When streaming in Nagle mode, we try to bundle small messages from user
as many as possible if there is one outstanding buffer, i.e. not ACK-ed
by the receiving side, which helps boost up the overall throughput. So,
the algorithm's effectiveness really depends on when Nagle ACK comes or
what the specific network latency (RTT) is, compared to the user's
message sending rate.
In a bad case, the user's sending rate is low or the network latency is
small, there will not be many bundles, so making a Nagle ACK or waiting
for it is not meaningful.
For example: a user sends its messages every 100ms and the RTT is 50ms,
then for each messages, we require one Nagle ACK but then there is only
one user message sent without any bundles.
In a better case, even if we have a few bundles (e.g. the RTT = 300ms),
but now the user sends messages in medium size, then there will not be
any difference at all, that says 3 x 1000-byte data messages if bundled
will still result in 3 bundles with MTU = 1500.
When Nagle is ineffective, the delay in user message sending is clearly
wasted instead of sending directly.
Besides, adding Nagle ACKs will consume some processor load on both the
sending and receiving sides.
This commit adds a test on the effectiveness of the Nagle algorithm for
an individual connection in the network on which it actually runs.
Particularly, upon receipt of a Nagle ACK we will compare the number of
bundles in the backlog queue to the number of user messages which would
be sent directly without Nagle. If the ratio is good (e.g. >= 2), Nagle
mode will be kept for further message sending. Otherwise, we will leave
Nagle and put a 'penalty' on the connection, so it will have to spend
more 'one-way' messages before being able to re-enter Nagle.
In addition, the 'ack-required' bit is only set when really needed that
the number of Nagle ACKs will be reduced during Nagle mode.
Testing with benchmark showed that with the patch, there was not much
difference in throughput for small messages since the tool continuously
sends messages without a break, so Nagle would still take in effect.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit enables dumping the statistics of a broadcast-receiver link
like the traditional 'broadcast-link' one (which is for broadcast-
sender). The link dumping can be triggered via netlink (e.g. the
iproute2/tipc tool) by the link flag - 'TIPC_NLA_LINK_BROADCAST' as the
indicator.
The name of a broadcast-receiver link of a specific peer will be in the
format: 'broadcast-link:<peer-id>'.
For example:
Link <broadcast-link:1001002>
Window:50 packets
RX packets:7841 fragments:2408/440 bundles:0/0
TX packets:0 fragments:0/0 bundles:0/0
RX naks:0 defs:124 dups:0
TX naks:21 acks:0 retrans:0
Congestion link:0 Send queue max:0 avg:0
In addition, the broadcast-receiver link statistics can be reset in the
usual way via netlink by specifying that link name in command.
Note: the 'tipc_link_name_ext()' is removed because the link name can
now be retrieved simply via the 'l->name'.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
In some environment, broadcast traffic is suppressed at high rate (i.e.
a kind of bandwidth limit setting). When it is applied, TIPC broadcast
can still run successfully. However, when it comes to a high load, some
packets will be dropped first and TIPC tries to retransmit them but the
packet retransmission is intentionally broadcast too, so making things
worse and not helpful at all.
This commit enables the broadcast retransmission via unicast which only
retransmits packets to the specific peer that has really reported a gap
i.e. not broadcasting to all nodes in the cluster, so will prevent from
being suppressed, and also reduce some overheads on the other peers due
to duplicates, finally improve the overall TIPC broadcast performance.
Note: the functionality can be turned on/off via the sysctl file:
echo 1 > /proc/sys/net/tipc/bc_retruni
echo 0 > /proc/sys/net/tipc/bc_retruni
Default is '0', i.e. the broadcast retransmission still works as usual.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the previous commit ("tipc: add Gap ACK blocks support for broadcast
link"), we have removed the following link trace events due to the code
changes:
- tipc_link_bc_ack
- tipc_link_retrans
This commit adds them back along with some minor changes to adapt to
the new code.
