| Age | Commit message (Collapse) | Author |
|---|
|
Since there's now no way to select maximum receive unit size the
hardware is programmed to accept frame sizes up to 9216 which is
now the maximum (down from 15.5K since this is supposed to work
in all advanced configurations and gives slightly better flow
control watermark ranges) and split all frames larger 2K into
multiple fragments (code was already there but wasn't enabled).
Tested on 82599 (SFP+) and X540 (10GBaseT).
With input from dlg@.
|
|
|
|
|
|
ok bcallah@ millert@
|
|
build flags into conditionals in the Makefile, fixing a few inconsistencies
in the process.
\o/ miod@
|
|
reboot and to cope with an upcoming change to /var/tmp.
Noted by and OK halex@
OK deraadt@
|
|
|
|
|
|
This is not necessary, as confirmed by tedu@ and kettenis@.
Solve the major contention problem seen on my Dual G5. Reduce the
build time of a kernel with two jobs from 5h+ to 18min and expose more
easily pmap/memroy corruptions 8)
ok kettenis@
|
|
changed. This saves a lot of I/O when attaching/detaching devices
and might help with some timing related problems.
Has been in snap for ten days, committing now so that people testing
xhci(4) test the same thing w/ snapshots and their own kernel.
|
|
less threatening and more informative.
e.g.
DHCPACK on <null address> to 5c:51:4f:56:81:c3 via em0
changes to
DHCPACK to 192.168.19.62 (3c:97:0e:0c:0c:d7) via em0
Issue noted and diff tested by Marc Peters. Thanks!
ok yasuoka@ millert@
|
|
return the number of items read of written.
When you intend to return the number of bytes actually processed, it is
wise to pass 1 as the item size and the size as the number of items.
But in *some* places, the OpenSSL does the opposite, and has extra logic
to change a successful return of 1 (item processed) into the real size.
And, guess why it does that? Because of old VMS, for they (used to) have a
substandard stdio implementation.
Note that this change causes the return values of BIO_dump_fp() and
BIO_dump_indent_fp() to no longer be useless (actual number of callback calls),
but actual bytes output. Given the irrelevance of the return value before,
it is unlikely that anything depends upon it (and if something does, it
probably has other problems in need for a fix...)
ok tedu@ beck@ jsing@
|
|
that contained at least one match in order to not prefer mdoc(1) from
ports over mdoc(7). As a bonus, this results in a speedup.
|
|
ok deraadt, guenther
|
|
bugs that panic the kernel and it's unclear if any such device works at all.
Anyone in possession of a working AR9300 device that stops working with this
commit please talk to me. I've only seen evidence of this code not working.
ok kirby@ mpi@ phessler@ dcoppa@
|
|
fixed as /logs within the chroot. As this httpd is properly privesp'ed this
has the nice property of allowing us to put the logs outside the chroot if
we want to.
ok reyk@
|
|
CryptAcquireContext and CryptGenRandom returns zero (FALSE) if fails.
From: Dongsheng Song <dongsheng.song@gmail.com>
|
|
the second "Set Address" command is issued. This would lead the
HC to reprocess TRBs corresponding to completed transfers.
This was the cause of the "xhci0: NULL xfer pointer" message that
could be seen after attaching a device and reported by naddy@.
|
|
since we might have tweaked the addresses.
Problem reported and fix test by Bastien Durel <bastien at geekwu ! org>,
thanks! OK henning
|
|
This should return the proper length again.
OK phessler, benno
|
|
to ASPLAIN.
Of course, you can still input AS numbers in either format.
OK henning@, claudio@, benno@, sthen@
|
|
other allocations in the same block couldn't.
problem pointed out by David Ramos on the openssl-dev list
ok miod@ doug@
|
|
* don't close the socket unless it was opened by the function
* note (in the comments) that the client is responsible for closing
the socket if they opened it, or they didn't use CLNT_DESTROY()
fixes a couple of unnecessary closing of already-closed sockets.
noted by: Matthias Drochner <M.Drochner@fz-juelich.de>"
tested by many in snaps
ok schwarze@
|
|
|
|
Usually, -h output is short, so the pager is just a nuisance.
Also, traditional man(1) does not use a pager for -h.
Triggered by a remark of deraadt@ on ICB.
|
|
requested by tedu@
|
|
|
|
mention ControlPersist=0 is the same as ControlPersist=yes
recommend that ControlPath sockets be placed in a og-w directory
|
|
not be in the cse code itself, actually). This is quite the pessimistic hammer
but that's the best we can do until someone with enough motivation debugs
this.
Behind-the scenes tedious research of the issue by afresh1@ trying to get
perl 5.20 to run on vax.
|
|
|
|
at the beginning of the first item of an enclosing list
right at the beginning of a new section or subsection;
minibug reported by Steffen Nurpmeso <sdaoden at yandex dot com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ok miod@
|
|
pool_setipl(9) has been called. This avoids the panic introduced in rev 1.139
(which was subsequently backed out) while still effectively guaranteeing a
consistent snapshot. Pools used from interrupt handlers should use the
appropriate pool IPL.
ok dlg@, deraadt@
|
|
types ICMP_ECHO and ICMP6_ECHO_REQUEST have their special meaning
only if the protocol matches.
Put an #ifdef INET6 around ICMP6_ECHO_REQUEST to make the kernel
without IPv6 compile.
OK henning@
|
|
|
|
|
|
This change fixes up SFP+ module detection during "ifconfig up"
after the machine has been booted without the modules plugged in.
|
|
since most of them can't be retrieved otherwise. This
comes with a slight but measurable performance increase
as well.
Also since the hardware has a single counter for missed
packets including those caused by the insufficient DMA
buffers available, this makes it hard to decipher actual
errors when used with Rx ring length limiting mechanisms
like if_rxr or mclgeti.
|
|
|
|
|
|
This code is violating various layers of abstraction, just like ehci(4)
does. Transaction translators need a bit more love.
|
|
|
