dteske [Sat, 2 Mar 2013 16:18:40 +0000 (16:18 +0000)]
Update FTP directories to include new layout. Previously discussed with
kensmith and others as this alleviates the need for the symbolic link upkeep
which is currently manually maintained.
melifaro [Sat, 2 Mar 2013 15:11:20 +0000 (15:11 +0000)]
Merge
* r233937 - Improve BPF locking model
* r233938 - Improve performace for writer-only BPF users
* r233946 - Fix build
* r235744 - Fix (new) panic on attaching to non-existent interface
* r235745 - Fix old panic when BPF consumer attaches to destroying interface
* r235746 - Call bpf_jitter() before acquiring BPF global lock
* r235747 - Make most BPF ioctls() SMP-safe.
* r236231 - Fix BPF_JITTER code broken by r235746.
* r236251 - Fix shim for BIOCSETF to drop all packets buffered on the descriptor.
* r236261 - Save the previous filter right before we set new one.
* r236262 - Fix style(9) nits, reduce unnecessary type castings.
* r236559 - Fix panic introduced by r235745
* r236806 - Fix typo introduced in r236559.
r233937
- Improve BPF locking model.
Interface locks and descriptor locks are converted from mutex(9) to rwlock(9).
This greately improves performance: in most common case we need to acquire 1
reader lock instead of 2 mutexes.
- Remove filter(descriptor) (reader) lock in bpf_mtap[2]
This was suggested by glebius@. We protect filter by requesting interface
writer lock on filter change.
- Cover struct bpf_if under BPF_INTERNAL define. This permits including bpf.h
without including rwlock stuff. However, this is is temporary solution,
struct bpf_if should be made opaque for any external caller.
r233938
- Improve performace for writer-only BPF users.
Linux and Solaris (at least OpenSolaris) has PF_PACKET socket families to send
raw ethernet frames. The only FreeBSD interface that can be used to send raw
frames is BPF. As a result, many programs like cdpd, lldpd, various dhcp stuff
uses BPF only to send data. This leads us to the situation when software like
cdpd, being run on high-traffic-volume interface significantly reduces overall
performance since we have to acquire additional locks for every packet.
Here we add sysctl that changes BPF behavior in the following way:
If program came and opens BPF socket without explicitly specifyin read filter
we assume it to be write-only and add it to special writer-only per-interface
list. This makes bpf_peers_present() return 0, so no additional overhead is
introduced. After filter is supplied, descriptor is added to original
per-interface list permitting packets to be captured.
Unfortunately, pcap_open_live() sets catch-all filter itself for the purpose
of setting snap length.
Fortunately, most programs explicitly sets (event catch-all) filter after
that. tcpdump(1) is a good example.
So a bit hackis approach is taken: we upgrade description only after second
BIOCSETF is received.
Sysctl is named net.bpf.optimize_writers and is turned off by default.
- While here, document all sysctl variables in bpf.4
r233946
Fix build broken by r233938.
r235744
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.
r235745
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)
r235746
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.
r235747
Make most BPF ioctls() SMP-safe.
r236559
Fix panic introduced by r235745. Panic occurs after first packet traverse
renamed interface.
Add several comments on locking
r236231
Fix BPF_JITTER code broken by r235746.
r236251
Fix 32-bit shim for BIOCSETF to drop all packets buffered on the descriptor
and reset statistics as it should.
r236261
- Save the previous filter right before we set new one.
- Reduce duplicate code and make it little easier to read.
r236262
Fix style(9) nits, reduce unnecessary type castings, etc., for bpf_setf().
dim [Sat, 2 Mar 2013 14:39:48 +0000 (14:39 +0000)]
MFC r247205:
Pull in r175962 from upstream llvm trunk:
X86: Disable cmov-memory patterns on subtargets without cmov.
Fixes PR15115.
For the i386 arch, this should enable cmov instructions only on
-march=pentiumpro and higher. Since our default CPU is i486, cmov
instructions will now be disabled by default.
ian [Fri, 1 Mar 2013 20:48:53 +0000 (20:48 +0000)]
MFC r246121 ...
