting increased every
time the disk was revalidated leading to potential for overflows.
The other is a regression oops fix for some of our last merge window
code"
* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
sd: Fix max transfer length for 4k disks
scsi: fix device handler detach oops
Pull drm fixes from Dave Airlie:
"Radeon and amdkfd fixes.
Radeon ones mostly for oops in some test/benchmark functions since
fencing changes, and one regression fix for old GPUs,
There is one cirrus regression fix, the 32bpp broke userspace, so this
hides it behind a module option for the few users who care.
I'm off for a few days, so this is probably the final pull I have, if
I see fixes from Intel I'll forward the pull as I should have email"
* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
drm/cirrus: Limit modes depending on bpp option
drm/radeon: fix the crash in test functions
drm/radeon: fix the crash in benchmark functions
drm/radeon: properly set vm fragment size for TN/RL
drm/radeon: don't init gpuvm if accel is disabled (v3)
drm/radeon: fix PLLs on RS880 and older v2
drm/amdkfd: Don't create BUG due to incorrect user parameter
drm/amdkfd: max num of queues can't be 0
drm/amdkfd: Fix bug in accounting of queues
Pull spi fixes from Mark Brown:
"A couple of driver specific fixes:
- Disable DMA mode for i.MX6DL chips due to a hardware bug.
- Don't use devm_kzalloc() outside of bind/unbind paths in the
fsl-dspi driver, fixing memory leaks"
* tag 'spi-v3.19-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi:
spi: imx: use pio mode for i.mx6dl
spi: spi-fsl-dspi: Remove usage of devm_kzalloc
Pull ACPI power management fix from Rafael Wysocki:
"This is a revert of an ACPI Low-power Subsystem (LPSS) driver change
that was supposed to improve power management of the LPSS DMA
controller, but introduced more serious problems.
Since fixing them turns out to be non-trivial, it is better to revert
the commit in question at this point and try to fix the original issue
differently in the next cycle"
* tag 'pm+acpi-3.19-fin' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
Revert "ACPI / LPSS: introduce a 'proxy' device to power on LPSS for DMA"
A shost_printk() statement was left over from debugging.
It can safely be removed; the same information is displayed
in the debugging message some lines further down.
Cc: Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
return type of wait_for_completion_timeout is unsigned long not int, this
patch uses the return value of wait_for_completion_timeout in the condition
directly rather than adding a additional appropriately typed variable.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
return type of wait_for_completion_timeout is unsigned long not int, this
patch uses the return value of wait_for_completion_timeout in the condition
directly rather than assigning it to an incorrect type variable.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch adds new regulator driver to support max77843
MFD(Multi Function Device) chip`s regulators.
The Max77843 has two voltage regulators for USB safeout.
Signed-off-by: Jaewon Kim <jaewon02.kim@samsung.com>
Signed-off-by: Beomho Seo <beomho.seo@samsung.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Make use of a new interface provided by iov_iter, backed by
scatter-gather list of iovec, instead of the old interface based on
sg_iovec. Also use iov_iter_advance() instead of manual iteration.
This commit should contain only literal replacements, without
functional changes.
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Doug Gilbert <dgilbert@interlog.com>
Cc: "James E.J. Bottomley" <JBottomley@parallels.com>
Signed-off-by: Kent Overstreet <kmo@daterainc.com>
[dpark: add more description in commit message]
Signed-off-by: Dongsu Park <dongsu.park@profitbricks.com>
[hch: fixed to do a deep clone of the iov_iter, and to properly use
the iov_iter direction]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
If Xenstore sends back a XS_ERROR for TRANSACTION_END, the driver BUGs
because it cannot find the matching transaction in the list. For
TRANSACTION_START, it leaks memory.
Check the message as returned from xenbus_dev_request_and_reply(), and
clean up for TRANSACTION_START or discard the error for
TRANSACTION_END.
