marius [Thu, 21 Apr 2011 21:56:28 +0000 (21:56 +0000)]
- Use the streaming cache unless BUS_DMA_COHERENT is specified. Since
r220375 all drivers enabled in the sparc64 GENERIC should be either
correctly using bus_dmamap_sync(9) calls or supply BUS_DMA_COHERENT
when appropriate or as a workaround for missing bus_dmamap_sync(9)
calls (sound(4) drivers and partially sym(4)). In at least some
configurations taking advantage of the streaming cache results in
a modest performance improvement.
- Remove the memory barrier for BUS_DMASYNC_PREREAD which as the
comment already suggested is bogus.
- Add my copyright for having implemented several things like support
for the Fire and Oberon IOMMUs, taking over PROM IOMMU mappings etc.
According to ATA specifications, when ATAPI master is the only device, it
should respond with all zeroes to any access to slave registers. Test with
PATA devices confirmed such behavior. Unluckily, Intel SATA controllers in
legacy emulation mode behave differently, not making any difference between
ATA and ATAPI devices. It causes false positive slave device detection and,
as result, command timeouts.
To workaround this problem, mask result of legacy-emulated soft-reset with
the device presence information received from the SATA-specific registers.
Remove the nm_mtx mutex locking from the test for
nm_maxfilesize. This value rarely, if ever, changes
and the nm_mtx mutex is locked/unlocked earlier in
the function, which should be sufficient to avoid
getting a stale cached value for it. There is a
discussion w.r.t. what these tests should be, but
I've left them basically the same as the regular
NFS client for now.
simon [Thu, 21 Apr 2011 16:40:34 +0000 (16:40 +0000)]
Expand / correct newsyslog regression tests:
- Test newslog with clasic naming of rotates files to actually test
the correct number of log files as newsyslog now does the correct
thing post r220926.
- Add some more newsyslog tests which tests keeping 0, 1, and 2
logfiles.
simon [Thu, 21 Apr 2011 16:31:05 +0000 (16:31 +0000)]
Fix an old bug in newsyslog where we kept one log file more than was
requested in newsyslog.conf. This was only the case using the non-time
based filenames (.0, .1, .2 etc.).
The change also makes newsyslog clean clean up the old extra logfile so
users don't end up with a single stale logfile which won't be rotated
out.
This change also cleans up some code a bit to avoid more copy / paste
code and removes some old copy / paste code in the process.
Add some more bit definitions:
- TCO_MESSAGEx: TCO specific regs providing the ability to monitor BIOS
bootup activity.
- TCO_NEWCENTURY: reporting RTC year roll over.
- TCO_NMI2SMI_EN, TCO_NMI_NOW: controlling SMIs conversion to NMIs and
NMI trigger.
- SMI_GBL_EN: Enabling SMI delivery for all the northbridge controller.
If number of keys for the given provider doesn't exceed the limit,
allocate all of them at attach time. This allows to avoid moving
keys around in the most-recently-used queue and needs no mutex
synchronization nor refcounting.
Instead of allocating memory for all the keys at device attach,
create reasonably large cache for the keys that is filled when
needed. The previous version was problematic for very large providers
(hundreds of terabytes or serval petabytes). Every terabyte of data
needs around 256kB for keys. Make the default cache limit big enough
to fit all the keys needed for 4TB providers, which will eat at most
1MB of memory.
- Fix mapping of the last two SATA ports on 6-port Intel controllers.
This improves hard-reset and hot-plug on these ports.
- Device with ID 0x29218086 is a 2-port variant of ICH9 in legacy mode.
Skip probing for nonexistent slave devices there.
Make PATA-like soft-reset in ata(4) more strict in checking disk signature.
It allows to avoid false positive device detection under Xen, that caused
long probe delays due to subsequent IDENTIFY command timeouts.
- Allows using full device name paths, such as /dev/ad0 or /dev/mirror/gm0 in config files
- Fixes some issues creating gmirror devices, including on GPT partitions
- Bugfixes for ZFS mirroring
- Enhanced GELI to work with a passphrase only, or key-file only
- Bugfix to prevent crashing of PC-BSD Live media when checking for upgrade partitions
adrian [Thu, 21 Apr 2011 03:59:37 +0000 (03:59 +0000)]
Implement very basic ALQ logging for net80211.
