emax [Wed, 23 May 2012 18:56:29 +0000 (18:56 +0000)]
Tweak condition for disabling allocation from per-CPU buckets in
low memory situation. I've observed a situation where per-CPU
allocations were disabled while there were enough free cached pages.
Basically, cnt.v_free_count was sitting stable at a value lower
than cnt.v_free_min and that caused massive performance drop.
jhb [Wed, 23 May 2012 13:45:52 +0000 (13:45 +0000)]
Rework the previous change to honor MADT processor IDs when probing
processor objects. Instead of forcing the new-bus CPU objects to use
a unit number equal to pc_cpuid, adjust acpi_pcpu_get_id() to honor the
MADT IDs by default. As with the previous change, setting
debug.acpi.cpu_unordered to 1 in the loader will revert to the old
behavior.
jhb [Wed, 23 May 2012 13:41:12 +0000 (13:41 +0000)]
Only check to see if a memory resource is a PCI ROM BAR when activating
and deactivating PCI resources. Previously, if a device had more than
48 MSI interrupts, then activating message 48 (which has a rid == PCIR_BIOS)
would incorrectly try to enable the PCI ROM BAR.
Tested by: Olivier Cinquin ocinquin uci edu
MFC after: 3 days
yongari [Wed, 23 May 2012 03:35:08 +0000 (03:35 +0000)]
Don't force max payload size to 128. Root complex and Endpoint will
negotiate with each other on the TLP payload size so blindly
forcing the size to 128 can cause a completion error which in turn
will stop device.
Reported by: Geans Pin < geanspin <> broadcom dot com >
MFC after: 5 days
yongari [Wed, 23 May 2012 01:20:25 +0000 (01:20 +0000)]
Make IPMI work in the bce driver even when the interface is
configured down. Formerly, IPMI communication was lost whenever the
interface was not up. The reason was that the BCE_EMAC_MODE
register was not configured with the correct media settings. There
are two parts to the fix.
First, resetting the chip in bce_reset() causes the BCE_EMAC_MODE
register to be initialized to a default value that does not
necessarily correspond to the actual media settings. The fix
implemented here is a bit of a hack. Ideally, at the end of
bce_reset() we would poll the PHY to determine the negotiated media,
and then we would set the BCE_EMAC_MODE register accordingly. That
is difficult, since the PHY is abstracted behind the MII layer and is
not supposed to be queried directly from the MAC driver. Instead,
we read the BCE_EMAC_MODE register at the beginning of bce_reset()
and then restore its media bits to their original values before
returning. If IPMI is up and running, then the link is already
established and the BCE_EMAC_MODE register is already set appropriately
when bce_reset() is called. If IPMI is not running, no harm is
done by preserving the BCE_EMAC_MODE settings. The driver will set
the register properly once the interface is configured up and link
is established.
Second, bce_miibus_statchg() is sometimes called when the link is
down. In that case, the reported media settings are invalid.
Formerly, the driver used them anyway to setup the BCE_EMAC_MODE
register. We now avoid changing any MAC registers unless link is
active and the reported media settings are valid.
Submitted by: jdp
Tested by: jdp
MFC after: 5 days
adrian [Tue, 22 May 2012 19:50:21 +0000 (19:50 +0000)]
Re-up the TX ath_buf limit from 128 to 512.
I'll have to leave this high for now, until I've done some significant
surgery with how ath_bufs (and descriptors) are handled.
This should significantly cut down on the opportunities for a full TX
queue hanging traffic. I'll continue making things work though; I'm
mostly doing this for users. :)
adrian [Tue, 22 May 2012 19:37:12 +0000 (19:37 +0000)]
Fix some corner cases in the ieee80211_send_bar() handling.
* If the first call succeeded but failed to transmit, a timer would
reschedule it via bar_timeout(). Unfortunately bar_timeout() didn't
check the return value from the ieee80211_send_bar() reattempt and
if that failed (eg the driver ic_raw_xmit() failed), it would never
re-arm the timer.
