sxlock, via the sx_{s, x}lock_sig() interface, or plain lockmgr), will
leave the waiters flag on forcing the owner to do a wakeup even when if
the waiter queue is empty.
That operation may lead to a deadlock in the case of doing a fake wakeup
on the "preferred" (based on the wakeup algorithm) queue while the other
queue has real waiters on it, because nobody is going to wakeup the 2nd
queue waiters and they will sleep indefinitively.
A similar bug, is present, for lockmgr in the case the waiters are
sleeping with LK_SLEEPFAIL on. In this case, even if the waiters queue
is not empty, the waiters won't progress after being awake but they will
just fail, still not taking care of the 2nd queue waiters (as instead the
lock owned doing the wakeup would expect).
In order to fix this bug in a cheap way (without adding too much locking
and complicating too much the semantic) add a sleepqueue interface which
does report the actual number of waiters on a specified queue of a
waitchannel (sleepq_sleepcnt()) and use it in order to determine if the
exclusive waiters (or shared waiters) are actually present on the lockmgr
(or sx) before to give them precedence in the wakeup algorithm.
This fix alone, however doesn't solve the LK_SLEEPFAIL bug. In order to
cope with it, add the tracking of how many exclusive LK_SLEEPFAIL waiters
a lockmgr has and if all the waiters on the exclusive waiters queue are
LK_SLEEPFAIL just wake both queues.
The sleepq_sleepcnt() introduction and ABI breakage require
__FreeBSD_version bumping.
Reported by: avg, kib, pho
Reviewed by: kib
Tested by: pho
handlers. This is primarily intended as a way to allow devices that use
multiple interrupts (e.g. MSI) to meaningfully distinguish the various
interrupt handlers.
- Add a new BUS_DESCRIBE_INTR() method to the bus interface to associate
a description with an active interrupt handler setup by BUS_SETUP_INTR.
It has a default method (bus_generic_describe_intr()) which simply passes
the request up to the parent device.
- Add a bus_describe_intr() wrapper around BUS_DESCRIBE_INTR() that supports
printf(9) style formatting using var args.
- Reserve MAXCOMLEN bytes in the intr_handler structure to hold the name of
an interrupt handler and copy the name passed to intr_event_add_handler()
into that buffer instead of just saving the pointer to the name.
- Add a new intr_event_describe_handler() which appends a description string
to an interrupt handler's name.
- Implement support for interrupt descriptions on amd64 and i386 by having
the nexus(4) driver supply a custom bus_describe_intr method that invokes
a new intr_describe() MD routine which in turn looks up the associated
interrupt event and invokes intr_event_describe_handler().
Requested by: many
Reviewed by: scottl
MFC after: 2 weeks
the kthread_create(9) man page to the kproc(9) page as it had migrated and
people looking for it may need a hand to find its new name.
MFC after: 1 week
better semantics if a request to append an address range to an existing list
fails.
- When cloning an sglist, properly set the length in the new sglist instead of
leaving the new list empty.
- Properly compute the amount of data added to an sglist via
_sglist_append_buf(). This allows sglist_consume_uio() to properly update
uio_resid.
- When a request to append an address range to a scatter/gather list fails,
restore the sglist to the state it had at the start of the function call
instead of resetting it to an empty list.
Requested by: np (3)
Approved by: re (kib)
replace it with wrappers around our taskqueue(9).
To make it possible implement taskqueue_member() function which returns 1
if the given thread was created by the given taskqueue.
Approved by: re (kib)
things a bit:
- use dpcpu data to track the ifps with packets queued up,
- per-cpu locking and driver flags
- along with .nh_drainedcpu and NETISR_POLICY_CPU.
- Put the mbufs in flight reference count, preventing interfaces
from going away, under INVARIANTS as this is a general problem
of the stack and should be solved in if.c/netisr but still good
to verify the internal queuing logic.
- Permit changing the MTU to virtually everythinkg like we do for loopback.
Hook epair(4) up to the build.
Approved by: re (kib)
- update for getrlimit(2) manpage;
- support for setting RLIMIT_SWAP in login class;
- addition to the limits(1) and sh and csh limit-setting builtins;
- tuning(7) documentation on the sysctls controlling overcommit.
In collaboration with: pho
Reviewed by: alc
Approved by: re (kensmith)
Actually, as it did receive few tuning, the support is disabled by
default, but it can opt-in with the option ADAPTIVE_LOCKMGRS.
Due to the nature of lockmgrs, adaptive spinning needs to be
selectively enabled for any interested lockmgr.