Signed-off-by: Jennifer Herbert <Jennifer.Herbert@citrix.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
The bug fixes around GPE races have been done to the EC driver by the
previous commits. This patch increases the revision to 3 to indicate the
behavior differences between the old and the new drivers. The
copyright/authorship notices are also updated.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Timeout in the ec_poll() doesn't refer to the last register access time. It
thus can win the competition against the acpi_ec_gpe_handler() if a
transaction takes longer than 1ms but individual register accesses are less
than 1ms. In some cases, it can make the following silicon bug easier to
be triggered:
GPE EN is not wired to the GPE trigger line, so when GPE STS is already
set when 1 is written to GPE EN, no GPE can be triggered.
This patch adds register access timestamp reference support for ec_poll()
to reduce the number of ec_poll() invocations.
Reported-by: Venkat Raghavulu <venkat.raghavulu@intel.com>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch switches EC driver into ACPI_GPE_DISPATCH_RAW_HANDLER mode where
the GPE lock is not held for acpi_ec_gpe_handler() and the ACPICA internal
GPE enabling/disabling/clearing operations are bypassed so that further
improvements are possible with the GPE APIs.
There are 2 strong reasons for deploying raw GPE handler mode in the EC
driver:
1. Some hardware logics can control their interrupts via their own
registers, so their interrupts can be disabled/enabled/acknowledged
without using the super IRQ controller provided functions. While there
is no mean (EC commands) for the EC driver to achieve this.
2. During suspending, the EC driver is still working for a while to
complete the platform firmware provided functionailities using ec_poll()
after all GPEs are disabled (see acpi_ec_block_transactions()), which
means the EC driver will drive the EC GPE out of the GPE core's control.
Without deploying the raw GPE handler mode, we can see many races between
the EC driver and the GPE core due to the above restrictions:
1. There is a race condition due to ACPICA internal GPE
disabling/clearing/enabling logics in acpi_ev_gpe_dispatch():
Orignally EC GPE is disabled (EN=0), cleared (STS=0) before invoking a
GPE handler and re-enabled (EN=1) after invoking a GPE handler in
acpi_ev_gpe_dispatch(). When re-enabling appears, GPE may be flagged
(STS=1).
=================================================================
(event pending A)
=================================================================
acpi_ev_gpe_dispatch() ec_poll()
EN=0
STS=0
acpi_ec_gpe_handler()
*****************************************************************
(event handling A)
Lock(EC)
advance_transaction()
EC_SC read
=================================================================
(event pending B)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
*****************************************************************
(event handling B)
Lock(EC)
advance_transaction()
EC_SC read
=================================================================
(event pending C)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
EN=1
This race condition is the root cause of different issues on different
silicon variations.
A. Silicon variation A:
On some platforms, GPE will be triggered due to "writing 1 to EN when
STS=1". This is because both EN and STS lines are wired to the GPE
trigger line.
1. Issue 1:
We can see no-op acpi_ec_gpe_handler() invoked on such platforms.
This is because:
a. event pending B: An event can arrive after ACPICA's GPE
clearing performed in acpi_ev_gpe_dispatch(), this event may
fail to be detected by EC_SC read that is performed before its
arrival;
b. event handling B: The event can be handled in ec_poll() because
EC lock is released after acpi_ec_gpe_handler() invocation;
c. There is no code in ec_poll() to clear STS but the GPE can
still be triggered by the EN=1 write performed in
acpi_ev_finish_gpe(), this leads to a no-op EC GPE handler
invocation;
d. As no-op GPE handler invocations are counted by the EC driver
to trigger the command storming conditions, the wrong no-op
GPE handler invocations thus can easily trigger wrong command
storming conditions.
Note 1:
If we removed GPE disabling/enabling code from
acpi_ev_gpe_dispatch(), we could still see no-op GPE handlers
triggered by the event arriving after the GPE clearing and before
the GPE handling on both silicon variation A and B. This can only
occur if the CPU is very slow (timing slice between STS=0 write
and EC_SC read should be short enough before hardware sets another
GPE indication). Thus this is very rare and is not what we need to
fix.