Fix a descriptor leak in devd. Clients reading /var/run/devd.pipe can close
their socket connection any time, and devd only notices that when it gets an
error trying to write an event to the client. On a system with no device
change activity, clients could connect and disappear repeatedly without devd
noticing, leading to an ever-growing list of open socket descriptors in devd.
Now devd uses poll(2) looking for POLLHUP on all existing clients every time
a new client connection is established, and also periodically (once a minute)
to proactively find zombie clients and reap the socket descriptors. It also
now has a connection limit, configurable with a new -l <num> command line arg.
When the maximum number of connections is reached it stops accepting new
connections until some current clients drop off.
jhb [Fri, 1 Mar 2013 19:01:40 +0000 (19:01 +0000)]
MFC 246129:
Allow the address and ports to be separated by a colon or period rather
than a space to permit directly pasting the output of commands such as
netstat and sockstat on the command line.
jhb [Fri, 1 Mar 2013 18:39:46 +0000 (18:39 +0000)]
MFC 245849:
Don't assume that all Linux TCP-level socket options are identical to
FreeBSD TCP-level socket options (only the first two are). Instead,
using a mapping function and fail unsupported options as we do for other
socket option levels.
dim [Fri, 1 Mar 2013 17:37:57 +0000 (17:37 +0000)]
MFC r247166:
Pull in r172354 from upstream clang trunk:
Refactor the x86 CPU name logic in the driver and pass -march and -mcpu
flag information down from the Clang driver into the Gold linker plugin
for LTO. This allows specifying -march on the linker commandline and
should hopefully have it pass all the way through to the LTO optimizer.
Fixes PR14697.
Pull in r175919 from upstream clang trunk:
Driver: Pass down the -march setting down to -cc1as on x86 too.
The assembler historically didn't make use of any target features, but this has
changed when support for old CPUs that don't support long nops was added.
This should fix the long nops that still occurred in crt*.o, and
possibly other object files, if the system was compiled for a CPU that
does not support those, such as Geode.
Note that gcc on i386 also does not pass through any -march, -mcpu or
-mtune setting to gas, but this has not caused any trouble yet, because
gas defaults to i386.
jhb [Fri, 1 Mar 2013 16:18:40 +0000 (16:18 +0000)]
MFC 246035:
- Compute the correct size to reallocate when doubling the size of the
array of loaded objects to avoid a buffer overrun.
- Use reallocf() to avoid leaking memory if the realloc() fails.
jhb [Fri, 1 Mar 2013 14:54:26 +0000 (14:54 +0000)]
MFC 245577,245640:
Don't attempt to use clflush on the local APIC register window. Various
CPUs exhibit bad behavior if this is done (Intel Errata AAJ3, hangs on
Pentium-M, and trashing of the local APIC registers on a VIA C7). The
local APIC is implicitly mapped UC already via MTRRs, so the clflush isn't
necessary anyway.
jhb [Fri, 1 Mar 2013 03:04:57 +0000 (03:04 +0000)]
MFC 245823,245824,246210
- Use decimal values for UDP, TCP, and UNIX domain socket options rather
than hex to avoid implying that these constants should be treated as bit
masks.
- Add placeholder constants to reserve a portion of the socket option
name space for use by downstream vendors to add custom options.
jhb [Fri, 1 Mar 2013 01:02:28 +0000 (01:02 +0000)]
MFC 245610,245633,245635:
Include the thread name along with the command name when displaying the
command name of a thread from a multi-threaded process that doesn't have
an available argument list (such as kernel processes) and threads display
is enabled via -H.
jhb [Thu, 28 Feb 2013 21:57:38 +0000 (21:57 +0000)]
MFC 245508,245566,245568,245611,245909:
Various fixes to timestamps in NFS:
- Use the VA_UTIMES_NULL flag to detect when NULL was passed to utimes()
instead of comparing the desired time against the current time as a
heuristic.
