During vMotion and Clone VMware by default runs multiple sequential 4MB
XCOPY requests same time. If CTL issues reads sequentially in 1MB chunks
for each XCOPY command, reads from different commands are not detected
as sequential by serseq option code and allowed to execute simultaneously.
Such read pattern confused ZFS prefetcher, causing suboptimal disk access.
Issuing all reads same time make serseq code work properly, serializing
reads both within each XCOPY command and between them.
My tests with ZFS pool of 14 disks in RAID10 shows prefetcher efficiency
improved from 37% to 99.7%, copying speed improved by 10-60%, average
read latency reduced twice on HDD layer and by five times on zvol layer.
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
Clang emits SSE instructions on amd64 in the common path of
pthread_mutex_unlock. If the thread does not otherwise use SSE,
this usage incurs a context-switch of the FPU/SSE state, which
reduces the performance of multiple real-world applications by a
non-trivial amount (3-5% in one application).
Instead of this change, I experimented with eagerly switching the
FPU state at context-switch time. This did not help. Most of the
cost seems to be in the read/write of memory--as kib@ stated--and
not in the #NM handling. I tested on machines with and without
XSAVEOPT.
One counter-argument to this change is that most applications already
use SIMD, and the number of applications and amount of SIMD usage
are only increasing. This is absolutely true. I agree that--in
general and in principle--this change is in the wrong direction.
However, there are applications that do not use enough SSE to offset
the extra context-switch cost. SSE does not provide a clear benefit
in the current libthr code with the current compiler, but it does
provide a clear loss in some cases. Therefore, disabling SSE in
libthr is a non-loss for most, and a gain for some.
I refrained from disabling SSE in libc--as was suggested--because
I can't make the above argument for libc. It provides a wide variety
of code; each case should be analyzed separately.
https://lists.freebsd.org/pipermail/freebsd-current/2015-March/055193.html
Suggestions from: dim, jmg, rpaulo
Approved by: kib (mentor)
MFC after: 2 weeks
Sponsored by: Dell Inc.
CloudABI uses a structure called cloudabi_sockstat_t. Think of it as
'struct stat' for sockets. It is used by functions such as
getsockname(), getpeername(), some of the getsockopt() values, etc.
This change implements the sock_stat_get() system call that returns a
copy of this structure. The accept() system call should also return a
full copy of this structure eventually, but for now we're only
interested in the peer address. Add a TODO() to make sure this is
patched up later on.
Differential Revision: https://reviews.freebsd.org/D3218
On CloudABI we want to create file descriptors with just the minimal set
of Capsicum rights in place. The reason for this is that it makes it
easier to obtain uniform behaviour across different operating systems.
By explicitly whitelisting the operations, we can return consistent
error codes, but also prevent applications from depending OS-specific
behaviour.
Extend kern_kqueue() to take an additional struct filecaps that is
passed on to falloc_caps(). Update the existing consumers to pass in
NULL.
Differential Revision: https://reviews.freebsd.org/D3259
It looks like EVFILT_READ and EVFILT_WRITE trigger under the same
conditions as poll()'s POLLRDNORM and POLLWRNORM as described by POSIX.
The only difference is that POLLRDNORM has to be triggered on regular
files unconditionally, whereas EVFILT_READ only triggers when not EOF.
Introduce a new flag, NOTE_FILE_POLL, that can be used to make
EVFILT_READ and EVFILT_WRITE behave identically to poll(). This flag
will be used by cloudlibc's poll() function.
Reviewed by: jmg
Differential Revision: https://reviews.freebsd.org/D3303
This will help ensure we don't trash file descriptors that get used later on
in the daemon
Found via internal Coverity scan
MFC after: 2 weeks
Discussed with: cem, ed, markj
Differential Revision: https://reviews.freebsd.org/D3081
Submitted by: Miles Ohlrich <miles.ohlrich@isilon.com>
Sponsored by: EMC / Isilon Storage Division
vm_offset_t pmap_quick_enter_page(vm_page_t m)
void pmap_quick_remove_page(vm_offset_t kva)
These will create and destroy a temporary, CPU-local KVA mapping of a specified page.
Guarantees:
--Will not sleep and will not fail.
--Safe to call under a non-sleepable lock or from an ithread
Restrictions:
--Not guaranteed to be safe to call from an interrupt filter or under a spin mutex on all platforms
--Current implementation does not guarantee more than one page of mapping space across all platforms. MI code should not make nested calls to pmap_quick_enter_page.
--MI code should not perform locking while holding onto a mapping created by pmap_quick_enter_page
The idea is to use this in busdma, for bounce buffer copies as well as virtually-indexed cache maintenance on mips and arm.