This is destined to be a lightweight and optional set of ALQ
probes for debugging events which are just impossible to debug
with printf/log (eg packet TX/RX handling; AMPDU handling.)
The probes and operations themselves will appear in subsequent
commits.
Add a check for VI_DOOMED at the beginning of nfscl_request()
so that it won't try and use vp->v_mount to do an RPC during
a forced dismount. There needs to be at least one more kernel
commit, plus a change to the umount(8) command before forced
dismounts will work for the experimental NFS client.
Ring the freelist doorbell from within refill_fl. While here, fix a bug
that could have allowed the hardware pidx to reach the cidx even though
the freelist isn't empty. (Haven't actually seen this but it was there
waiting to happen..)
This is only a problem if IFS contains digits, which is unusual but valid.
Because of an incorrect fix for PR bin/12137, "${#parameter}" was treated
as ${#parameter}. The underlying problem was that "${#parameter}"
erroneously added CTLESC bytes before determining the length. This
was properly fixed for PR bin/56147 but the incorrect fix was not backed
out.
Reported by: Seeker on forums.freebsd.org
MFC after: 2 weeks
When we become primary, we connect to the remote and expect it to be in
secondary role. It is possible that the remote node is primary, but only
because there was a role change and it didn't finish cleaning up (unmounting
file systems, etc.). If we detect such situation, wait for the remote node
to switch the role to secondary before accepting I/Os. If we don't wait for
it in that case, we will most likely cause split-brain.
r179417 introduced a bug into pthread_once(). Previously pthread_once()
used a global pthread_mutex_t for synchronization. r179417 replaced that
with an implementation that directly used atomic instructions and thr_*
syscalls to synchronize callers to pthread_once. However, calling
pthread_mutex_lock on the global mutex implicitly ensured that
_thr_check_init() had been called but with r179417 this was no longer
guaranteed. This meant that if you were unlucky enough to have your first
call into libthr be a call to pthread_once(), you would segfault when
trying to access the pointer returned by _get_curthread().
The fix is to explicitly call _thr_check_init() from pthread_once().
When calling XPT_REL_SIMQ to ajust number of openings, do not try to really
release device. We haven't frozen the device before and attempt to release
it will at least cause warning message from kernel.
Compile in in_cksum* implementations for both IPv6 and IPv6.
While in_pseudo() etc. is often used in offloading feature support,
in_cksum() is mostly used to fix some broken hardware.
Keeping both around for the moment allows us to compile NIC drivers
even in an IPv6 only environment without the need to mangle them
with #ifdef INETs in a way they are not prepared for. This will
leave some dead code paths that will not be exercised for IPv6.
Reviewed by: gnn
Sponsored by: The FreeBSD Foundation
Sponsored by: iXsystems
MFC after: 3 days
Not compiling in and not initializing from inetsw from in_proto.c for
IPv6 only, we need to initialize upper layer protocols from inet6sw.
Make sure to not initialize them twice in a Dual-Stack
environment but only conditionally on no INET as we have done for
TCP for a long time. Otherwise we would leak resources.
Reviewed by: gnn
Sponsored by: The FreeBSD Foundation
Sponsored by: iXsystems
MFC after: 3 days
Move the ipport_tick_callout and related functions from ip_input.c
to in_pcb.c. The random source port allocation code has been merged
and is now local to in_pcb.c only.
Use a SYSINIT to get the callout started and no longer depend on
initialization from the inet code, which would not work in an IPv6
only setup.
Reviewed by: gnn
Sponsored by: The FreeBSD Foundation
Sponsored by: iXsystems
MFC after: 4 days
Modify the offset + size checks for read and write in the
experimental NFS client to take care of overflows for the calls
above the buffer cache layer in a manner similar to r220876.
Thanks go to dillon at apollo.backplane.com for providing the
snippet of code that does this.
Modify the offset + size checks for read and write in the
experimental NFS client to take care of overflows. Thanks
go to dillon at apollo.backplane.com for providing the
snippet of code that does this.
Pull some features out of the firmware:
- If a ENH_SENS TLV section exit the firmware is capable of doing
enhanced sensitivity calibration.