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
As achieved through commit 9195948fbf ("tipc: improve TIPC throughput
by Gap ACK blocks"), we apply the same mechanism for the broadcast link
as well. The 'Gap ACK blocks' data field in a 'PROTOCOL/STATE_MSG' will
consist of two parts built for both the broadcast and unicast types:
31 16 15 0
+-------------+-------------+-------------+-------------+
| bgack_cnt | ugack_cnt | len |
+-------------+-------------+-------------+-------------+ -
| gap | ack | |
+-------------+-------------+-------------+-------------+ > bc gacks
: : : |
+-------------+-------------+-------------+-------------+ -
| gap | ack | |
+-------------+-------------+-------------+-------------+ > uc gacks
: : : |
+-------------+-------------+-------------+-------------+ -
which is "automatically" backward-compatible.
We also increase the max number of Gap ACK blocks to 128, allowing upto
64 blocks per type (total buffer size = 516 bytes).
Besides, the 'tipc_link_advance_transmq()' function is refactored which
is applicable for both the unicast and broadcast cases now, so some old
functions can be removed and the code is optimized.
With the patch, TIPC broadcast is more robust regardless of packet loss
or disorder, latency, ... in the underlying network. Its performance is
boost up significantly.
For example, experiment with a 5% packet loss rate results:
$ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000
real 0m 42.46s
user 0m 1.16s
sys 0m 17.67s
Without the patch:
$ time tipc-pipe --mc --rdm --data_size 123 --data_num 1500000
real 8m 27.94s
user 0m 0.55s
sys 0m 2.38s
Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: Tuong Lien <tuong.t.lien@dektech.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
In older FW versions the completion flag was treated as the ack flag in
edpm messages. Expose the FW option of setting which mode the QP is in
by adding a flag to the qedr <-> qed API.
Flag is added for backward compatibility with libqedr.
This flag will be set by qedr after determining whether the libqedr is
using the updated version.
Fixes: f109394033 ("qed: Add support for QP verbs")
Signed-off-by: Yuval Basson <yuval.bason@marvell.com>
Signed-off-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I missed the fact that tcp_v4_err() differs from tcp_v6_err().
After commit 4d1a2d9ec1 ("Rename skb to icmp_skb in tcp_v4_err()")
the skb argument has been renamed to icmp_skb only in one function.
I will in a future patch reconciliate these functions to avoid
this kind of confusion.
Fixes: 45af29ca76 ("tcp: allow traceroute -Mtcp for unpriv users")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull i.MX clk driver updates from Shawn Guo:
- A few patches from Abel Vesa as preparation of adding audiomix clock
support
- A couple of cleanups from Anson Huang on clk-sscg-pll and clk-pllv3
driver
- Update imx7ulp clock driver to use imx_clk_hw_cpu() for making the
change of ARM core clock easier
- Drop dependency on ARM64 for i.MX8M clock driver, as there is a move
to support aarch32 mode on aarch64 hardware
- A series from Peng Fan to improve i.MX8M clock drivers, using
composite clock for core and bus clk slice
- Set a better parent clock for flexcan on i.MX6UL to support CiA102
defined bit rates
* tag 'clk-imx-5.8' of git://git.kernel.org/pub/scm/linux/kernel/git/shawnguo/linux:
clk: imx: use imx8m_clk_hw_composite_bus for i.MX8M bus clk slice
clk: imx: add imx8m_clk_hw_composite_bus
clk: imx: add mux ops for i.MX8M composite clk
clk: imx8m: migrate A53 clk root to use composite core
clk: imx8mp: use imx8m_clk_hw_composite_core to simplify code
clk: imx8mp: Define gates for pll1/2 fixed dividers
clk: imx: imx8mp: fix pll mux bit
clk: imx8m: drop clk_hw_set_parent for A53
dt-bindings: clocks: imx8mp: Add ids for audiomix clocks
clk: imx: Add helpers for passing the device as argument
clk: imx: pll14xx: Add the device as argument when registering
clk: imx: gate2: Allow single bit gating clock
clk: imx: clk-pllv3: Use readl_relaxed_poll_timeout() for PLL lock wait
clk: imx: clk-sscg-pll: Remove unnecessary blank lines
clk: imx: drop the dependency on ARM64 for i.MX8M
clk: imx7ulp: make it easy to change ARM core clk
clk: imx: imx6ul: change flexcan clock to support CiA bitrates
Since the switch of floppy driver to blk-mq, the contended (fdc_busy) case
in floppy_queue_rq() is not handled correctly.