* If BARPEND is cleared (which ieee80211_send_bar() will do if it can't
TX), then re-arming the timer isn't enough - once bar_timeout() occurs,
it'll see BARPEND is 0 and not run through the rest of the routine.
So when rearming the timer, also set that flag.
* If the TX wasn't occuring, bar_tx_complete() wouldn't be called and the
driver callback wouldn't be called either. So the driver had no idea
that the BAR TX attempt had failed. In the ath(4) case, TX would stay
paused.
(There's no callback to indicate that BAR TX had failed or not;
only a "BAR TX was attempted". That's a separate, later problem.)
So call the driver callback (ic_bar_response()) before the ADDBA session
is torn down, so it has a chance of being notified that things didn't
quite go to plan.
I've verified that yes, this does suspend traffic for ath(4), retry BAR
TX even if the driver is failing ic_raw_xmit(), and then eventually giving
up and sending a DELBA. I'll address the "out of ath_buf" issue in ath(4)
in a subsequent commit - this commit just fixes the edge case where any
driver is (way) out of internal buffers/descriptors and fails frame TX.
bapt [Tue, 22 May 2012 16:33:10 +0000 (16:33 +0000)]
Fix world after byacc import:
- old yacc(1) use to magicially append stdlib.h, while new one don't
- new yacc(1) do declare yyparse by itself, fix redundant declaration of
'yyparse'
trasz [Tue, 22 May 2012 15:58:27 +0000 (15:58 +0000)]
Fix panic with RACCT that could occur in low memory (or out of swap)
situations, due to fork1() calling racct_proc_exit() without calling
racct_proc_fork() first.
Submitted by: Mateusz Guzik <mjguzik at gmail dot com> (earlier version)
Reviewed by: Mateusz Guzik <mjguzik at gmail dot com>
kib [Tue, 22 May 2012 11:07:44 +0000 (11:07 +0000)]
Add the code for new Intel GPU driver, which supports GEM, KMS and
works with new generations of GPUs (IronLake, SandyBridge and
supposedly IvyBridge).
The driver is not connected to the build yet.
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
kib [Tue, 22 May 2012 10:59:26 +0000 (10:59 +0000)]
A rewrite of the i810 bits of the agp(4) driver. New driver supports
operations required by GEMified i915.ko. It also attaches to SandyBridge
and IvyBridge CPU northbridges now.
Sponsored by: The FreeBSD Foundation
MFC after: 1 month
trasz [Tue, 22 May 2012 10:54:42 +0000 (10:54 +0000)]
Fix enforcement of file size limit with O_APPEND on ZFS.
vn_rlimit_fsize takes uio->uio_offset and uio->uio_resid into account
when determining whether given write would exceed RLIMIT_FSIZE.
When APPEND flag is specified, ZFS updates uio->uio_offset to point to the
end of file.
But this happens after a call to vn_rlimit_fsize, so vn_rlimit_fsize check
can be rendered ineffective by thread that opens some file with O_APPEND
and lseeks below RLIMIT_FSIZE before calling write.
harti [Tue, 22 May 2012 09:59:49 +0000 (09:59 +0000)]
Fix a compilation error with some compilers: __attribute__
requires two parenthesis for its argument, but instead of using
__attribute__ directly, use the appropriate __nonnull macro
from cdefs.h.
harti [Tue, 22 May 2012 07:23:41 +0000 (07:23 +0000)]
Make dumptid non-static. It is used by libkvm to detect whether
this is a VNET-kernel or not. gcc used to put the static symbol into
the symbol table, clang does not. This fixes the 'netstat: no namelist'
error seen on clang+VNET systems.
andrew [Tue, 22 May 2012 07:04:23 +0000 (07:04 +0000)]
Fix booting on ARM.
In PHYS_TO_VM_PAGE() when VM_PHYSSEG_DENSE is set the check if we are past
the end of vm_page_array was incorrect causing it to return NULL. This
value is then used in vm_phys_add_page causing a data abort.
adrian [Tue, 22 May 2012 06:31:03 +0000 (06:31 +0000)]
Fix up some corner cases with aggregation handling.