- Remove unused nfs_curusec().
- Use vfs_timestamp() to set file timestamps rather than invoking
getmicrotime() or getnanotime() directly in NFS.
- Use NFSD_MONOSEC (which maps to time_uptime) instead of the seconds
portion of wall-time stamps to manage timeouts on events.
- Remove unused nd_starttime from the per-request structure in the new
NFS server.
- Use nanotime() for the modification time on a delegation to get as
precise a time as possible.
- Use time_second instead of extracting the second from a call to
getmicrotime().
jhb [Thu, 28 Feb 2013 21:24:10 +0000 (21:24 +0000)]
MFC 245238:
Add an option to not drop options from the third retransmitted SYN. If the
SYNs (or SYN/ACK replies) are dropped due to network congestion, then the
remote end of the connection may act as if options such as window scaling
are enabled but the local end will think they are not. This can result in
very slow data transfers in the case of window scaling disagreements.
Note that the unlike HEAD the existing behavior is preserved by default,
but it can be disabled by setting the
net.inet.tcp.rexmit_drop_options sysctl to zero.
jhb [Thu, 28 Feb 2013 19:00:57 +0000 (19:00 +0000)]
MFC 245476:
- More properly handle interrupted NFS requests on an interruptible mount
by returning an error of EINTR rather than EACCES.
- While here, bring back some (but not all) of the NFS RPC statistics lost
when krpc was committed.
jhb [Thu, 28 Feb 2013 18:58:37 +0000 (18:58 +0000)]
MFC 245471:
Do not require a filter-only interrupt handler for puc ports that are not
serial devices (such as printer ports). This allows ppc devices attached
to puc to correctly setup an interrupt handler and work.
jhb [Thu, 28 Feb 2013 18:49:31 +0000 (18:49 +0000)]
MFC 241748:
When checking to see if a video output's _ADR matches an entry in the
parent adapter's _DOD list, only check the low 16 bits of both _ADR and
_DOD. The language in the ACPI spec seems to indicate that the _ADR values
should exactly match the entries in _DOD. However, I assume that the
masking added to _DOD values was added to work around some known busted
machines (the commit history doesn't indicate either way), and the ACPI
spec does require that the low 16 bits are unique for all video outputs,
so only checking the low 16 bits should be fine.
This fixes recognition of video outputs that use the new standardized
device ID scheme in ACPI 3.0 that set bit 31 such as certain Dell laptops.
jhb [Thu, 28 Feb 2013 18:22:41 +0000 (18:22 +0000)]
MFC 241027:
- Re-shuffle the <machine/pc/bios.h> headers to move all kernel-specific
bits under #ifdef _KERNEL but leave definitions for various structures
defined by standards ($PIR table, SMAP entries, etc.) available to
userland.
- Consolidate duplicate SMBIOS table structure definitions in ipmi(4)
and smbios(4) in <machine/pc/bios.h> and make them available to
userland.
jhb [Thu, 28 Feb 2013 18:10:20 +0000 (18:10 +0000)]
MFC 240968:
Merge similar fixes from 223198 from igb to ixgbe:
- Use a dedicated task to handle deferred transmits from the if_transmit
method instead of reusing the existing per-queue interrupt task.
Reusing the per-queue interrupt task could result in both an interrupt
thread and the taskqueue thread trying to handle received packets on a
single queue resulting in out-of-order packet processing and lock
contention.
- Don't define ixgbe_start() at all when if_transmit is used.
gallatin [Thu, 28 Feb 2013 16:38:28 +0000 (16:38 +0000)]
MFC r247133 and associated fixes (r247151, r247152):
r247133:
Improve mxge's receive performance for IPv6:
- Add support for IPv6 rx csum offload
- Finally switch mxge from using its own driver lro, to
using tcp_lro
r247151:
Fix build.
r247152:
Try harder to make mxge safe for all combinations of INET and INET6
- Re-fix build by restoring local removed in r247151, but protected
by #if defined(INET) || defined(INET6) so that the compile
succeeds in the !(INET||INET6) case.