The support is bi-directional, or, in other ways, it will work in both
cases if the lock is held in read or write way. In particular, the
read path is passible of further tunning using the sysctls
debug.lockmgr.retries and debug.lockmgr.loops . Ideally, such sysctls
should be axed or compiled out before release.
Addictionally note that adaptive spinning doesn't cope well with
LK_SLEEPFAIL. The reason is that many (and probabilly all) consumers
of LK_SLEEPFAIL are mainly interested in knowing if the interlock was
dropped or not in order to reacquire it and re-test initial conditions.
This directly interacts with adaptive spinning because lockmgr needs
to drop the interlock while spinning in order to avoid a deadlock
(further details in the comments inside the patch).
Final note: finding someone willing to help on tuning this with
relevant workloads would be either very important and appreciated.
Tested by: jeff, pho
Requested by: many
queue was drained. It will never fire for a directly dispatched packet.
You will most likely never want to use this for any ordinary netisr usage
and you will never blame netisr in case you try to use it and it does
not work as expected.
Reviewed by: rwatson
probe. The current device order is unchanged. This commit just adds the
infrastructure and ABI changes so that it is easier to merge later changes
into 8.x.
- Driver attachments now have an associated pass level. Attachments are
not allowed to probe or attach to drivers until the system-wide pass level
is >= the attachment's pass level. By default driver attachments use the
"last" pass level (BUS_PASS_DEFAULT). Driver's that wish to probe during
an earlier pass use EARLY_DRIVER_MODULE() instead of DRIVER_MODULE() which
accepts the pass level as an additional parameter.
- A new method BUS_NEW_PASS has been added to the bus interface. This
method is invoked when the system-wide pass level is changed to kick off
a rescan of the device tree so that drivers that have just been made
"eligible" can probe and attach.
- The bus_generic_new_pass() function provides a default implementation of
BUS_NEW_PASS(). It first allows drivers that were just made eligible for
this pass to identify new child devices. Then it propogates the rescan to
child devices that already have an attached driver by invoking their
BUS_NEW_PASS() method. It also reprobes devices without a driver.
- BUS_PROBE_NOMATCH() is only invoked for devices that do not have
an attached driver after being scanned during the final pass.
- The bus_set_pass() function is used during boot to raise the pass level.
Currently it is only called once during root_bus_configure() to raise
the pass level to BUS_PASS_DEFAULT. This has the effect of probing all
devices in a single pass identical to previous behavior.
Reviewed by: imp
Approved by: re (kib)
Each list describes a logical memory object that is backed by one or more
physical address ranges. To minimize locking, the sglist objects
themselves are immutable once they are shared.
These objects may be used in the future to facilitate I/O requests using
physically-addressed buffers. For the immediate future I plan to use them
to implement a new type of VM object and pager.
Reviewed by: jeff, scottl
MFC after: 1 month
permissions, such as VWRITE_ACL. For a filsystems that don't
implement it, there is a default implementation, which works
as a wrapper around VOP_ACCESS.
Reviewed by: rwatson@
- Add rm_init_flags() and accept extended options only for that variation.
- Add a flags space specifically for rm_init_flags(), rather than borrowing
the lock_init() flag space.
- Define flag RM_RECURSE to use instead of LO_RECURSABLE.
- Define flag RM_NOWITNESS to allow an rmlock to be exempt from WITNESS
checking; this wasn't possible previously as rm_init() always passed
LO_WITNESS when initializing an rmlock's struct lock.
- Add RM_SYSINIT_FLAGS().
- Rename embedded mutex in rmlocks to make it more obvious what it is.
- Update consumers.
- Update man page.
Introduce for this operation the reverse NO_ADAPTIVE_SX option.
The flag SX_ADAPTIVESPIN to be passed to sx_init_flags(9) gets suppressed
and the new flag, offering the reversed logic, SX_NOADAPTIVE is added.
Additively implements adaptive spininning for sx held in shared mode.
The spinning limit can be handled through sysctls in order to be tuned
while the code doesn't reach the release, after which time they should
be dropped probabilly.
This change has made been necessary by recent benchmarks where it does
improve concurrency of workloads in presence of high contention
(ie. ZFS).
KPI breakage is documented by __FreeBSD_version bumping, manpage and
UPDATING updates.
Requested by: jeff, kmacy
Reviewed by: jeff
Tested by: pho
Add support for kernel fault injection using KFAIL_POINT_* macros and
fail_point_* infrastructure. Add example fail point in vfs_bio.c to
simulate VM buf pressure.
Approved by: dfr (mentor)