I've come across a weird scenario in net80211 where two TX streams will
happily attempt to setup an aggregation session together.
If we're very lucky, it happens concurrently on separate CPUs and the
total lack of locking in the net80211 aggregation code causes this stuff
to race. Badly.
So >1 call would occur to the ath(4) addba start, but only one call would
complete to addba complete or timeout. The TID would thus stay paused.
The real fix is to implement some proper per-node (or maybe per-TID)
locking in net80211, which then could be leveraged by the ath(4) TX
aggregation code.
Whilst I'm at it, shuffle around the debugging messages a bit.
I like to keep people on their toes.
jimharris [Mon, 21 May 2012 22:54:33 +0000 (22:54 +0000)]
Wait until completion context unwinds before retrying CCBs that have been
queued internally. This works around issue in the isci HAL where it cannot
accept new I/O to a device after a resetting->ready state transition until
the completion context has unwound.
This issue was found by submitting non-tagged CCBs through pass(4) interface
to a SATA disk with an extremely small timeout value (5ms). This would trigger
internal resets with I/O in the isci(4) internal queues.
The small timeout value had not been intentional (and original reporter has
since changed his test to use 5sec instead), but it did uncover this corner
case that would result in a hung disk.
Sponsored by: Intel
Reported and tested by: Ravi Pokala <rpokala at panasas dot com>
Reviewed by: scottl (earlier version)
MFC after: 1 week
melifaro [Mon, 21 May 2012 22:19:19 +0000 (22:19 +0000)]
Call bpf_jitter() before acquiring BPF global lock due to malloc() being used inside bpf_jitter.
Eliminate bpf_buffer_alloc() and allocate BPF buffers on descriptor creation and BIOCSBLEN ioctl.
This permits us not to allocate buffers inside bpf_attachd() which is protected by global lock.
melifaro [Mon, 21 May 2012 22:17:29 +0000 (22:17 +0000)]
Fix old panic when BPF consumer attaches to destroying interface.
'flags' field is added to the end of bpf_if structure. Currently the only
flag is BPFIF_FLAG_DYING which is set on bpf detach and checked by bpf_attachd()
Problem can be easily triggered on SMP stable/[89] by the following command (sort of):
'while true; do ifconfig vlan222 create vlan 222 vlandev em0 up ; tcpdump -pi vlan222 & ; ifconfig vlan222 destroy ; done'
Fix possible use-after-free when BPF detaches itself from interface, freeing bpf_bif memory,
while interface is still UP and there can be routes via this interface.
Freeing is now delayed till ifnet_departure_event is received via eventhandler(9) api.
Convert bpfd rwlock back to mutex due lack of performance gain (currently checking if packet
matches filter is done without holding bpfd lock and we have to acquire write lock if packet matches)
melifaro [Mon, 21 May 2012 22:13:48 +0000 (22:13 +0000)]
Fix panic on attaching to non-existent interface (introduced by r233937, pointed by hrs@)
Fix panic on tcpdump being attached to interface being removed (introduced by r233937, pointed by hrs@ and adrian@)
Protect most of bpf_setf() by BPF global lock
Add several forgotten assertions (thanks to adrian@)
Document current locking model inside bpf.c
Document EVENTHANDLER(9) usage inside BPF.
ghelmer [Mon, 21 May 2012 21:10:00 +0000 (21:10 +0000)]
Add checks for memory allocation failures in appropriate places, and
avoid creating bad entries in the grp list as a result of memory allocation
failures while building new entries.
