This allocator uses a binary buddy system with a twist. First and
foremost, this allocator is required to support the implementation of
superpages. As a side effect, it enables a more robust implementation
of contigmalloc(9). Moreover, this reimplementation of
contigmalloc(9) eliminates the acquisition of Giant by
contigmalloc(..., M_NOWAIT, ...).
The twist is that this allocator tries to reduce the number of TLB
misses incurred by accesses through a direct map to small, UMA-managed
objects and page table pages. Roughly speaking, the physical pages
that are allocated for such purposes are clustered together in the
physical address space. The performance benefits vary. In the most
extreme case, a uniprocessor kernel running on an Opteron, I measured
an 18% reduction in system time during a buildworld.
This allocator does not implement page coloring. The reason is that
superpages have much the same effect. The contiguous physical memory
allocation necessary for a superpage is inherently colored.
Finally, the one caveat is that this allocator does not effectively
support prezeroed pages. I hope this is temporary. On i386, this is
a slight pessimization. However, on amd64, the beneficial effects of
the direct-map optimization outweigh the ill effects. I speculate
that this is true in general of machines with a direct map.
Approved by: re
eradication in/from userland path, countless locking fixes, etc.
- General sleep call through msleep(9) has been converted to condvar(9)
with better consistencies.
- Heavily guard every possible "slow path" entries (open(), close(),
few ioctl()s, sysctls), but once it entering "fast path" (io, interrupt
started), they are free to fly on their own.
- Rearrange locking sequences, resulting better concurrency and
serialization. Large part doesn't even need locking at all, and will be
removed in future. Less clutter, except in few places due to lock
ordering.
- Anonymous mixer object creation/deletion to simplify mixer handling
beyond typical mixer ioctls.
Submitted by: chibis (with modifications)
- Add few mix_[get|set|..] functions to avoid calling mixer_ioctl()
directly using cryptic arguments.
- Locking fixes to avoid possible deadlock with (still under Giant) USB.
- Better simplex/duplex device handling.
- Recover mmap() functionality for recording, which has been lost
since 2.2.x - 3.x (the introduction of newpcm). Full-duplex mmap still
doesn't work (due to VM/page design), but people still can mmap
both by opening each direction separately. mmaped playback is guarantee
to work either way.
- New sysctl: "hw.snd.compat_linux_mmap" to allow PROT_EXEC page
mapping, due to recent changes in linux compatibility layer which
require it. All linux applications that using sound + mmap() (mostly games)
require this to be enabled. Disabled by default.
- Other goodies.. too many, that will increase releng7 shareholder value
and make users of releng6 (and below) cry ;)
* This commit should be atomic. If anything goes wrong (not counting problem
originated from elsewhere), I will not hesitate to revert everything back
within 12 hours. This substantial changes itself not a rocket science
and the process has begun for almost 2 years, and lots of incremental
changes are already in place during that period of time.
* Some issues does occur in snd_emu10kx (note the 'x') due to various
internal locking issues and it is currently being worked on by chibis.
Tested by: chibis (Yuriy Tsibizov), joel, Alexandre Vieira,
many innocent souls...
Without bus_dma clean up and increment of number of Tx descriptors
it's hard to guarantee correct Tx operation in TSO case. The TSO
support would be enabled again when I get more feeback from re(4)
patch posted to current.
Please note that, this is currently considered as an
experimental feature so there could be some rough
edges. Consult http://wiki.freebsd.org/TMPFS for
more information.
For now, connect tmpfs to build on i386 and amd64
architectures only. Please let us know if you have
success with other platforms.
This work was developed by Julio M. Merino Vidal
for NetBSD as a SoC project; Rohit Jalan ported it
from NetBSD to FreeBSD. Howard Su and Glen Leeder
are worked on it to continue this effort.
Obtained from: NetBSD via p4
Submitted by: Howard Su (with some minor changes)
Approved by: re (kensmith)
- Remove unnecessary timestamps.
- Return CAM_RESRC_UNAVAIL for ORB shortage.
- Fix a lock problem when doorbell is used.
- Fix a potential bug for unordered execution.
'result' is still NULL and we do not need to free anything.
That allows us to gc the entire goto parts and a now unused variable.
Found with: Coverity Prevent(tm)
CID: 2519
do not continue with a NULL pointer. [1]
While here change the return of the error handling code path above.
I cannot see why we should always return 0 there. Neither does KAME
nor do we in here for the similar check in all the other functions.
Found with: Coverity Prevent(tm) [1]
CID: 2521
114 bytes of cmos ram in the PC clock chip. The big difference between
this and the Linux version is that we do not recalculate the checksums
for bytes 16..31.
We use this at work when cloning identical machines - we can copy the
bios settings as well. Reading /dev/nvram gives 114 bytes of data but
you can seek/read/write whichever bytes you like.
Yes, this is a "foot, gun, fire!" type of device.
without an mtag in ipsec4_common_input_cb.
So in case of !IPCOMP (AH,ESP) only change the m_tag_id if an mtag
was passed to ipsec4_common_input_cb.
Found with: Coverity Prevent(tm)
CID: 2523
handle, document those sprotos using an IPSEC_ASSERT so that it will
be clear that 'spi' will always be initialized when used the first time.
Found with: Coverity Prevent(tm)
CID: 2533
- In tdq_choose() only assert that a thread does not have too high a
priority (low value) for the queue we removed it from. This will catch
bugs in priority elevation. It's not a serious error for the thread
to have too low a priority as we don't change queues in this case as
an optimization.
Reported by: kris
thinking it had the whole chunk. This could cause a crash if
a large packet drop came in. Fixed by adjusting the trunc length
down to the limit.
- Large sacks with lots of segments could also have same issue. Changed
duplicate and segment handling to use proper get_m_ptr function to
pull each block from mbuf chains.
priorities, etc.) in the NICE field:
Use a combination of pri_native and pri_user instead of pri_level to
guess the original realtime priority. Using pri_level here has been
wrong since 2001/02/12. Using only pri_native here would be correct
if the kernel actually initialized it reasonably. (The kernel exports
its raw td_base_priority as pri_native, but userland mostly wants a
refined base priority). Give up on waiting pri_native to work correctly
and only use it when there is nothing better (for kthreads).
This should reduce printing of bizarre pseudo-nice values. Bizarre
values are still printed if we observe a transient borrowed priority
for a kthread (transient borrowing is the main thing that makes the
raw td_base_priority almost useless in userland), or if there is a
kernel bug. One current kernel bug involves the kernel idprio thread
pagezero permanently changing its priority from PRI_MAX_IDLE (255) to
PUSER (160). Then the bizarre value "ki-6" is printed instead of
"ki31". Here "-6" is PRI_MIN_IDLE - PUSER = -64 truncated to 2
characters. We are observing a transient borrowed priority that has
become permanent due to a bug.
ps/print.c:priorityr() needs similar changes (including ones in stage 2
here).
