| Age | Commit message (Collapse) | Author |
|---|
|
Now that RPC replies are processed in a workqueue, there's no need
to disable IRQs when managing send and receive buffers. This saves
noticeable overhead per RPC.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
Clean up: The reply tasklet is no longer used.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
The reply tasklet is fast, but it's single threaded. After reply
traffic saturates a single CPU, there's no more reply processing
capacity.
Replace the tasklet with a workqueue to spread reply handling across
all CPUs. This also moves RPC/RDMA reply handling out of the soft
IRQ context and into a context that allows sleeps.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
The rb_send_bufs and rb_recv_bufs arrays are used to implement a
pair of stacks for keeping track of free rpcrdma_req and rpcrdma_rep
structs. Replace those arrays with free lists.
To allow more than 512 RPCs in-flight at once, each of these arrays
would be larger than a page (assuming 8-byte addresses and 4KB
pages). Allowing up to 64K in-flight RPCs (as TCP now does), each
buffer array would have to be 128 pages. That's an order-6
allocation. (Not that we're going there.)
A list is easier to expand dynamically. Instead of allocating a
larger array of pointers and copying the existing pointers to the
new array, simply append more buffers to each list.
This also makes it simpler to manage receive buffers that might
catch backwards-direction calls, or to post receive buffers in
bulk to amortize the overhead of ib_post_recv.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
Clean up: The error cases in rpcrdma_reply_handler() almost never
execute. Ensure the compiler places them out of the hot path.
No behavior change expected.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
Commit 8301a2c047cc ("xprtrdma: Limit work done by completion
handler") was supposed to prevent xprtrdma's upcall handlers from
starving other softIRQ work by letting them return to the provider
before all CQEs have been polled.
The logic assumes the provider will call the upcall handler again
immediately if the CQ is re-armed while there are still queued CQEs.
This assumption is invalid. The IBTA spec says that after a CQ is
armed, the hardware must interrupt only when a new CQE is inserted.
xprtrdma can't rely on the provider calling again, even though some
providers do.
Therefore, leaving CQEs on queue makes sense only when there is
another mechanism that ensures all remaining CQEs are consumed in a
timely fashion. xprtrdma does not have such a mechanism. If a CQE
remains queued, the transport can wait forever to send the next RPC.
Finally, move the wcs array back onto the stack to ensure that the
poll array is always local to the CPU where the completion upcall is
running.
Fixes: 8301a2c047cc ("xprtrdma: Limit work done by completion ...")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
ib_req_notify_cq(IB_CQ_REPORT_MISSED_EVENTS) returns a positive
value if WCs were added to a CQ after the last completion upcall
but before the CQ has been re-armed.
Commit 7f23f6f6e388 ("xprtrmda: Reduce lock contention in
completion handlers") assumed that when ib_req_notify_cq() returned
a positive RC, the CQ had also been successfully re-armed, making
it safe to return control to the provider without losing any
completion signals. That is an invalid assumption.
Change both completion handlers to continue polling while
ib_req_notify_cq() returns a positive value.
Fixes: 7f23f6f6e388 ("xprtrmda: Reduce lock contention in ...")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Devesh Sharma <devesh.sharma@avagotech.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
After adding a swapfile on an NFS/RDMA mount and removing the
normal swap partition, I was able to push the NFS client well
into swap without any issue.
I forgot to swapoff the NFS file before rebooting. This pinned
the NFS mount and the IB core and provider, causing shutdown to
hang. I think this is expected and safe behavior. Probably
shutdown scripts should "swapoff -a" before unmounting any
filesystems.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Tested-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
Unsignaled send WRs can get flushed as part of normal unmount, so don't
log them as warnings.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
|
|
There's no need to record the time tracepoints take when tracing is off.
This is because:
1) We cannot see these records since ring_buffer record is off at that
moment.
2) If tracing is off and benchmark tracepoint is enabled, the time
tracepoint takes is fewer than the same situation when tracing is on,
since the tracepoints need to be wrote into ring_buffer, it would
take more time. If turn on tracing at this moment, the average and
standard deviation cannot exactly present the time that tracepoints
take to write data into ring_buffer.
Link: http://lkml.kernel.org/r/1445947933-27955-1-git-send-email-zhang.chunyan@linaro.org
Signed-off-by: Chunyan Zhang <zhang.chunyan@linaro.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
|
|
Although, the default value of rel_type_nop is zero, and the value
of R_386_NONE/R_X86_64_NONE is zero too, but it should be assigned
a meaningful value explicitly, otherwise it looks confused.