B. Silicon variation B:
On other platforms, GPE may not be triggered due to "writing 1 to EN
when STS=1". This is because only STS line is wired to the GPE
trigger line.
2. Issue 2:
We can see GPE loss on such platforms. This is because:
a. event pending B vs. event handling A: An event can arrive after
ACPICA's GPE handling performed in acpi_ev_gpe_dispatch(), or
event pending C vs. event handling B: An event can arrive after
Linux's GPE handling performed in ec_poll(),
these events may fail to be detected by EC_SC read that is
performed before their arrival;
b. The GPE cannot be triggered by EN=1 write performed in
acpi_ev_finish_gpe();
c. If no polling mechanism is implemented in the driver for the
pending event (for example, SCI_EVT), this event is lost due to
no GPE being triggered.
Note 2:
On most platforms, there might be another rule that GPE may not be
triggered due to "writing 1 to STS when STS=1 and EN=1".
Then on silicon variation B, an even worse case is if the issue 2
event loss happens, further events may never trigger GPE again on
such platforms due to being blocked by the current STS=1. Unless
someone clears STS, all events have to be polled.
2. There is a race condition due to lacking in GPE status checking in EC
driver:
Originally, GPE status is checked in ACPICA core but not checked in
the GPE handler. Thus since the status checking and handling is not
locked, it can be interrupted by another handling path.
=================================================================
(event pending A)
=================================================================
acpi_ev_gpe_detect() ec_poll()
if (EN==1 && STS==1)
*****************************************************************
(event handling A)
Lock(EC)
advance_transaction()
EC_SC read
EC_SC handled
Unlock(EC)
*****************************************************************
acpi_ev_gpe_dispatch()
EN=0
STS=0
acpi_ec_gpe_handler()
*****************************************************************
(event handling B)
Lock(EC)
advance_transaction()
EC_SC read
Unlock(EC)
*****************************************************************
3. Issue 3:
We can see no-op acpi_ec_gpe_handler() invoked on both silicon
variation A and B. This is because:
a. event pending A: An event can arrive to trigger an EC GPE and
ACPICA checks it and is about to invoke the EC GPE handler;
b. event handling A: The event can be handled in ec_poll() because
EC lock is not held after the GPE status checking;
c. event handling B: Then when the EC GPE handler is invoked, it
becomes a no-op GPE handler invocation.
d. As no-op GPE handler invocations are counted by the EC driver
to trigger the command storming conditions, the wrong no-op
GPE handler invocations thus can easily trigger wrong command
storming conditions.
Note 3:
This no-op GPE handler invocation is rare because the time between
the IRQ arrival and the acpi_ec_gpe_handler() invocation is less than
the timeout value waited in ec_poll(). So most of the no-op GPE
handler invocations are caused by the reason described in issue 1.
3. There is a race condition due to ACPICA internal GPE clearing logic in
acpi_enable_gpe():
During runtime, acpi_enable_gpe() can be invoked by the EC storming
prevention code. When it is invoked, GPE may be flagged (STS=1).
=================================================================
(event pending A)
=================================================================
acpi_ev_gpe_dispatch() acpi_ec_transaction()
EN=0
STS=0
acpi_ec_gpe_handler()
*****************************************************************
(event handling A)
Lock(EC)
advance_transaction()
EC_SC read
EC_SC handled
Unlock(EC)
*****************************************************************
EN=1 ?
Lock(EC)
Unlock(EC)
=================================================================
(event pending B)
=================================================================
acpi_enable_gpe()
STS=0
EN=1
4. Issue 4:
We can see GPE loss on both silicon variation A and B platforms.