NOTE: the non-i386, non-amd64 implementations of these functions still need review and testing.
Reviewed by: kib
Approved by: kib (mentor)
Differential Revision: http://reviews.freebsd.org/D3013
This config is already building all modules, so we don't need the
MODULES_EXTRA definition. It was also causing problems to users who
rely on MODULES_OVERRIDE to do the right thing.
Discussed with: ian
defines the keys differently than NIST does, so we have to muck with
key lengths and nonce/IVs to be standard compliant...
Remove the iv from secasvar as it was unused...
Add a counter protected by a mutex to ensure that the counter for GCM
and ICM will never be repeated.. This is a requirement for security..
I would use atomics, but we don't have a 64bit one on all platforms..
Fix a bug where IPsec was depending upon the OCF to ensure that the
blocksize was always at least 4 bytes to maintain alignment... Move
this logic into IPsec so changes to OCF won't break IPsec...
In one place, espx was always non-NULL, so don't test that it's
non-NULL before doing work..
minor style cleanups...
drop setting key and klen as they were not used...
Enforce that OCF won't pass invalid key lengths to AES that would
panic the machine...
This was has been tested by others too... I tested this against
NetBSD 6.1.5 using mini-test suite in
https://github.com/jmgurney/ipseccfgs and the only things that don't
pass are keyed md5 and sha1, and 3des-deriv (setkey syntax error),
all other modes listed in setkey's man page... The nice thing is
that NetBSD uses setkey, so same config files were used on both...
Reviewed by: gnn
overflows the stack during root mount in some configurations.
Tested by: Fabian Keil <freebsd-listen@fabiankeil.de> (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
of the timehands, from the kern_tc.c implementation to vdso. Add
comments giving hints where to look for the algorithm explanation.
To compensate the removal of rmb() in userspace binuptime(), add
explicit lfence instruction before rdtsc. On i386, add usual
complications to detect SSE2 presence; assume that old CPUs which do
not implement SSE2 also execute rdtsc almost in order.
Reviewed by: alc, bde (previous version)
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
It looks like umtx_key_get() has the addition and subtraction the wrong
way around, meaning that it fails to match in certain cases. This causes
the cloudlibc unit tests to deadlock in certain cases.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D3287
The filter is called from the network hot path and must not sleep.
The filter runs with the descriptor lock held and does not manipulates the
buffers, so it is not necessary sleep when the hold buffer is in use.
Just ignore the hold buffer contents when it is being copied to user space
(when hold buffer in use is set).
This fix the "Sleeping thread owns a non-sleepable lock" panic when the
userland thread is too busy reading the packets from bpf(4).
PR: 200323
MFC after: 2 weeks
Sponsored by: Rubicon Communications (Netgate)
a pcb from stoppcbs[] rather than the thread's PCB. However, exited threads
retained td_oncpu from the last time they ran, and newborn threads had their
CPU fields cleared to zero during fork and thread creation since they are
in the set of fields zeroed when threads are setup. To fix, explicitly
update the CPU fields for exiting threads in sched_throw() to reflect the
switch out and reset the CPU fields for new threads in sched_fork_thread()
to NOCPU.
Reviewed by: kib
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D3193
which constitute the majority of the pages that are processed by
vm_fault_dontneed(), are already near the tail of the inactive queue. Only
the pages at faulting virtual addresses are actually moved by
vm_page_advise(..., MADV_DONTNEED). However, vm_page_advise(...,
MADV_DONTNEED) is simultaneously too aggressive and passive for the moved
pages. It makes most of these pages too easily reclaimable, and at the same
time it leaves enough pages in the active queue to trigger pageouts by the
page daemon. Instead, with this change, the pages at faulting virtual
addresses are moved to the tail of the inactive queue, where they are
relatively close to the pages prefetched by the same page fault.
Discussed with: jeff
Sponsored by: EMC / Isilon Storage Division
as it is now, devfs ignores MNT_FORCE anyway, so it needs to be modified
to trigger the panic) with consumers still opened.
Note that this still results in a leak of r/w/e counters. It seems
to be harmless, though. If anyone knows a better way to approach
this - please tell.
Discussed with: kib@, mav@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3050
For example, without this patch, the following three lines
in /boot/loader.conf would result in /boot/root.img being preloaded
twice, and two md(4) devices - md0 and md1 - being created.
initmd_load="YES"
initmd_type="md_image"
initmd_name="/boot/root.img"
Reviewed by: marcel@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3204
When doing a data abort from userland it is possible to get
more than one data abort inside the same exception level.
Add an appropriate exception number to allow nesting of
data_abort handler for EL0.
Reviewed by: andrew
Obtained from: Semihalf
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3276