Assign R_386_NONE to rel_type_nop for 386, assign R_X86_64_NONE
to rel_type_nop for x86_64.
Link: http://lkml.kernel.org/r/1446020606-16352-1-git-send-email-huawei.libin@huawei.com
Signed-off-by: Li Bin <huawei.libin@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
|
|
set_event_pid
For the case where pids are already in set_event_pid, and one is added or
removed then each CPU should be checked to make sure that the new or old pid
is on or not on a CPU.
For example:
# echo 123 >> set_event_pid
or
# echo '!123' >> set_event_pid
Link: http://lkml.kernel.org/r/20151030061643.GA19480@cac
Suggested-by: Jiaxing Wang <hello.wjx@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
|
|
On some NUMA system, after dom0 up, we see below warning even if there are
enough pfn ranges that could be used for remapping:
"Unable to find available pfn range, not remapping identity pages"
Fix it to avoid getting a memory region of zero size in xen_find_pfn_range.
Signed-off-by: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
|
|
Add the "sigma" vendor prefix for Sigma Designs, Inc.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Signed-off-by: Rob Herring <robh@kernel.org>
|
|
During the migration to HDA core code, we lost the workaround for 4k
BDL boundary. The flag exists in the new hdac_bus, but it's never
set. This resulted in the sudden sound stall on some controllers that
require this workaround like Creative Recon3D.
This patch fixes the issue by setting the flag for such controllers
properly.
Fixes: ccc98865aa44 ('ALSA: hda - Migrate more hdac_stream codes')
Cc: <stable@vger.kernel.org> # v4.2+
Signed-off-by: Takashi Iwai <tiwai@suse.de>
|
|
Intuos Pen in wireless mode does not have the same report id (2) as
when it is in USB mode (17).
This patch also moves WIRELESS next to REMOTE in type enum so we
can group devices with similar features easily.
Reported-by: Dale Brewe <dlbrewe@hotmail.com>
Tested-by: Dale Brewe <dlbrewe@hotmail.com>
Signed-off-by: Ping Cheng <pingc@wacom.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
|
|
This reverts commit d871cd2ec5abf8715774bcb90aa32ae5b750b587 as it causes
regression in BBB
Reported-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
This reverts commit e3faf2b8826b8ac58cdaad7f801e59e389320f0e as it causes
regression in BBB
Reported-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
|
|
Unused amdgpu_mn functions threw warnings for every file that includes
amdgpu.h. It makes sense to inline this amdgpu_mn stubs to avoid the warning.
Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Cc: stable@vger.kernel.org
|
|
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Diamond core 212 is a generic purpose core without full MMU used for
sample noMMU configuration.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Not having HAVE_FUTEX_CMPXCHG makes futex_detect_cmpxchg probe
cmpxchg_futex_value_locked with NULL address. It's not guaranteed to
fault without MMU, instead it locks up on Xtensa when there's no RAM at
address 0.
Select HAVE_FUTEX_CMPXCHG in noMMU Xtensa configurations.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
RAM starts at 0x60000000 on noMMU cores, not at 0x40000000. Fix the
default.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
KIO region location is different for noMMU cores. Provide different
default physical address and make KIO virtual address equal to physical.
Move xtensa_get_kio_paddr function close to XCHAL_KIO_PADDR definition
and define it not only for MMUv3, but for all MMU options except MMUv2.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
There's no kernel/user separation in noMMU and PS.RING may not exist.
Even if it exists it should not be used because TLB entries are not set
up for user ring on user pages.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
This fixes the following warning when default memory region crosses
0x80000000:
arch/xtensa/include/asm/processor.h:40:47: warning:
integer overflow in expression [-Woverflow]
#define TASK_SIZE (PLATFORM_DEFAULT_MEM_START + PLATFORM_DEFAULT_MEM_SIZE)
^
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
- make cache-related assembly macros empty if core doesn't have
corresponding cache type;
- don't initialize cache attributes in instruction/data TLB entries if
there's no corresponding cache type.
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Build-time fixes:
- make lbeg/lend/lcount save/restore conditional on kernel entry;
- don't clear lcount in platform_restart functions unconditionally.