This is because:
a. event pending B: An event can arrive right before ACPICA's GPE
clearing performed in acpi_enable_gpe();
b. If the GPE is cleared when GPE is disabled, then EN=1 write in
acpi_enable_gpe() cannot trigger this GPE;
c. If no polling mechanism is implemented in the driver for this
event (for example, SCI_EVT), this event is lost due to no GPE
being triggered.
Note 4:
Currently we don't have this issue, but after we switch the EC
driver into ACPI_GPE_DISPATCH_RAW_HANDLER mode, we need to take care
of handling this because the EN=1 write in acpi_ev_gpe_dispatch()
will be abandoned.
There might be more race issues for the current GPE handler usages. This is
because the EC IRQ's enabling/disabling/checking/clearing/handling
operations should be locked by a single lock that is under the EC driver's
control to achieve the serialization. Which means we need to invoke GPE
APIs with EC driver's lock held and all ACPICA internal GPE operations
related to the GPE handler should be abandoned. Invoking GPE APIs inside of
the EC driver lock and bypassing ACPICA internal GPE operations requires
the ACPI_GPE_DISPATCH_RAW_HANDLER mode where the same lock used by the APIs
are released prior than invoking the handlers. Otherwise, we can see dead
locks due to circular locking dependencies (see Reference below).
This patch then switches the EC driver into the
ACPI_GPE_DISPATCH_RAW_HANDLER mode so that it can perform correct GPE
operations using the GPE APIs:
1. Bypasses EN modifications performed in acpi_ev_gpe_dispatch() by
using acpi_install_gpe_raw_handler() and invoking all GPE APIs with EC
spin lock held. This can fix issue 1 as it makes a non frequent GPE
enabling/disabling environment.
2. Bypasses STS clearing performed in acpi_enable_gpe() by replacing
acpi_enable_gpe()/acpi_disable_gpe() with acpi_set_gpe(). This can fix
issue 4. And this can also help to fix issue 1 as it makes a no sudden
GPE clearing environment when GPE is frequently enabled/disabled.
3. Ensures STS acknowledged before handling by invoking acpi_clear_gpe()
in advance_transaction(). This can finally fix issue 1 even in a
frequent GPE enabling/disabling environment. And this can also finally
fix issue 3 when issue 2 is fixed.
Note 3:
GPE clearing is edge triggered W1C, which means we can clear it right
before handling it. Since all EC GPE indications are handled in
advance_transaction() by previous commits, we can now move GPE clearing
into it to implement the correct GPE clearing.
Note 4:
We can use acpi_set_gpe() which is not shared GPE safer instead of
acpi_enable_gpe()/acpi_disable_gpe() because EC GPE is not shared by
other hardware, which is mentioned in the ACPI specification 5.0, 12.6
Interrupt Model: "OSPM driver treats this as an edge event (the EC SCI
cannot be shared)". So we can stop using shared GPE safer APIs
acpi_enable_gpe()/acpi_disable_gpe() in the EC driver. Otherwise
cleanups need to be made in acpi_ev_enable_gpe() to bypass the GPE
clearing logic before keeping acpi_enable_gpe().
This patch also invokes advance_transaction() when GPE is re-enabled in the
task context which:
1. Ensures EN=1 can trigger GPE by checking and handling EC status register
right after EN=1 writes. This can fix issue 2.
After applying this patch, without frequent GPE enablings considered:
=================================================================
(event pending A)
=================================================================
acpi_ec_gpe_handler() ec_poll()
*****************************************************************
(event handling A)
Lock(EC)
advance_transaction()
if STS==1
STS=0
EC_SC read
=================================================================
(event pending B)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
*****************************************************************
(event handling B)
Lock(EC)
advance_transaction()
if STS==1
STS=0
EC_SC read
=================================================================
(event pending C)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
The event pending for issue 1 (event pending B) can arrive as a next GPE
due to the previous IRQ context STS=0 write. And if it is handled by
ec_poll() (event handling B), as it is also acknowledged by ec_poll(), the
event pending for issue 2 (event pending C) can properly arrive as a next
GPE after the task context STS=0 write. So no GPE will be lost and never
triggered due to GPE clearing performed in the wrong position. And since
all GPE handling is performed after a locked GPE status checking, we can
hardly see no-op GPE handler invocations due to issue 1 and 3. We may still
see no-op GPE handler invocations due to "Note 1", but as it is inevitable,
it needn't be fixed.