In case we reach floppy_queue_rq() with fdc_busy set (i.e. with the floppy
locked due to another request still being in-flight), we put the request
on the list of requests and return BLK_STS_OK to the block core, without
actually scheduling delayed work / doing further processing of the
request. This means that processing of this request is postponed until
another request comes and passess uncontended.
Which in some cases might actually never happen and we keep waiting
indefinitely. The simple testcase is
for i in `seq 1 2000`; do echo -en $i '\r'; blkid --info /dev/fd0 2> /dev/null; done
run in quemu. That reliably causes blkid eventually indefinitely hanging
in __floppy_read_block_0() waiting for completion, as the BIO callback
never happens, and no further IO is ever submitted on the (non-existent)
floppy device. This was observed reliably on qemu-emulated device.
Fix that by not queuing the request in the contended case, and return
BLK_STS_RESOURCE instead, so that blk core handles the request
rescheduling and let it pass properly non-contended later.
Fixes: a9f38e1dec ("floppy: convert to blk-mq")
Cc: stable@vger.kernel.org
Tested-by: Libor Pechacek <lpechacek@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In order to create or activate a new node, lpfc_els_unsol_buffer() invokes
lpfc_nlp_init() or lpfc_enable_node() or lpfc_nlp_get(), all of them will
return a reference of the specified lpfc_nodelist object to "ndlp" with
increased refcnt.
When lpfc_els_unsol_buffer() returns, local variable "ndlp" becomes
invalid, so the refcount should be decreased to keep refcount balanced.
The reference counting issue happens in one exception handling path of
lpfc_els_unsol_buffer(). When "ndlp" in DEV_LOSS, the function forgets to
decrease the refcnt increased by lpfc_nlp_init() or lpfc_enable_node() or
lpfc_nlp_get(), causing a refcnt leak.
Fix this issue by calling lpfc_nlp_put() when "ndlp" in DEV_LOSS.
Link: https://lore.kernel.org/r/1590416184-52592-1-git-send-email-xiyuyang19@fudan.edu.cn
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Xiyu Yang <xiyuyang19@fudan.edu.cn>
Signed-off-by: Xin Tan <tanxin.ctf@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
vhost-scsi pre-allocates the maximum sg entries per command and if a
command requires more than VHOST_SCSI_PREALLOC_SGLS entries, then that
command is failed by it. This patch lets vhost communicate the max sg limit
when it registers vhost_scsi_ops with TCM. With this change, TCM would
report the max sg entries through "Block Limits" VPD page which will be
typically queried by the SCSI initiator during device discovery. By knowing
this limit, the initiator could ensure the maximum transfer length is less
than or equal to what is reported by vhost-scsi.
Link: https://lore.kernel.org/r/1590166317-953-1-git-send-email-sudhakar.panneerselvam@oracle.com
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Mike Christie <mchristi@redhat.com>
Signed-off-by: Sudhakar Panneerselvam <sudhakar.panneerselvam@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
the origin design will use varible of "attr->states" to save node
supported states on current gpu device, but for multi gpu device, when
probe second gpu device, the driver will check attribute node states
from previous gpu device wthether to create attribute node.
it will cause other gpu device create attribute node faild.
1. add member attr_list into amdgpu_device to link supported device attribute node.
2. add new structure "struct amdgpu_device_attr_entry{}" to track device attribute state.
3. drop member "states" from amdgpu_device_attr.
v2:
1. move "attr_list" into amdgpu_pm and rename to "pm_attr_list".
2. refine create & remove device node functions parameter.
fix:
drm/amdgpu: optimize amdgpu device attribute code
Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add support for the hardware ECC BCH engine.
Please mind that this engine has an important limitation:
BCH implementation does not inform the user when an uncorrectable ECC
error occurs. To workaround this, we avoid using the hardware engine
in the read path and do the computation with the software BCH
implementation, which is faster than mixing hardware (for correction)
and software (for verification).