Run-time fixes:
- use correct end of range register in __endla paired with __loopt, not
the unused temporary register. This fixes .bss zero-initialization.
Update comments in asmmacro.h;
- don't clobber a10 in the usercopy that leads to access to unmapped
memory.
Cc: <stable@vger.kernel.org>
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
|
|
Now that we strictly forbid absolute relocations in libstub code,
make sure that we don't emit any when CONFIG_MODVERSIONS is enabled,
by stripping the kcrctab sections from the object file. This fixes
a build problem under CONFIG_MODVERSIONS=y.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc into HEAD
There's nothing much in the way of new features this time; it's mostly
bug fixes, plus Nikunj has implemented support for KVM_CAP_NR_MEMSLOTS.
|
|
'x86/amd' into next
Conflicts:
drivers/iommu/amd_iommu_types.h
|
|
The function returns 0 on success, so check for the right
value.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
|
|
The hash_accept call fails to work on sockets that have not received
any data. For some algorithm implementations it may cause crashes.
This patch fixes this by ensuring that we only export and import on
sockets that have received data.
Cc: stable@vger.kernel.org
Reported-by: Harsh Jain <harshjain.prof@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Stephan Mueller <smueller@chronox.de>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into HEAD
KVM: s390: Bugfix and cleanups
There is one important bug fix for a potential memory corruption
and/or guest errors for guests with 63 or 64 vCPUs. This fix would
qualify for 4.3 but is some days too late giving that we are
about to release 4.3.
Given that this patch is cc stable >= 3.15 anyway, we can handle
it via 4.4. merge window.
This pull request also contains two cleanups.
|
|
Reported-by: Keith Webb <khwebb@gmail.com>
Suggested-by: Keith Webb <khwebb@gmail.com>
Cc: stable@vger.kernel.org
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=106671
Reviewed-by: Clint Taylor <Clinton.A.Taylor@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1446209424-28801-1-git-send-email-jani.nikula@intel.com
|
|
Now that we use memremap instead of ioremap, Use WRITE_ONCE / READ_ONCE
instead of iowrite / ioread.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
|
|
Do this until we know how much MOB memory to allocate for these surfaces.
v2: Mask also non-DX multisample.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
|
|
Per commit 2e586a7e017a "drm/vmwgfx: Map the fifo as cached" the driver
expects the fifo registers to be cacheable. In preparation for
deprecating ioremap_cache() convert its usage in vmwgfx to memremap().
Cc: David Airlie <airlied@linux.ie>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Sinclair Yeh <syeh@vmware.com>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
|
|
|
|
I'm getting unclaimed register writes when checking the WM registers
after the crtc is disabled. So I would imagine those are guarded by
the crtc power well. Fix this by not reading out wm state when the
power well is off.
Cc: stable@vger.kernel.org # v4.3
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92181
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
|
|
ESS Allegro (marked ES1989S), [125d:1988] (rev 10) is mute after
loading snd-maestro3 and running alsactl restore. Touching master
volume or mute makes it work.
Looks like a bug in the AC'97 codec integrated into the chip.
Write AC97_MASTER register twice to work-around the problem.
Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/scottwood/linux into next
Freescale updates from Scott:
"Highlights include 64-bit book3e kexec/kdump support, a rework of the
qoriq clock driver, device tree changes including qoriq fman nodes,
support for a new 85xx board, and some fixes.
Note that there is a trivial merge conflict with the clock tree's next
branch, in the clock Makefile."
|
|
It turns out that at least some versions of glibc end up reading
/proc/meminfo at every single startup, because glibc wants to know the
amount of memory the machine has. And while that's arguably insane,
it's just how things are.
And it turns out that it's not all that expensive most of the time, but
the vmalloc information statistics (amount of virtual memory used in the
vmalloc space, and the biggest remaining chunk) can be rather expensive
to compute.
The 'get_vmalloc_info()' function actually showed up on my profiles as
4% of the CPU usage of "make test" in the git source repository, because
the git tests are lots of very short-lived shell-scripts etc.
It turns out that apparently this same silly vmalloc info gathering
shows up on the facebook servers too, according to Dave Jones. So it's
not just "make test" for git.
We had two patches to just cache the information (one by me, one by
Ingo) to mitigate this issue, but the whole vmalloc information of of
rather dubious value to begin with, and people who *actually* want to
know what the situation is wrt the vmalloc area should just look at the
much more complete /proc/vmallocinfo instead.