- Protect call to in_pseudo() with an #ifdef INET, to allow
a kernel to link with mxge when INET is not compiled in.
- Also remove an errant (improperly commented) obsolete debugging printf
gallatin [Thu, 28 Feb 2013 16:32:36 +0000 (16:32 +0000)]
MFC r247104:
Fix tcp_lro_rx_ipv4() for drivers that do not set CSUM_IP_CHECKED.
Specifcially, in_cksum_hdr() returns 0 (not 0xffff) when the IPv4
checksum is correct. Without this fix, the tcp_lro code will reject
good IPv4 traffic from drivers that do not implement IPv4 header
harder csum offload.
Note that TCP_OFFLOAD is not enabled in 9 yet and so some of these MFCs
don't really affect functionality. But they do help future MFCs
(related to TCP_OFFLOAD or not) by minimizing diffs with the driver in
head.
r245243:
cxgbe(4): updates to the configuration file that controls how hardware
resources are partitioned.
- Reduce the number of virtual interfaces reserved for PF4. This leaves
spare room in the source MAC table and allows the driver to setup
filters that rewrite the source MAC address.
- Reduce the number of filters and use the freed up space for the CLIP
(Compressed Local IPv6 addresses) table. This is a prerequisite for
IPv6 TOE support which will follow separately in a series of commits.
r245274:
cxgbe(4): Add functions to help synchronize "slow" operations (those not
on the fast data path) and use them instead of frobbing the adapter lock
and busy flag directly.
Other changes made while reworking all slow operations:
- Wait for the reply to a filter request (add/delete). This guarantees
that the operation is complete by the time the ioctl returns.
- Tidy up the tid_info structure.
- Do not allow the tx queue size to be set to something that's not a
power of 2.
r245276:
Overhaul the stid allocator so that it can be used for IPv6 servers
too. The entry for an IPv6 server in the TCAM takes up the equivalent
of two ordinary stids and must be properly aligned too.
r245434:
cxgbe(4): Updates to the hardware L2 table management code.
- Add full support for IPv6 addresses.
- Read the size of the L2 table during attach. Do not assume that PCIe
physical function 4 of the card has all of the table to itself.
- Use FNV instead of Jenkins to hash L3 addresses and drop the private
copy of jhash.h from the driver.
r245441:
cxgbe/tom: Miscellaneous updates for TOE+IPv6 support (more to follow).
- Teach find_best_mtu_idx() to deal with IPv6 endpoints.
- Install correct protosw in offloaded TCP/IPv6 sockets when DDP is
enabled.
- Move set_tcp_ddp_ulp_mode to t4_tom.c so that t4_tom.h can be included
without having to drag in t4_msg.h too. This was bothering the iWARP
driver for some reason.
r245448:
cxgbe/tom: Basic CLIP table management.
This is the Compressed Local IPv6 table on the chip. To save space, the
chip uses an index into this table instead of a full IPv6 address in
some of its hardware data structures.
For now the driver fills this table with all the local IPv6 addresses
that it sees at the time the table is initialized. I'll improve this
later so that the table is updated whenever new IPv6 addresses are
configured or existing ones deleted.
r245467:
cxgbe/tom: Add support for fully offloaded TCP/IPv6 connections (active open).
r245468:
cxgbe/tom: Add support for fully offloaded TCP/IPv6 connections (passive open).
r245517:
cxgbe: Fix the for_each_foo macros -- the last argument should not share
its name with any member of struct sge.
r245518:
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl.
r245520:
Allow "ivlan" (inner VLAN) to be used as an alias for "vlan" when
specifying match criteria. "vlan" continues to be valid here, and it
continues to be valid when deleting, rewriting, inserting, or stacking
an 802.1q tag to a matching packet.
r245567:
cxgbe: Make the for_each macros safer to use by turning them
into a single statement each.