- Newer devices/firmwares have more calibration commands therefore
hardcoding the noise gain/reset commands no longer works. It is
supposed to use the next index after the newest calibration type
support. Read the command index of the TLV section if available.
Scenario:
- We have two nodes connected and synchronized (local counters on both sides
are 0).
- We take secondary down and recreate it.
- Primary connects to it and starts synchronization (but local counters are
still 0).
- We switch the roles.
- Synchronization restarts but data is synchronized now from new primary
(because local counters are 0) that doesn't have new data yet.
This fix this issue we bump local counter on primary when we discover that
connected secondary was recreated and has no data yet.
dim [Tue, 19 Apr 2011 18:09:21 +0000 (18:09 +0000)]
Remove support for the Intel C Compiler from the build infrastructure.
This support has not worked for several years, and is not likely to work
again, unless Intel decides to release a native FreeBSD version of their
compiler. ;)
- Rewrite functions that copyin/out NAT configuration, so that they
calculate required memory size dynamically.
- Fix races on chain re-lock.
- Introduce new field to ip_fw_chain - generation count. Now utilized
only in the NAT configuration, but can be utilized wider in ipfw.
- Get rid of NAT_BUF_LEN in ip_fw.h
ipdn_bound_var() functions is designed to bound a variable between
specified minimum and maximum. In case when specified default value
is out of bounds it does not work as expected and does not limit
variable. Check that default value is in range and limit it if needed.
Also bump max_hash_size value to 65536 to correspond with manual page.
According to specification. device should respond to COMRESET with COMINIT
in no more then 10ms. If we detected no device presence within that time,
there is no reason to wait longer.
Fix up handling of the nfsmount structure in read and write
within the experimental NFS client. Mostly add mutex locking
and use the same rsize, wsize during the operation by keeping
a local copy of it. This is another change that brings it
closer to the regular NFS client.
Revert r220761 since, as kib@ pointed out, the case of
adding the check to nfsrpc_close() isn't useful. Also,
the check in nfscl_getcl() must be more involved, since
it needs to check before and after the acquisition of
the refcnt on nfsc_lock, while the mutex that protects
the client state data is held.
Do not assume PM timer GAS type is I/O or memory. It may be an unsupported
type, i. e., a broken table. Also, do not hardcode ACPI timer frequency in
device description.
Make pmap_invalidate_cache_range() available for consumption on amd64.
Add pmap_invalidate_cache_pages() method on x86. It flushes the CPU
cache for the set of pages, which are not neccessary mapped. Since its
supposed use is to prepare the move of the pages ownership to a device
that does not snoop all CPU accesses to the main memory (read GPU in
GMCH), do not rely on CPU self-snoop feature.
amd64 implementation takes advantage of the direct map. On i386,
extract the helper pmap_flush_page() from pmap_page_set_memattr(), and
use it to make a temporary mapping of the flushed page.
Reviewed by: alc
Sponsored by: The FreeBSD Foundation
MFC after: 3 weeks
Changed "conscontrol unset" to accept an existing virtual
console device as an argument. Unsetting virtual console
using /dev/console seems to have never worked.
When checking to see if a window update should be sent to the remote peer,
don't force a window update if the window would not actually grow due to
window scaling. Specifically, if the window scaling factor is larger than
2 * MSS, then after the local reader has drained 2 * MSS bytes from the
socket, a window update can end up advertising the same window. If this
happens, the supposed window update actually ends up being a duplicate ACK.
This can result in an excessive number of duplicate ACKs when using a
higher maximum socket buffer size.
Add the posix_fallocate(2) syscall. The default implementation in
vop_stdallocate() is filesystem agnostic and will run as slow as a
read/write loop in userspace; however, it serves to correctly
implement the functionality for filesystems that do not implement a
VOP_ALLOCATE.
Note that __FreeBSD_version was already bumped today to 900036 for any
ports which would like to use this function.
Also reserve space in the syscall table for posix_fadvise(2).
adrian [Mon, 18 Apr 2011 14:14:54 +0000 (14:14 +0000)]
For now, only enable GTT. CST is firing very frequently during local tests;
I'll figure out what's going on before re-enabling this as it does add
to the interrupt load.