In fact, according to my testing - and perhaps more importantly,
according to that big search engine in the sky: Google - there is
nothing out there that actually cares about those two expensive fields:
VmallocUsed and VmallocChunk.
So let's try to just remove them entirely. Actually, this just removes
the computation and reports the numbers as zero for now, just to try to
be minimally intrusive.
If this breaks anything, we'll obviously have to re-introduce the code
to compute this all and add the caching patches on top. But if given
the option, I'd really prefer to just remove this bad idea entirely
rather than add even more code to work around our historical mistake
that likely nobody really cares about.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Merge file descriptor allocation speedup.
Eric Dumazet has a test-case for a fairly common network deamon load
pattern: openign and closing a lot of sockets that each have very little
work done on them. It turns out that in that case, the cost of just
finding the correct file descriptor number can be a dominating factor.
We've long had a trivial optimization for allocating file descriptors
sequentially, but that optimization ends up being not very effective
when other file descriptors are being closed concurrently, and the fd
patterns are not some simple FIFO pattern. In such cases we ended up
spending a lot of time just scanning the bitmap of open file descriptors
in order to find the next file descriptor number to open.
This trivial patch-series mitigates that by simply introducing a
second-level bitmap of which words in the first bitmap are already fully
allocated. That cuts down the cost of scanning by an order of magnitude
in some pathological (but realistic) cases.
The second patch is an even more trivial patch to avoid unnecessarily
dirtying the cacheline for the close-on-exec bit array that normally
ends up being all empty.
* fs-file-descriptor-optimization:
vfs: conditionally clear close-on-exec flag
vfs: Fix pathological performance case for __alloc_fd()
|
|
|
|
* pm-avs:
PM / AVS: rockchip-io: Fix module autoload for OF platform driver
* pm-clk:
PM / clk: Fix broken build due to non-matching code and header #ifdefs
* powercap:
powercap / RAPL: Enable Broxton RAPL support
|
|
* pm-opp:
PM / OPP: passing NULL to PTR_ERR()
PM / OPP: Move cpu specific code to opp/cpu.c
PM / OPP: Move opp core to its own directory
PM / OPP: Prefix exported opp routines with dev_pm_opp_
PM / OPP: Rename opp init/free table routines
PM / OPP: reuse of_parse_phandle()
|
|
* pm-domains:
PM / Domains: Merge measurements for PM QoS device latencies
PM / Domains: Don't measure ->start|stop() latency in system PM callbacks
PM / Domains: Rename *pm_genpd_poweron|poweroff()
PM / Domains: Remove pm_genpd_poweron() API
PM / Domains: Remove pm_genpd_poweroff_unused() API
soc: dove: Let genpd deal with disabling of unused PM domains
PM / Domains: Remove in_progress counter from struct generic_pm_domain
PM / domains: Drop unused label
PM / Domains: Remove cpuidle attach
PM / Domains: Remove name based API for genpd
|
|
* pm-cpufreq:
cpufreq: postfix policy directory with the first CPU in related_cpus
cpufreq: create cpu/cpufreq/policyX directories
cpufreq: remove cpufreq_sysfs_{create|remove}_file()
cpufreq: create cpu/cpufreq at boot time
cpufreq: Use cpumask_copy instead of cpumask_or to copy a mask
cpufreq: ondemand: Drop unnecessary locks from update_sampling_rate()
cpufreq: intel_pstate: Fix intel_pstate powersave min_perf_pct value
cpufreq: intel_pstate: Avoid calculation for max/min
Documentation: kernel_parameters for Intel P state driver
cpufreq: intel_pstate: Use ACPI perf configuration
cpufreq: intel-pstate: Use separate max pstate for scaling
cpufreq: intel_pstate: get P1 from TAR when available
cpufreq: Drop redundant check for inactive policies
cpufreq : powernv: Report Pmax throttling if capped below nominal frequency
cpufreq: imx: update the clock switch flow to support imx6ul
cpufreq: tegra20: remove superfluous CONFIG_PM ifdefs
cpufreq: conservative: remove 'enable' field
cpufreq: integrator: Fix module autoload for OF platform driver
* pm-cpuidle:
cpuidle: mvebu: disable the bind/unbind attributes and use builtin_platform_driver
cpuidle: mvebu: clean up multiple platform drivers
|