r245933:
cxgbe/tom: List IFCAP_TOE6 as supported now that all the required pieces
are in place. You still have to enable it explicitly, after loading the
t4_tom KLD.
r245935:
Add a couple of missing error codes. Treat CPL_ERR_KEEPALV_NEG_ADVICE as
negative advice and not a fatal error.
r245936:
Force the 404-BT card (4 x 1G) to use the "uwire" configuration file.
r245937:
Install an extra hold on the newly allocated synq entry so that it
cannot be freed while do_pass_accept_req is running. This closes a race
where do_pass_establish on another CPU (the driver chose a different
queue for the new tid) expands the synq entry into a full PCB and then
releases the only hold on it, all while do_pass_accept_req is still
running.
r246093:
Provide a statistic to track the number of drops in each of the port's
txq's buf_ring. The aggregate for all the queues of a port is already
provided in ifnet->if_snd.ifq_drops.
r246385:
Busy-wait when cold.
r246575:
Do not hold locks around hardware context reads.
r247062:
cxgbe(4): Assume that CSUM_TSO in the transmit path implies CSUM_IP and
CSUM_TCP too. They are all set explicitly by the kernel usually.
r247122:
cxgbe(4): Add sysctls to extract debug information from the chip:
r247291:
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating
them in a firmware message if it is able to do so. This works out
better for one of the FIFOs in the chip.
r247347:
cxgbe(4): Consider all the API versions of the interfaces exported by
the firmware (instead of just the main firmware version) when evaluating
firmware compatibility. Document the new "hw.cxgbe.fw_install" knob
being introduced here.
This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will
mostly do what was requested in the PR but it's a bit more intelligent
in that it won't reinstall the same firmware repeatedly if the knob is
left set.
r247355:
cxgbe(4): Report unusual out of band errors from the firmware.
r241733 (by ed@):
Prefer __containerof() over __member2struct().
The former works better with qualifiers, but also properly type checks
the input pointer.
jhb [Wed, 27 Feb 2013 22:08:18 +0000 (22:08 +0000)]
MFC 241130:
Rename the module for 'device enc' to "if_enc" to avoid conflicting with
the CAM "enc" peripheral (part of ses(4)). Previously the two modules
used the same name, so only one was included in a linked kernel causing
enc0 to not be created if you added IPSEC to GENERIC. The new module
name follows the pattern of other network interfaces (e.g. "if_loop").
jhb [Wed, 27 Feb 2013 22:02:40 +0000 (22:02 +0000)]
MFC 240921:
Add optional entropy harvesting for software interrupts in swi_sched()
as controlled by kern.random.sys.harvest.swi. SWI harvesting feeds into
the interrupt FIFO and each event is estimated as providing a single bit of
entropy.
jhb [Wed, 27 Feb 2013 21:03:09 +0000 (21:03 +0000)]
MFC 240692,241228:
Adjust the ioctl workaround from r234501:
- Ensure the native ioctl path always allocates a 4kb buffer if a request
uses a buffer size of 0.
- Rounding all small requests up to 32k swamped the controller causing
firmware hangs. Instead, round requests smaller than 64k up to the next
power of 2 as a general rule. To handle the one known special case of a
command that accepts a 12k buffer returning a 24k-ish reply, round
requests between 8k and 16k up to 32k rather than 16k. The result is
that commands less than 8k should now be rounded up to a smaller size
(either 4k or 8k) rather than 32k.
jhb [Wed, 27 Feb 2013 19:38:25 +0000 (19:38 +0000)]
MFC 240474:
- Denote PCI-e endpoints that support FLR.
- Make parsing of PCI-e extended capabilities assume that future version
numbers are backwards compatible.
- Add new AER error descriptions.
