feeder.h:
feeder.c:
- Implement scoring mechanisme to select best format for conversion.
This is actually part of newer format chaining procedures which
will be commited someday. Confusion during chaining process solved
by this scoring since it will try to reduce list of from/to formats
to a single, best format.
Related PR: kern/91683
channel.c:
- Simplify feeder building process since we have smarter format
chaining.
- (1) Fix DMA alignment, based on bytes per sample.
feeder_fmt.c:
- Add few more sign conversion feeders for 24 and 32 bit format.
feeder_rate.c:
- Force buffer / bytes allignment. Unaligned buffer may cause
panics during recording on pure 32bit sample format if it
involves feeder_rate as part of feeders chain. Handle strayed
bytes (mostly caused by #1) better.
Tested on: ATI IXP, force 32bit recording.
This DMA alignment issues are extremely hard to reproduce unless the user
happen to have a 32bit capable soundcards (ATI IXP) and knowledgeable enough
to force it to operate under pure 32bit operations on both record and play
directions.
The problem with it was that it swapped the relative order of IP
assignment and parent interface attachment. The present if_vlan
code gets certain flags from the parent, including those meaningful
to the upper layers. E.g., IP assignment to an interface with
IFF_BROADCAST goes somewhat differently from that to a non-broadcast
interface. Consequently, assigning IP before attaching the parent
results in a bogus or missing broadcast address on the vlan interface.
This bug is still here, but at least it won't be triggered by doing
both vlanX configuration steps, IP and vlan+vlandev, in a single
ifconfig invocation, which is usual to setting up vlans via rc.conf.
Work at the global issue is under way.
rev. 1.39 sys/net/if_gre.c
rev. 1.21 sys/netinet/ip_gre.c rev. 1.4 sys/netinet/ip_gre.h
rev. 1.12 sys/netipsec/keysock.c rev. 1.3 sys/netipsec/keysock.h
rev. 1.32 sys/netkey/keysock.c rev. 1.9 sys/netkey/keysock.h
Fix stack corruptions on amd64.
Vararg functions have a different calling convention than regular
functions on amd64. Casting a varag function to a regular one to
match the function pointer declaration will hide the varargs from
the caller and we will end up with an incorrectly setup stack.
Entirely remove the varargs from these functions and change the
functions to match the declaration of the function pointers.
Remove the now unnecessary casts.
Lots of explanations and help from: peter
PR: amd64/89261
Use thread0 instead of user's thread for sobind.
This fixes reconnect after, for example, tcp idle disconnection.
Previously this would fail if a normal user tried to bind to a privileged
port.
a slightly different mechanism to load geom_md (g_me) which was renamed at
some point. This has been tested on RELENG_6 by erwin and glebius.
Reviewed by: erwin, glebius
- hw.ata.wc can cause data loss on crashes too. This brings the
description of the hw.ata.wc in sync with tuning(7)
- Via VT8251 support.
- Support for Marvell chipsets.
Add a64l(), l64a(), and l64a_r() XSI extentions. These functions convert
between a 32-bit integer and a radix-64 ASCII string. The l64a_r() function
is a NetBSD addition.
PR: 51209 (based on submission, but very different)
Reviewed by: bde, ru
date: 2006/01/20 03:21:25; author: csjp; state: Exp; lines: +1 -0
Destroy the dynamic rule zone in the event that we fail to insert the
initial default rule.
MFC revision 1.123
date: 2006/01/20 05:35:27; author: csjp; state: Exp; lines: +25 -10
- Change the return type for init_tables from void to int so we can propagate
errors from rn_inithead back to the ipfw initialization function.
- Check return value of rn_inithead for failure, if table allocation has
failed for any reason, free up any tables we have created and return ENOMEM
- In ipfw_init check the return value of init_tables and free up any mutexes or
UMA zones which may have been created.
- Assert that the supplied table is not NULL before attempting to dereference.
This fixes panics which were a result of invalid memory accesses due to failed
table allocation. This is an issue mainly because the R_Zalloc function is a
malloc(M_NOWAIT) wrapper, thus making it possible for allocations to fail.