After applying this patch, with frequent GPE enablings considered:
=================================================================
(event pending A)
=================================================================
acpi_ec_gpe_handler() acpi_ec_transaction()
*****************************************************************
(event handling A)
Lock(EC)
advance_transaction()
if STS==1
STS=0
EC_SC read
=================================================================
(event pending B)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
*****************************************************************
(event handling B)
Lock(EC)
EN=1
if STS==1
advance_transaction()
if STS==1
STS=0
EC_SC read
=================================================================
(event pending C)
=================================================================
EC_SC handled
Unlock(EC)
*****************************************************************
The event pending for issue 2 can be manually handled by
advance_transaction(). And after the STS=0 write performed in the manual
triggered advance_transaction(), GPE can always arrive. So no GPE will be
lost due to frequent GPE disabling/enabling performed in the driver like
issue 4.
Note 5:
It's ideally when EN=1 write occurred, an IRQ thread should be woken up to
handle the GPE when the GPE was raised. But this requires the IRQ thread to
contain the poller code for all EC GPE indications, while currently some of
the indications are handled in the user tasks. It then is very hard for the
code to determine whether a user task should be invoked or the poller work
item should be scheduled. So we have to invoke advance_transaction()
directly now and it leaves us such a restriction for the GPE re-enabling:
it must be performed in the task context to avoid starving the GPEs.
As a conclusion: we can see the EC GPE is always handled in serial after
deploying the raw GPE handler mode:
Lock(EC)
if (STS==1)
STS=0
EC_SC read
EC_SC handled
Unlock(EC)
The EC driver specific lock is responsible to make the EC GPE handling
processes serialized so that EC can handle its GPE from both IRQ and task
contexts and the next IRQ can be ensured to arrive after this process.
Note 6:
We have many EC_FLAGS_MSI qurik users in the current driver. They all seem
to be suffering from unexpected GPE triggering source lost. And they are
false root caused to a timing issue. Since EC communication protocol has
already flow control defined, timing shouldn't be the root cause, while
this fix might be fixing the root cause of the old bugs.
Link: https://lkml.org/lkml/2014/11/4/974
Link: https://lkml.org/lkml/2014/11/18/316
Link: https://www.spinics.net/lists/linux-acpi/msg54340.html
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit da9a83e1a845f2d7332bdbc0632466b2595e5424
For acpi_set_gpe()/acpi_enable_gpe(), our target is to purify them to be APIs
that can be used for various GPE handling models, so we need them to be
pure GPE enabling APIs. GPE enabling/disabling has 2 use cases:
1. Driver may permanently enable/disable GPEs according to the usage
counts.
1. When upper layers (the users of the driver) submit requests to the
driver, it means they care about the underlying hardware. GPE need
to be enabled for the first request submission and disabled for the
last request completion.
2. When the GPE is shared between 2+ silicon logics. GPE need to be
enabled for either silicon logic's driver and disabled when all of
the drivers are not started.
For these cases, acpi_enable_gpe()/acpi_disable_gpe() should be used. When
the usage count is increased from 0 to 1, the GPE is enabled and it is
disabled when the usage count is decrased from 1 to 0.
2. Driver may temporarily disables the GPE to enter an GPE polling mode and
wants to re-enable it later.
1. Prevent GPE storming: when a driver cannot fully solve the condition
that triggered the GPE in the GPE context, in order not to trigger
GPE storm, driver has to disable GPE to switch into the polling mode
and re-enables it in the non interrupt context after the storming
condition is cleared.