- Add descriptions for more PCI-e extended capabilities.
jhb [Tue, 26 Feb 2013 21:04:58 +0000 (21:04 +0000)]
MFC 240467:
Ignore stop and continue signals sent to an exiting process. Stop signals
set p_xstat to the signal that triggered the stop, but p_xstat is also
used to hold the exit status of an exiting process. Without this change,
a stop signal that arrived after a process was marked P_WEXIT but before
it was marked a zombie would overwrite the exit status with the stop signal
number.
jhb [Tue, 26 Feb 2013 20:19:19 +0000 (20:19 +0000)]
MFC 239512:
Add a BUS_CHILD_DELETED() method that a bus can hook to allow it to cleanup
any bus-specific state (such as ivars) when a child device is deleted.
jhb [Tue, 26 Feb 2013 19:14:05 +0000 (19:14 +0000)]
MFC 239356,239564:
Teach dhclient to track changes in link state and to enter the reboot
state when the link on an interface goes up causing dhclient to attempt
to renew its existing lease.
dim [Tue, 26 Feb 2013 18:31:03 +0000 (18:31 +0000)]
MFC r246880:
Since clang 3.2 now has an option to suppress warnings about implicitly
promoted K&R parameters, remove the workarounds added for sendmail
components in r228558.
jhb [Tue, 26 Feb 2013 16:55:44 +0000 (16:55 +0000)]
MFC 234927:
- Don't log messages saying that accounting is being disabled and enabled
if the accounting log file is atomically replaced with a new file
(such as during log rotation).
- Simplify accounting log rotation a bit. There is no need to re-run
accton(8) after renaming the new log file to it's real name.
jilles [Mon, 25 Feb 2013 18:07:20 +0000 (18:07 +0000)]
MFC r241010: libc/fts: Use O_CLOEXEC for internal file descriptors.
Because fts keeps internal file descriptors open across calls, making such
descriptors close-on-exec helps not only multi-threaded applications but
also single-threaded applications.
In particular, this prevents passing a temporary file descriptor for saving
the current directory to processes created via find -exec.
dteske [Mon, 25 Feb 2013 16:13:21 +0000 (16:13 +0000)]
Oops, r55392 (over 13 years ago) forgot to classify the new HTTP Proxy for the
options dialog (causing "<unknown>" to be displayed for the HTTP media type).
Direct commit to stable/9 (no sysinstall in head).
pfg [Sun, 24 Feb 2013 02:47:19 +0000 (02:47 +0000)]
MFC r246563, r246564, r246634:
ext2fs: Miscellaneous cleanups and fixes.
make e2fs_maxcontig local and remove tautological check.
Replace redundant EXT2_MIN_BLOCK with EXT2_MIN_BLOCK_SIZE.
Use prototype declarations for function definitions
jilles [Fri, 22 Feb 2013 20:16:16 +0000 (20:16 +0000)]
MFC r246522: sh: Simplify mksyntax and make it fit for cross-compiling.
Now it outputs fixed files, which use constants provided by the C standard
library to determine appropriate values for the target machine.
Before, mksyntax inspected the host machine which resulted in subtle
breakage if e.g. char is signed on the host and unsigned on the target such
as when cross-compiling on x86 for ARM.
dim [Fri, 22 Feb 2013 18:35:40 +0000 (18:35 +0000)]
MFC r247003:
Pull in r175360 from upstream llvm trunk:
MCParser: Reject .balign with non-pow2 alignments.
GNU as rejects them and there are configure scripts in the wild that
check if the assembler rejects ".align 3" to determine whether the
alignment is in bytes or powers of two.
dim [Fri, 22 Feb 2013 18:33:42 +0000 (18:33 +0000)]
MFC r246858:
Pull in r175057 from upstream llvm trunk:
X86: Disable generation of rep;movsl when %esi is used as a base pointer.
This happens when there is both stack realignment and a dynamic alloca in the
function. If we overwrite %esi (rep;movsl uses fixed registers) we'll lose the
base pointer and the next register spill will write into oblivion.
