| Age | Commit message (Collapse) | Author |
|---|
|
the pci write buffers.
|
|
bridge interrupts dance when trying to access an uninplemented ioc3 register.
Makes PIC handling simpler as a bonus.
|
|
don't return the error back to the caller, but also don't turn on
unsolicited responses for the volume knob. chances are, it isn't
acutally used. lets sony viao tz play audio through suspend/resume
cycle.
* add some debug messages for other volume knob errors.
|
|
1. though shalt not hold a mutex while sleeping, which kthread_create can
do. instead of holding the wq mutex over the kthread_create and increasing
the number of running threads after kthread_create succeeds, this counts
the thread and drops the mutex before kthread_create. after the call it
takes the mutex again and decrements the number of threads if the call
failed.
2. if a workq is created during autoconf, the actual thread create
is deferred to when the scheduler is running. if the workq is
destroyed before then, the wq memory gets freed and then the deferred
thread creation will be using freed memory. we now have a workq
state variable so we can do the right thing at the different stages
of the workqs lifetime.
ok matthew@
|
|
from alexey suslikov
ok marco@
|
|
|
|
from alexey suslikov
ok marco@
|
|
|
|
For masks of identical length rn_lexobetter() did not stop on the
first non-equal byte. This leads rn_addroute() to not detecting
duplicate entries and thus we might create a very long list of masks
to check for each node.
This can have a huge impact on IPsec performance, where non-contiguous
masks are used for the flow lookup. In a setup with 1300 flows we
saw 400 duplicate masks and only a third of the expected throughput.
Lots of help in narrowing this down from markus@.
Improved comments from claudio@.
OK markus@, claudio@
|
|
|
|
|
|
|
|
|
|
ok miod@ (some time ago)
|
|
and let exec_elf.h do the right thing.
ok miod@ (some time ago)
|
|
steal a page from dlg@ and embed a workq_task struct directly in
the aoe_handler struct so that we won't fail when enqueueing a task
while here, create real debugging printfs vice commenting out regular ones,
and kill with fire the excessive number of includes
ok marco@ tedu@
|
|
bothering to check the mount type when FFS2 support is omitted.
ok krw@, jasper@; "i like it" tedu@
|
|
ok krw@, "of course" tedu@
|
|
found via the clang static analyser
|
|
|
|
It makes even the smallest run(4) devices look ridiculously huge.
|
|
|
|
|
|
getting stupid ideas like optimizing away stores to the descriptor that we're
setting. This may be overkill, but this code is far from performance
critical and it may prevent future surprises. Fixes instant reboots
with bsd.rd on Pentiums with the F00F bug.
Thanks to espie@, for narrowing the issue down enough for me to find the
problem. ok deraadt@
|
|
machdep.console_device that's only implemented on a few architectures.
ok deraadt@, miod@
|
|
making DEVNAME() usage a lot more consistent.
ok krw@
|
|
* use the correct variable in a DPRINTF
|
|
complete outstanding requests, so use uaudio_drain in uaudio_detach.
* the channel alt index can still be -1 at detach. check for this.
|
|
linux emulation layer, nor will they in the future.
ok miod@
|
|
|
|
|
|
restore the 100us delay after writing to an analog register since
it turns out that it is required.
|
|
tested by ray@
- no periodic PA calibration for the Osprey (AR9380)
|
|
|
|
ok jsing@
|
|
|
|
|
|
them explicitely pending before triggering the softintr; I am ashamed I did
not notice this when changing the soft interrupt code 18 months ago.
Noticed by claudio@ and beck@
|
|
(deepest sleep) state upon suspend, and restore power upon resume.
ok deraadt@
|
|
1.5, about 5 years ago. My bad.
ok deraadt@
|
|
over the PROM page tables on sun4c; also crank the tape kernel generous
size estimate to 4MB; crank version.
ok deraadt@
|
|
from cdboot to slim it down below the 32k limit.
|
|
"go ahead" kettenis@
|
|
|
|
has enough space for all the fragments on it.
this check was snuck in by itojun under an unrelated commit. it broke when
i set the virtual interface send queue depths to 1, which beck had to
special case at n2k10. without this code we avoid these dubious checks
along with another splnet/splx pair, and it should make future work on
manipulating send queues easier.
ive been running this in production since n2k10 (~7months ago).
ok claudio@ henning@ deraadt@
|
|
no binary change.
|
|
on AR9285 and AR9287.
|
|
|
|
compensate Tx gain for temperature changes.
|
|
value in the Tx gain table.
This fixes AR9280/AR9281 chips in open-loop power control mode, like
the high power solution found in the Acer Aspire One.
tested by ray@
|