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Link: https://lore.kernel.org/linux-mtd/20200519074549.23673-9-miquel.raynal@bootlin.com
io_close() was punting async manually to skip grabbing files. Use
REQ_F_NO_FILE_TABLE instead, and pass it through the generic path
with -EAGAIN.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_commit_cqring() assembly doesn't look good with extra code handling
drained requests. IOSQE_IO_DRAIN is slow and discouraged to be used in
a hot path, so try to minimise its impact by putting it into a helper
and doing a fast check.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
SQEs are user writable, don't read sqe->off twice in io_timeout_prep()
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move spin_lock_irq() earlier to have only 1 call site of it in
io_timeout(). It makes the flow easier.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In io_uring_cancel_files(), after refcount_sub_and_test() leaves 0
req->refs, it calls io_put_req(), which would also put a ref. Call
io_free_req() instead.
Cc: stable@vger.kernel.org
Fixes: 2ca10259b4 ("io_uring: prune request from overflow list on flush")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
An invariant of cap_bprm_set_creds is that every field in the new cred
structure that cap_bprm_set_creds might set, needs to be set every
time to ensure the fields does not get a stale value.
The field cap_ambient is not set every time cap_bprm_set_creds is
called, which means that if there is a suid or sgid script with an
interpreter that has neither the suid nor the sgid bits set the
interpreter should be able to accept ambient credentials.
Unfortuantely because cap_ambient is not reset to it's original value
the interpreter can not accept ambient credentials.
Given that the ambient capability set is expected to be controlled by
the caller, I don't think this is particularly serious. But it is
definitely worth fixing so the code works correctly.
I have tested to verify my reading of the code is correct and the
interpreter of a sgid can receive ambient capabilities with this
change and cannot receive ambient capabilities without this change.
Cc: stable@vger.kernel.org
Cc: Andy Lutomirski <luto@kernel.org>
Fixes: 58319057b7 ("capabilities: ambient capabilities")
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
- The default root is where we can create v2 cgroups.
- The __DEVEL__sane_behavior mount option has been removed long long ago.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
There will likely be exception handlers that can sleep, which rules
out the usual approach of invoking rcu_nmi_enter() on entry and also
rcu_nmi_exit() on all exit paths. However, the alternative approach of
just not calling anything can prevent RCU from coaxing quiescent states
from nohz_full CPUs that are looping in the kernel: RCU must instead
IPI them explicitly. It would be better to enable the scheduler tick
on such CPUs to interact with RCU in a lighter-weight manner, and this
enabling is one of the things that rcu_nmi_enter() currently does.
What is needed is something that helps RCU coax quiescent states while
not preventing subsequent sleeps. This commit therefore splits out the
nohz_full scheduler-tick enabling from the rest of the rcu_nmi_enter()
logic into a new function named rcu_irq_enter_check_tick().
[ tglx: Renamed the function and made it a nop when context tracking is off ]
[ mingo: Fixed a CONFIG_NO_HZ_FULL assumption, harmonized and fixed all the
comment blocks and cleaned up rcu_nmi_enter()/exit() definitions. ]
Suggested-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200521202116.996113173@linutronix.de
Originally spi_write_then_read() used a fixed statically allocated
buffer which limited the maximum message size it could handle. This
restriction was removed a while ago so that we could dynamically
allocate a buffer if required but the kerneldoc was not updated to
reflect this, do so.
Reported-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20200525133120.57273-1-broonie@kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
Stefano reported a crash with using SQPOLL with io_uring:
BUG: kernel NULL pointer dereference, address: 00000000000003b0
CPU: 2 PID: 1307 Comm: io_uring-sq Not tainted 5.7.0-rc7 #11
RIP: 0010:task_numa_work+0x4f/0x2c0
Call Trace:
task_work_run+0x68/0xa0
io_sq_thread+0x252/0x3d0
kthread+0xf9/0x130
ret_from_fork+0x35/0x40
which is task_numa_work() oopsing on current->mm being NULL.
The task work is queued by task_tick_numa(), which checks if current->mm is
NULL at the time of the call. But this state isn't necessarily persistent,
if the kthread is using use_mm() to temporarily adopt the mm of a task.
Change the task_tick_numa() check to exclude kernel threads in general,
as it doesn't make sense to attempt ot balance for kthreads anyway.
Reported-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/865de121-8190-5d30-ece5-3b097dc74431@kernel.dk
The PXA2xx SPI driver uses devm_spi_register_controller() on bind.
As a consequence, on unbind, __device_release_driver() first invokes
pxa2xx_spi_remove() before unregistering the SPI controller via
devres_release_all().