2. Meet throughput requirement: some IO drivers need to poll hardware
again and again until nothing indicated instead of just handling once
for one interruption, this need to be done in the polling mode or the
IO flood may prevent the GPE handler from returning.
3. Meet realtime requirement: in order not to block CPU to handle higher
realtime prioritized GPEs, lower priority GPEs can be handled in the
polling mode.
For these cases, acpi_set_gpe() should be used to switch to/from the
polling mode.
This patch adds unconditional GPE enabling support into acpi_set_gpe() so
that this API can be used by the drivers to switch back from the GPE
polling mode unconditionally.
Originally this function includes GPE clearing logic in it.
First, the GPE clearing is typically used in the GPE handling code to:
1. Acknowledge the GPE when we know there is an edge triggered GPE raised
and is about to handle it, otherwise the unexpected clearing may lead to
a GPE loss;
2. Issue actions after we have handled a level triggered GPE, otherwise
the unexpected clearing may trigger unwanted OSPM actions to the
hardware (for example, clocking in out-dated write FIFO data).
Thus the GPE clearing is not suitable to be used in the GPE enabling APIs.
Second, the combination of acknowledging and enabling may also not be
expected by the hardware drivers. For GPE clearing, we have a seperate API
acpi_clear_gpe(). There are cases drivers do want the 2 operations to be
split. So splitting these 2 operations could facilitates drivers the
maximum possibilities to achieve success. For a combined one, we already
have acpi_finish_gpe() ready to be invoked.
Given the fact that drivers should complete all outstanding requests before
putting themselves into the sleep states, as this API is executed for
outstanding requests, it should also have nothing to do with the
"RUN"/"WAKE" distinguishing. That's why the acpi_set_gpe(ACPI_GPE_ENABLE)
should not be implemented by acpi_hw_low_set_gpe(ACPI_GPE_CONDITIONAL_ENABLE).
This patch thus converts acpi_set_gpe(ACPI_GPE_ENABLE) into
acpi_hw_low_set_gpe(ACPI_GPE_ENABLE) to achieve a seperate GPE enabling API.
Drivers then are encouraged to use this API when they need to switch
to/from the GPE polling mode.
Note that the acpi_set_gpe()/acpi_finish_gpe() should be first introduced to
Linux using a divergence reduction patch before sending a linuxized version
of this patch. Lv Zheng.
Link: https://github.com/acpica/acpica/commit/da9a83e1
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: David E. Box <david.e.box@linux.intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This can help to reduce source code differences between Linux and ACPICA
upstream. Further driver cleanups also require these APIs to eliminate GPE
storms.
1. acpi_set_gpe(): An API that driver should invoke in the case it wants
to disable/enable IRQ without honoring the reference
count implemented in the acpi_disable_gpe() and
acpi_enable_gpe(). Note that this API should only be
invoked inside a acpi_enable_gpe()/acpi_disable_gpe()
pair.
2. acpi_finish_gpe(): Drivers returned ACPI_REENABLE_GPE unset from the
GPE handler should use this API instead of
invoking acpi_set_gpe()/acpi_enable_gpe() to
re-enable the GPE.
The GPE APIs can be invoked inside of a GPE handler or in the task context
with a driver provided lock held. This driver provided lock is safe to be
held in the GPE handler by the driver.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit 199cad16530a45aea2bec98e528866e20c5927e1
Since whether the GPE should be disabled/enabled/cleared should only be
determined by the GPE driver's state machine:
1. GPE should be disabled if the driver wants to switch to the GPE polling
mode when a GPE storm condition is indicated and should be enabled if
the driver wants to switch back to the GPE interrupt mode when all of
the storm conditions are cleared. The conditions should be protected by
the driver's specific lock.
2. GPE should be enabled if the driver has accepted more than one request
and should be disabled if the driver has completed all of the requests.
The request count should be protected by the driver's specific lock.