Fixes PR15249 and unbreaks firefox on i386/freebsd. Mozilla uses dynamic allocas
and freebsd a 4 byte stack alignment.
mav [Fri, 22 Feb 2013 09:47:21 +0000 (09:47 +0000)]
MFC r242852, r243069:
Several optimizations to sched_idletd():
- Do not try to steal load from other CPUs if there was no context switches
on this CPU (i.e. it was idle all the time and woke up just for bus mastering
or TLB shutdown). If current CPU was idle, then it is quite unlikely that some
other CPU has load to steal. Under high I/O rate, when TLB shutdowns cause
numerous CPU wakeups, on 24-CPU system load stealing code may consume up to
25% of all CPU time without giving any benefits.
- Change code that implements spinning for load to restart spin in case of
context switch. Previous code periodically called cpu_idle() even under
high interrupt/context switch rate.
- Rise spinning threshold to 10KHz, where it gives at least some effect
that may worth consumed power.
mav [Thu, 21 Feb 2013 19:02:29 +0000 (19:02 +0000)]
MFC r244014 (by ken):
Fix a device departure bug for the the pass(4), enc(4), sg(4) and ch(4)
drivers.
The bug occurrs when a userland process has the driver instance
open and the underlying device goes away. We get the devfs
callback that the device node has been destroyed, but not all of
the closes necessary to fully decrement the reference count on the
CAM peripheral.
The reason is that once devfs calls back and says the device has
been destroyed, it is moved off to deadfs, and devfs guarantees
that there will be no more open or close calls. So the solution
is to keep track of how many outstanding open calls there are on
the device, and just release that many references when we get the
callback from devfs.
scsi_pass.c,
scsi_enc.c,
scsi_enc_internal.h: Add an open count to the softc in these
drivers. Increment it on open and
decrement it on close.
When we get a devfs callback to say that
the device node has gone away, decrement
the peripheral reference count by the
number of still outstanding opens.
Make sure we don't access the peripheral
with cam_periph_unlock() after what might
be the final call to
cam_periph_release_locked(). The
peripheral might have been freed, and we
will be dereferencing freed memory.
scsi_ch.c,
scsi_sg.c: For the ch(4) and sg(4) drivers, add the
same changes described above, and in
addition, fix another bug that was
previously fixed in the pass(4) and enc(4)
drivers.
These drivers were calling destroy_dev()
from their cleanup routine, but that could
cause a deadlock because the cleanup
routine could be indirectly called from
the driver's close routine. This would
cause a deadlock, because the device node
is being held open by the active close
call, and can't be destroyed.
mav [Thu, 21 Feb 2013 18:56:09 +0000 (18:56 +0000)]
MFC r237328 (by ken) for recently merged scsi_enc.c:
Fix several reference counting and object lifetime issues between
the pass(4) and enc(4) drivers and devfs.
The pass(4) driver uses the destroy_dev_sched() routine to
schedule its device node for destruction in a separate thread
context. It does this because the passcleanup() routine can get
called indirectly from the passclose() routine, and that would
cause a deadlock if the close routine tried to destroy its own
device node.
In any case, once a particular passthrough driver number, e.g.
pass3, is destroyed, CAM considers that unit number (3 in this
case) available for reuse.
The problem is that devfs may not be done cleaning up the previous
instance of pass3, and will panic if isn't done cleaning up the
previous instance.
The solution is to get a callback from devfs when the device node
is removed, and make sure we hold a reference to the peripheral
until that happens.
Testing exposed some other cases where we have reference counting
issues, and those were also fixed in the pass(4) driver.
cam_periph.c: In camperiphfree(), reorder some of the operations.
The peripheral destructor needs to be called before
the peripheral is removed from the peripheral is
removed from the list. This is because once we
remove the peripheral from the list, and drop the
topology lock, the peripheral number may be reused.
But if the destructor hasn't been called yet, there
may still be resources hanging around (like devfs
nodes) that haven't been fully cleaned up.
cam_xpt.c: Add an argument to xpt_remove_periph() to indicate
whether the topology lock is already held.
scsi_enc.c: Acquire an extra reference to the peripheral during
registration, and release it once we get a callback
from devfs indicating that the device node is gone.