Found by: Coverity Prevent (tm)
Coverity ID: CID79
MFC after: 1 week
from HEAD to RELENG_6:
Add a "comconsole_speed" loader variable that can be used to change
the serial console speed (i386 and amd64 only). If the previous
stage boot loader requested a serial console (RB_SERIAL or RB_MULTIPLE)
then the default speed is determined from the current serial port
speed. Otherwise it is set to 9600 or the value of BOOT_COMCONSOLE_SPEED
at compile time.
Approved by: rwatson (mentor)
MFC 1.88:
Improve handling flags that must be propagated
to the parent interface, such as IFF_PROMISC and
IFF_ALLMULTI. In addition, vlan(4) gains ability
to migrate from one parent to another w/o losing
its own flags.
PR: kern/81978
Mega update to the LSI MegaRAID driver:
1. Implement a large set of ioctl shims so that the Linux management apps
from LSI will work. This includes infrastructure to support adding, deleting
and rescanning arrays at runtime. This is based on work from Doug Ambrosko,
heavily augmented by LSI and Yahoo.
2. Implement full 64-bit DMA support. Systems with more than 4GB of RAM
can now operate without the cost of bounce buffers. Cards that cannot do
64-bit DMA will automatically revert to using bounce buffers. This option
can be forced off by setting the 'hw.amr.force_sg32" tunable in the loader.
It should only be turned off for debugging purposes. This work was sponsored
by Yahoo.
3. Streamline the command delivery and interrupt handler paths after
much discussion with Dell and LSI. The logic now closely matches the
intended design, making it both more robust and much faster. Certain
i/o failures under heavy load should be fixed with this.
4. Optimize the locking. In the interrupt handler, the card can be checked
for completed commands without any locks held, due to the handler being
implicitely serialized and there being no need to look at any shared data.
Only grab the lock to return the command structure to the free pool. A
small optimization can still be made to collect all of the completions
together and then free them together under a single lock.
Items 3 and 4 significantly increase the performance of the driver. On an
LSI 320-2X card, transactions per second went from 13,000 to 31,000 in my
testing with these changes. However, these changes are still fairly
experimental and shouldn't be merged to 6.x until there is more testing.
Thanks to Doug Ambrisko, LSI, Dell, and Yahoo for contributing towards
this.
These have now had more testing and Scott requested them to be merged
before the freeze.
o Add workaround for dropped interrupts on VIA and ATI controllers
o Add various identification strings and vendor IDs
o Set sc_dying to 1 when detaching
o Don't bother clearing the unused transfer `hcpriv' field
them to userland. [1]
MFC rev. 1.16 of sys/contrib/pf/net/pf_norm.c: Correct an error in pf
handling of IP packet fragments which could result in a kernel panic.
[2]
Security: FreeBSD-SA-06:06.kmem [1]
Security: FreeBSD-SA-06:07.pf [2]
Bring ATA up to -current standards:
Fix SiS SATA support, the SATA registers was off.
Update the ICH7 support so it deals better with chips without AHCI.
Unbreak hotplug support on the ICH6 and ICH7 chipsets.
Add support for VIA VT8251 southbridge.
Add new nVidia nForce4 chips.
Add support for the Marvell 88SX[56]0[48][01] series of SATA chips.
Fix the promise modesetting for old chips.
Get rid of the advertising clause in the copyright.
Add support for using DMA on dump, greatly speeds up the dump process.
When IOCATAGPARM is called, update the capabilities page that is stored
in the kernel and return the new values.
Fix rebuilds of arrays that got stuck.
Add dump support in ataraid.
Add support for for reading and writing SiS metadata.
Add support for writing VIA metadata.
Add support for writing Intel metadata.
Correct calculation of RAID0 sizes on VIA RAID arrays.
Update Intel MatrixRAID support to be able to pick up RAID0+1 (RAID10)
rmdir() fails with EINVAL if the last component of the path is "..".
Document the previously undocumented behaviour of rmdir(".") returning
EINVAL as required by POSIX and SUSv3. Bump the man page change date.
undelete() fails with EINVAL if the last component of the path is "..".
Bump the man page change date.
Return EINVAL from lookup() if cn_nameiop is DELETE or RENAME and
the last component of the path name is "..". This keeps VOP_LOOKUP()
from locking vnodes in reverse order.
In kern_unlink(), remap EINVAL errors returned from namei() to EPERM
to match existing (and POSIX required) behaviour.