3. GPE should be cleared either when the driver is about to handle an edge
triggered GPE or when the driver has completed to handle a level
triggered GPE. The handling code should be protected by the driver's
specific lock.
Thus the GPE enabling/disabling/clearing operations are likely to be
performed with the driver's specific lock held while we currently cannot do
this. This is because:
1. We have the acpi_gbl_gpe_lock held before invoking the GPE driver's
handler. Driver's specific lock is likely to be held inside of the
handler, thus we can see some dead lock issues due to the reversed
locking order or recursive locking. In order to solve such dead lock
issues, we need to unlock the acpi_gbl_gpe_lock before invoking the
handler. BZ 1100.
2. Since GPE disabling/enabling/clearing should be determined by the GPE
driver's state machine, we shouldn't perform such operations inside of
ACPICA for a GPE handler to mess up the driver's state machine. BZ 1101.
Originally this patch includes a logic to flush GPE handlers, it is dropped
due to the following reasons:
1. This is a different issue;
2. Linux OSL has fixed this by flushing SCI in acpi_os_wait_events_complete().
We will pick up this topic when the Linux OSL fix turns out to be not
sufficient.
Note that currently the internal operations and the acpi_gbl_gpe_lock are
also used by ACPI_GPE_DISPATCH_METHOD and ACPI_GPE_DISPATCH_NOTIFY. In
order not to introduce regressions, we add one
ACPI_GPE_DISPATCH_RAW_HANDLER type to be distiguished from
ACPI_GPE_DISPATCH_HANDLER. For which the acpi_gbl_gpe_lock is unlocked before
invoking the GPE handler and the internal enabling/disabling operations are
bypassed to allow drivers to perform them at a proper position using the
GPE APIs and ACPI_GPE_DISPATCH_RAW_HANDLER users should invoke acpi_set_gpe()
instead of acpi_enable_gpe()/acpi_disable_gpe() to bypass the internal GPE
clearing code in acpi_enable_gpe(). Lv Zheng.
Known issues:
1. Edge-triggered GPE lost for frequent enablings
On some buggy silicon platforms, GPE enable line may not be directly
wired to the GPE trigger line. In that case, when GPE enabling is
frequently performed for edge-triggered GPEs, GPE status may stay set
without being triggered.
This patch may maginify this problem as it allows GPE enabling to be
parallel performed during the process the GPEs are handled.
This is an existing issue, because:
1. For task context:
Current ACPI_GPE_DISPATCH_METHOD practices have proven that this
isn't a real issue - we can re-enable edge-triggered GPE in a work
queue where the GPE status bit might already be set.
2. For IRQ context:
This can even happen when the GPE enabling occurs before returning
from the GPE handler and after unlocking the GPE lock.
Thus currently no code is included to protect this.
Link: https://github.com/acpica/acpica/commit/199cad16
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit 7926d5ca9452c87f866938dcea8f12e1efb58f89
There is an issue in acpi_install_gpe_handler() and acpi_remove_gpe_handler().
The code to obtain the GPE dispatcher type from the Handler->original_flags
is wrong:
if (((Handler->original_flags & ACPI_GPE_DISPATCH_METHOD) ||
(Handler->original_flags & ACPI_GPE_DISPATCH_NOTIFY)) &&
ACPI_GPE_DISPATCH_NOTIFY is 0x03 and ACPI_GPE_DISPATCH_METHOD is 0x02, thus
this statement is TRUE for the following dispatcher types:
0x01 (ACPI_GPE_DISPATCH_HANDLER): not expected
0x02 (ACPI_GPE_DISPATCH_METHOD): expected
0x03 (ACPI_GPE_DISPATCH_NOTIFY): expected
There is no functional issue due to this because Handler->original_flags is
only set in acpi_install_gpe_handler(), and an earlier checker has excluded
the ACPI_GPE_DISPATCH_HANDLER:
if ((gpe_event_info->Flags & ACPI_GPE_DISPATCH_MASK) ==
ACPI_GPE_DISPATCH_HANDLER)
{
Status = AE_ALREADY_EXISTS;
goto free_and_exit;
}
...