Call destroy_dev_sched_cb() in enc_oninvalidate()
instead of calling destroy_dev() in the cleanup
routine.
scsi_pass.c: Add reference counting to handle peripheral and
devfs object lifetime issues.
Add a reference to the peripheral and the devfs
node in the peripheral registration.
Don't attempt to add a physical path alias if the
peripheral has been marked invalid.
Release the devfs reference once the initial
physical path alias taskqueue run has completed.
Schedule devfs node destruction in the
passoninvalidate(), and release our peripheral
reference in a new routine, passdevgonecb() once
the devfs node is gone. This allows the peripheral
to fully go away, and the peripheral destructor,
passcleanup(), will get called.
mav [Thu, 21 Feb 2013 18:49:05 +0000 (18:49 +0000)]
MFC r236138 (by ken) for recently merged scsi_enc.c:
Work around a race condition in devfs by changing the way closes
are handled in most CAM peripheral drivers that are not handled by
GEOM's disk class.
The usual character driver open and close semantics are that the
driver gets N open calls, but only one close, when the last caller
closes the device.
CAM peripheral drivers expect that behavior to be honored to the
letter, and the CAM peripheral driver code (specifically
cam_periph_release_locked_busses()) panics if it is done incorrectly.
Since devfs has to drop its locks while it calls a driver's close
routine, and it does not have a way to delay or prevent open calls
while it is calling the close routine, there is a race.
The sequence of events, simplified a bit, is:
- devfs acquires a lock
- devfs checks the reference count, and if it is 1, continues to close.
- devfs releases the lock
- 2nd process open call on the device happens here
- devfs calls the driver's close routine
- devfs acquires a lock
- devfs decrements the reference count
- devfs releases the lock
- 2nd process close call on the device happens here
At the second close, we get a panic in
cam_periph_release_locked_busses(), complaining that peripheral
has been released when the reference count is already 0. This is
because we have gotten two closes in a row, which should not
happen.
The fix is to add the D_TRACKCLOSE flag to the driver's cdevsw, so
that we get a close() call for each open(). That does happen
reliably, so we can make sure that our reference counts are
correct.
Note that the sa(4) and pt(4) drivers only allow one context
through the open routine. So these drivers aren't exposed to the
same race condition.
scsi_ch.c,
scsi_enc.c,
scsi_enc_internal.h,
scsi_pass.c,
scsi_sg.c:
For these drivers, change the open() routine to
increment the reference count for every open, and
just decrement the reference count in the close.
Call cam_periph_release_locked() in some scenarios
to avoid additional lock and unlock calls.
scsi_pt.c: Call cam_periph_release_locked() in some scenarios
to avoid additional lock and unlock calls.
mav [Thu, 21 Feb 2013 16:59:28 +0000 (16:59 +0000)]
MFC r238379, r238382 (by bruefer):
Renamed the kern.cam.da.da_send_ordered sysctl and tunable to
kern.cam.da.send_ordered, more in line with the other da sysctls/tunables.
Renamed the kern.cam.ada.ada_send_ordered sysctl and tunable to
kern.cam.ada.send_ordered, more in line with the other da sysctls/tunables.
hselasky [Thu, 21 Feb 2013 07:48:07 +0000 (07:48 +0000)]
MFC r246616 and r246759:
- Move scratch data from the USB bus structure to the USB device
structure so that simultaneous access cannot happen. Protect scratch
area using the enumeration lock.
- Reduce stack usage in usbd_transfer_setup() by moving some big stack
members to the scratch area. This saves around 200 bytes of stack.
- Fix a whitespace.
- Protect control requests using the USB device enumeration lock.
- Make sure all callers of usbd_enum_lock() check the return value.
- Remove the control transfer specific lock.
- Bump the FreeBSD version number, hence external USB modules may need
to be recompiled due to a USB device structure change.
projects / FreeBSD / stable / 9.git / log