This order is incorrect: pxa2xx_spi_remove() disables the chip,
rendering the SPI bus inaccessible even though the SPI controller is
still registered. When the SPI controller is subsequently unregistered,
it unbinds all its slave devices. Because their drivers cannot access
the SPI bus, e.g. to quiesce interrupts, the slave devices may be left
in an improper state.
As a rule, devm_spi_register_controller() must not be used if the
->remove() hook performs teardown steps which shall be performed after
unregistering the controller and specifically after unbinding of slaves.
Fix by reverting to the non-devm variant of spi_register_controller().
An alternative approach would be to use device-managed functions for all
steps in pxa2xx_spi_remove(), e.g. by calling devm_add_action_or_reset()
on probe. However that approach would add more LoC to the driver and
it wouldn't lend itself as well to backporting to stable.
The improper use of devm_spi_register_controller() was introduced in 2013
by commit a807fcd090 ("spi: pxa2xx: use devm_spi_register_master()"),
but all earlier versions of the driver going back to 2006 were likewise
broken because they invoked spi_unregister_master() at the end of
pxa2xx_spi_remove(), rather than at the beginning.
Fixes: e0c9905e87 ("[PATCH] SPI: add PXA2xx SSP SPI Driver")
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: stable@vger.kernel.org # v2.6.17+
Cc: Tsuchiya Yuto <kitakar@gmail.com>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=206403#c1
Link: https://lore.kernel.org/r/834c446b1cf3284d2660f1bee1ebe3e737cd02a9.1590408496.git.lukas@wunner.de
Signed-off-by: Mark Brown <broonie@kernel.org>
The Designware SPI driver uses devm_spi_register_controller() on bind.
As a consequence, on unbind, __device_release_driver() first invokes
dw_spi_remove_host() before unregistering the SPI controller via
devres_release_all().
This order is incorrect: dw_spi_remove_host() shuts down the chip,
rendering the SPI bus inaccessible even though the SPI controller is
still registered. When the SPI controller is subsequently unregistered,
it unbinds all its slave devices. Because their drivers cannot access
the SPI bus, e.g. to quiesce interrupts, the slave devices may be left
in an improper state.
As a rule, devm_spi_register_controller() must not be used if the
->remove() hook performs teardown steps which shall be performed after
unregistering the controller and specifically after unbinding of slaves.
Fix by reverting to the non-devm variant of spi_register_controller().
An alternative approach would be to use device-managed functions for all
steps in dw_spi_remove_host(), e.g. by calling devm_add_action_or_reset()
on probe. However that approach would add more LoC to the driver and
it wouldn't lend itself as well to backporting to stable.
Fixes: 04f421e7b0 ("spi: dw: use managed resources")
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: stable@vger.kernel.org # v3.14+
Cc: Baruch Siach <baruch@tkos.co.il>
Link: https://lore.kernel.org/r/3fff8cb8ae44a9893840d0688be15bb88c090a14.1590408496.git.lukas@wunner.de
Signed-off-by: Mark Brown <broonie@kernel.org>
Commit 1ca3dec2b2 ("powerpc/xive: Prevent page fault issues in the
machine crash handler") fixed an issue in the FW assisted dump of
machines using hash MMU and the XIVE interrupt mode under the POWER
hypervisor. It forced the mapping of the ESB page of interrupts being
mapped in the Linux IRQ number space to make sure the 'crash kexec'
sequence worked during such an event. But it didn't handle the
un-mapping.
This mapping is now blocking the removal of a passthrough IO adapter
under the POWER hypervisor because it expects the guest OS to have
cleared all page table entries related to the adapter. If some are
still present, the RTAS call which isolates the PCI slot returns error
9001 "valid outstanding translations".
Remove these mapping in the IRQ data cleanup routine.
Under KVM, this cleanup is not required because the ESB pages for the
adapter interrupts are un-mapped from the guest by the hypervisor in
the KVM XIVE native device. This is now redundant but it's harmless.
Fixes: 1ca3dec2b2 ("powerpc/xive: Prevent page fault issues in the machine crash handler")
Cc: stable@vger.kernel.org # v5.5+
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200429075122.1216388-2-clg@kaod.org
The code patching code wants to get the value of a struct ppc_inst as
a u64 when the instruction is prefixed, so we can pass the u64 down to
__put_user_asm() and write it with a single store.
The optprobes code wants to load a struct ppc_inst as an immediate
into a register so it is useful to have it as a u64 to use the
existing helper function.
