The 'physio' fast-path used by AIO requests on md(4) devices, is not
gated on the unsafe_aio knob. Prior to r327755, some AIO requests could
fail the fast-path and fall back to the slow-path (requests for devices
not supporting unmapped I/O and requests which failed with EFAULT during
the fast-path). However, those cases now return a suitable error rather
than using the slow-path.
PR: 217261
Reviewed by: asomers
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D14742
ECN (ABE)" proposal to the New Reno congestion control algorithm module.
ABE reduces the amount of congestion window reduction in response to
ECN-signalled congestion relative to the loss-inferred congestion response.
More details about ABE can be found in the Internet-Draft:
https://tools.ietf.org/html/draft-ietf-tcpm-alternativebackoff-ecn
The implementation introduces four new sysctls:
- net.inet.tcp.cc.abe defaults to 0 (disabled) and can be set to non-zero to
enable ABE for ECN-enabled TCP connections.
- net.inet.tcp.cc.newreno.beta and net.inet.tcp.cc.newreno.beta_ecn set the
multiplicative window decrease factor, specified as a percentage, applied to
the congestion window in response to a loss-based or ECN-based congestion
signal respectively. They default to the values specified in the draft i.e.
beta=50 and beta_ecn=80.
- net.inet.tcp.cc.abe_frlossreduce defaults to 0 (disabled) and can be set to
non-zero to enable the use of standard beta (50% by default) when repairing
loss during an ECN-signalled congestion recovery episode. It enables a more
conservative congestion response and is provided for the purposes of
experimentation as a result of some discussion at IETF 100 in Singapore.
The values of beta and beta_ecn can also be set per-connection by way of the
TCP_CCALGOOPT TCP-level socket option and the new CC_NEWRENO_BETA or
CC_NEWRENO_BETA_ECN CC algo sub-options.
Submitted by: Tom Jones <tj@enoti.me>
Tested by: Tom Jones <tj@enoti.me>, Grenville Armitage <garmitage@swin.edu.au>
Relnotes: Yes
Differential Revision: https://reviews.freebsd.org/D11616
The former is fairly vague; these are FDT overlays to be applied to the
running system, so /boot/dtb is a sensible location to put it without
cluttering up /boot/dtb even further if desired.
The console may have been set for different colors before lualoader kicks
in; notably, a black-on-white color scheme is not necessarily what we're
expecting.
While here, make color.default() a composition of color.escape() instead of
rewriting the escape sequence to make it more obvious what it's achieving: a
white-on-black color scheme with no attributes set.
Reported by: emaste, whose eyes may rest easily
These problems have only been observed with boards using U-Boot (e.g. ARM)
where virtual addresses are already set in the memory map by the firmware
and the firmware is expecting a call to SetVirtualAddressMap to be made.
I refrain from mentioning this in the note because this could also be the
case on some not-yet-tested firmware on amd64 and it's not a bad
recommendation for the general case.
There's a fair amount of duplication between MD linuxulator files.
Make indentation and comments consistent between the three versions of
linux_sysvec.c to reduce diffs when comparing them.
Sponsored by: Turing Robotic Industries Inc.
about the chip including the erase block size at attach time.
Also add myself to the copyrights since at this point svn blame would point
to me as the culprit for much of this.
all over the place. Also pass the softc as the arg to all the internal
functions instead of passing a device_t and calling device_get_softc() in
each function.
before starting them.
Using the wait-before logic would make sense if there was useful time-
consuming work that could be done between the end of one write and the
beginning of the next, but it also requires doing the wait-for-ready before
reading, because a prior write or erase could still be in progress. Reading
is the far more common case, so adding a whole extra bus transaction to
check for ready before each read would soak up any small gains that might be
had from doing async writes.
Such pages are re-enqueued at the end of the fault handler, preserving
LRU. Rather than performing two separate operations per fault, simply
requeue the page at the end of the fault (or bump its activation count
if it resides in PQ_ACTIVE, avoiding the page queue lock entirely).
This elides some page lock and page queue lock operations in common
cases, e.g., CoW faults.
Note that we must still dequeue the source page for "optimized" CoW
faults since the page may not remain enqueued while it is moved to
another object.
Reviewed by: alc, kib
Tested by: pho
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D14625
In many cases the page is not enqueued so the change will have no
effect. However, the change is needed to support an optimization in
the fault handler and in some cases (sendfile, the buffer cache) it
was being emulated by the caller anyway.
Reviewed by: alc
Tested by: pho
MFC after: 2 weeks
X-Differential Revision: https://reviews.freebsd.org/D14625
The new page does not belong to a VM object, but the page daemon does
not expect to encounter such pages.
Reviewed by: alc, kib
Tested by: pho
MFC after: 1 week
X-Differential Revision: https://reviews.freebsd.org/D14625
for the same condition that the preceeding lines checked for and would have
returned EIO, so the assert could never possibly trigger (sc_sectorsize must
inherently be an integer multiple of FLASH_PAGE_SIZE).
transfers data in both directions at once. When writing to the device,
use a dummy buffer for the incoming data, not the same buffer as the
outgoing data. Writes are done in FLASH_PAGE_SIZE chunks, which is only
256 bytes, so just put the dummy buffer into the softc.
no longer relevant (read: most of what was there) and adds some
quick links to point newcomers in the right direction.
Reviewed by: imp@
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D14680
Remove obsolete example. All services has they own example.
This example also show old type of limiting method which is
not recommended to use.
Reviewed by: bcr@
Differential Revision: https://reviews.freebsd.org/D14607
Link __bswap[ds]i2() intrinsics in to libzstd for riscv, where the C runtime
apparently lacks such intrinsics.
Broken in r330894.
Reported by: asomers
Sponsored by: Dell EMC Isilon
The slow path is always taken when lockstat is enabled. This induces
rdtsc (or other) calls to get the cycle count even when there was no
contention.
Still go to the slow path to not mess with the fast path, but avoid
the heavy lifting unless necessary.
This reduces sys and real time during -j 80 buildkernel:
before: 3651.84s user 1105.59s system 5394% cpu 1:28.18 total
after: 3685.99s user 975.74s system 5450% cpu 1:25.53 total
disabled: 3697.96s user 411.13s system 5261% cpu 1:18.10 total
So note this is still a significant hit.
LOCK_PROFILING results are not affected.
where we had not hit global dirty limits but a single queue was starved
for space by dirty buffers. A single buf_daemon is maintained for now.
Add a bd_speedup() when we are low on bufspace. This can happen due to SUJ
keeping many bufs locked until a cg block is written. Document this with
a comment.
Fix sysctls to work with per-domain variables. Add more ddb debugging.
Reported by: pho
Reviewed by: kib
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D14705