PR: bin/83340
Reviewed by: delphij (prior version of patch)
imp [Mon, 21 May 2012 04:24:58 +0000 (04:24 +0000)]
Another minor re-arrangement of the code: calcualte the master clock
frequency in the at91_pmc_clock_init rather than passing it in. Allow
for frequencies >= 21MHz by rounding to the nearest 500Hz (Idea from
Ian Lapore whose company uses a similar arrangement in their product).
at91_pmc_clock_init() is now nearly independent of the rest of the pmc
driver (which means we may be able to call it much earlier in boot
soon to eliminate the master clock config file requirement for printf
to work during early boot and also eliminate some interdependencies
with the device ordering which requires pmc to be the first device
added).
imp [Sun, 20 May 2012 20:50:40 +0000 (20:50 +0000)]
Minor cleanup before some more major changes:
o main_clock_hz isn't used, eliminate it
o move main clock calculation code and table so we have only one ifdef.
alc [Sun, 20 May 2012 18:25:45 +0000 (18:25 +0000)]
Replace all uses of the vm page queues lock by a r/w lock that is private
to this pmap.c. This new r/w lock is used primarily to synchronize access
to the PV lists. However, it will be used in a somewhat unconventional
way. As finer-grained PV list locking is added to each of the pmap
functions that acquire this r/w lock, its acquisition will be changed from
write to read, enabling concurrent execution of the pmap functions with
finer-grained locking.
kientzle [Sun, 20 May 2012 18:07:35 +0000 (18:07 +0000)]
Determine kernel load address dynamically from u-boot memory map.
The generic ELF loading code maps the kernel into low memory
by subtracting KERN_BASE. So the copyin/copyout/readin functions
are always called with low addresses. This code finds the largest
DRAM block from the U-Boot memory map and adds that base to
the addresses.
In particular, this fixes ubldr on AM3358, which has DRAM
mapped to 0x80000000 at power-on.
nwhitehorn [Sun, 20 May 2012 14:33:28 +0000 (14:33 +0000)]
Replace the list of PVOs owned by each PMAP with an RB tree. This simplifies
range operations like pmap_remove() and pmap_protect() as well as allowing
simple operations like pmap_extract() not to involve any global state.
This substantially reduces lock coverages for the global table lock and
improves concurrency.
marius [Sun, 20 May 2012 05:12:31 +0000 (05:12 +0000)]
Rewrite nd6_sysctl_{d,p}rlist() to avoid misaligned accesses to char arrays
casted to structs by getting rid of these buffers entirely. In r169832, it
was tried to paper over this issue by 32-bit aligning the buffers. Depending
on compiler optimizations that still was insufficient for 64-bit architectures
with strong alignment requirements though.
While at it, add comments regarding the total lack of locking in this area.
adrian [Sun, 20 May 2012 02:49:42 +0000 (02:49 +0000)]
Migrate the TDMA management functions out of if_ath.c into if_ath_tdma.c.
There's some TX path TDMA code in if_ath_tx.c which should be migrated
out, but first I should likely try and verify/fix/repair the TDMA support
in 9.x and -HEAD.
adrian [Sun, 20 May 2012 02:05:10 +0000 (02:05 +0000)]
Migrate the bulk of the RX routines out from if_ath.c to if_ath_rx.[ch].
* migrate the rx processing out into if_ath_rx.c
* migrate the TSF functions into if_ath_tsf.h, as inlines
This is in prepration for supporting the EDMA RX routines, required to
support the AR93xx series NICs.
TODO:
* ath_start() shouldn't be private, but it's called as part of
the RX path. I should likely migrate ath_rx_tasklet() back into
if_ath.c and then return this to be 'static'. The RX code really
shouldn't need to see TX routines (and vice versa.)
bz [Sat, 19 May 2012 23:25:57 +0000 (23:25 +0000)]
MFp4 bz_ipv6_fast:
Allow tso4 and tso6 be set individually given we have the bits.
This will help with drivers not working as expected during the
transition time and later.
Sponsored by: The FreeBSD Foundation
Sponsored by: iXsystems
marcel [Sat, 19 May 2012 18:05:00 +0000 (18:05 +0000)]
Add build opton MK_LS_COLORS to control whether ls(1) supports colors
(and thus needs to depend on libtermcap). Embedded systems may not
want or need colors.