Currently this is a bit awkward because the value differs based on the
CPU endianness, so add a helper to do the conversion.
This fixes the usage in arch_prepare_optimized_kprobe() which was
previously incorrect on big endian.
Fixes: 650b55b707 ("powerpc: Add prefixed instructions to instruction data type")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Jordan Niethe <jniethe5@gmail.com>
Link: https://lore.kernel.org/r/20200526072630.2487363-1-mpe@ellerman.id.au
In a few places we want to calculate the address of the next
instruction. Previously that was simple, we just added 4 bytes, or if
using a u32 * we incremented that pointer by 1.
But prefixed instructions make it more complicated, we need to advance
by either 4 or 8 bytes depending on the actual instruction. We also
can't do pointer arithmetic using struct ppc_inst, because it is
always 8 bytes in size on 64-bit, even though we might only need to
advance by 4 bytes.
So add a ppc_inst_next() helper which calculates the location of the
next instruction, if the given instruction was located at the given
address. Note the instruction doesn't need to actually be at the
address in memory.
Although it would seem natural for the value to be passed by value,
that makes it too easy to write a loop that will read off the end of a
page, eg:
for (; src < end; src = ppc_inst_next(src, *src),
dest = ppc_inst_next(dest, *dest))
As noticed by Christophe and Jordan, if end is the exact end of a
page, and the next page is not mapped, this will fault, because *dest
will read 8 bytes, 4 bytes into the next page.
So value is passed by reference, so the helper can be careful to use
ppc_inst_read() on it.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Jordan Niethe <jniethe5@gmail.com>
Link: https://lore.kernel.org/r/20200522133318.1681406-1-mpe@ellerman.id.au
Merge our fixes branch from this cycle. It contains several important
fixes we need in next for testing purposes, and also some that will
conflict with upcoming changes.
Merge Christophe's large series to use huge pages for the linear
mapping on 8xx.
From his cover letter:
The main purpose of this big series is to:
- reorganise huge page handling to avoid using mm_slices.
- use huge pages to map kernel memory on the 8xx.
The 8xx supports 4 page sizes: 4k, 16k, 512k and 8M.
It uses 2 Level page tables, PGD having 1024 entries, each entry
covering 4M address space. Then each page table has 1024 entries.
At the time being, page sizes are managed in PGD entries, implying
the use of mm_slices as it can't mix several pages of the same size
in one page table.
The first purpose of this series is to reorganise things so that
standard page tables can also handle 512k pages. This is done by
adding a new _PAGE_HUGE flag which will be copied into the Level 1
entry in the TLB miss handler. That done, we have 2 types of pages:
- PGD entries to regular page tables handling 4k/16k and 512k pages
- PGD entries to hugepd tables handling 8M pages.
There is no need to mix 8M pages with other sizes, because a 8M page
will use more than what a single PGD covers.
Then comes the second purpose of this series. At the time being, the
8xx has implemented special handling in the TLB miss handlers in order
to transparently map kernel linear address space and the IMMR using
huge pages by building the TLB entries in assembly at the time of the
exception.
As mm_slices is only for user space pages, and also because it would
anyway not be convenient to slice kernel address space, it was not
possible to use huge pages for kernel address space. But after step
one of the series, it is now more flexible to use huge pages.
This series drop all assembly 'just in time' handling of huge pages
and use huge pages in page tables instead.
Once the above is done, then comes icing on the cake:
- Use huge pages for KASAN shadow mapping
- Allow pinned TLBs with strict kernel rwx
- Allow pinned TLBs with debug pagealloc
Then, last but not least, those modifications for the 8xx allows the
following improvement on book3s/32:
- Mapping KASAN shadow with BATs
- Allowing BATs with debug pagealloc
All this allows to considerably simplify TLB miss handlers and associated
initialisation. The overhead of reading page tables is negligible
compared to the reduction of the miss handlers.
While we were at touching pte_update(), some cleanup was done
there too.
Tested widely on 8xx and 832x. Boot tested on QEMU MAC99.