Handler->original_flags = (u8) (gpe_event_info->Flags &
(ACPI_GPE_XRUPT_TYPE_MASK | ACPI_GPE_DISPATCH_MASK));
We need to clean this up before modifying the GPE dispatcher type values.
In order to prevent such issue from happening in the future, this patch
introduces ACPI_GPE_DISPATCH_TYPE() macro to be used to obtain the GPE
dispatcher types. Lv Zheng.
Link: https://github.com/acpica/acpica/commit/7926d5ca
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: David E. Box <david.e.box@linux.intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit 04f25acdd4f655ae33f83de789bb5f4b7790171c
This patch follows acpi_ev_fixed_event_detect(), which invokes
acpi_gbl_global_event_handler instead of invoking it in
acpi_ev_fixed_event_dispatch(), moves acpi_gbl_global_event_handler from
acpi_ev_gpe_dispatch() to acpi_ev_gpe_detect(). This makes further cleanups
around acpi_ev_gpe_dispatch() simpler. Lv Zheng.
Link: https://github.com/acpica/acpica/commit/04f25acd
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: David E. Box <david.e.box@linux.intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit 8823b44ff53859ab24ecfcfd3fba8cc56b17d223
Currently we rely on the logic that GPE blocks will never be deleted,
otherwise we can be broken by the race between acpi_ev_create_gpe_block(),
acpi_ev_delete_gpe_block() and acpi_ev_gpe_detect().
On the other hand, if we want to protect GPE block creation/deletion, we
need to use a different synchronization facility to protect the period
between acpi_ev_gpe_dispatch() and acpi_ev_asynch_enable_gpe(). Which leaves us
no choice but abandoning the ACPI_MTX_EVENTS used during this period.
This patch removes ACPI_MTX_EVENTS used during this period and the
acpi_ev_valid_gpe_event() to reflect current restriction. Lv Zheng.
Link: https://github.com/acpica/acpica/commit/8823b44f
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: David E. Box <david.e.box@linux.intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPICA commit ca10324788bc9bdaf47fa9e51145129c1299144d
This patch deletes a sanity check from acpi_ev_enable_gpe().
This kind of check is already done in
acpi_enable_gpe()/acpi_remove_gpe_handler()/acpi_update_all_gpes() before invoking
acpi_ev_enable_gpe():
1. acpi_enable_gpe(): same check (skip if DISPATCH_NONE) is now implemented.
2. acpi_remove_gpe_handler(): a more strict check (skip if !DISPATCH_HANDLER)
is implemented.
3. acpi_update_all_gpes(): a more strict check (skip if DISPATCH_NONE ||
DISPATCH_HANDLER || CAN_WAKE)
4. acpi_set_gpe(): since it is invoked by the OSPM driver where the GPE
handler is known to be available, such check isn't needed.
So we can simply remove this duplicated check from acpi_ev_enable_gpe().
Lv Zheng.
Link: https://github.com/acpica/acpica/commit/ca103247
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: David E. Box <david.e.box@linux.intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This is a back port result of the Linux commit:
Commit c50f13c672
Subject: ACPICA: Save current masks of enabled GPEs after enable register writes
Besides of the indent divergences, only a missing prototype added due to
the ACPICA internal coding style.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It's possible for invalidate_range_start mmu notifier callback to race
against userptr object release. If the gem object was released prior to
obtaining the spinlock in invalidate_range_start we're hitting null
pointer dereference.
Testcase: igt/gem_userptr_blits/stress-mm-invalidate-close
Testcase: igt/gem_userptr_blits/stress-mm-invalidate-close-overlap
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@vger.kernel.org
[Jani: added code comment suggested by Chris]
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
