- There is no globally visible scheduler lock any longer. For now the
watchdog can only check Giant. This model of checking particular locks
is flawed and should be revisited. Other metrics should be considered.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Use sched_throw() rather than replicating the same cpu_throw() code for
each architecture. This also allows the scheduler to use any locking it
may want to.
- Use the thread_lock() rather than sched_lock when preempting.
- The scheduler lock is not required to synchronize release_aps.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Add new spinlocks to support thread_lock() and adjust ordering.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Attempt to return the ttyinfo() selection algorithm to something sane
as it has been broken and disabled for some time. Adapt this algorithm
in such a way that it does not conflict with per-cpu scheduler locking.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Use a global umtx spinlock to protect the sleep queues now that there
is no global scheduler lock.
- Use thread_lock() to protect thread state.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Use thread_lock() rather than sched_lock for per-thread scheduling
sychronization.
- Use the per-process spinlock rather than the sched_lock for per-process
scheduling synchronization.
- Use a global kse spinlock to protect upcall and thread assignment. The
per-process spinlock can not be used because this lock must be acquired
via mi_switch() where we already hold a thread lock. The kse spinlock
is a leaf lock ordered after the process and thread spinlocks.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Use thread_lock() rather than sched_lock for per-thread scheduling
sychronization.
- Use the per-process spinlock rather than the sched_lock for per-process
scheduling synchronization.
- Replace the tail-end of fork_exit() with a scheduler specific routine
which can do the appropriate lock manipulations.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Protect the cp_time tick counts with atomics instead of a global lock.
There will only be one atomic per tick and this allows all processors
to execute softclock concurrently.
- In softclock, protect access to rusage and td_*tick data with the
thread_lock(), expanding the scope of the thread lock over the whole
function.
- Do some creative re-arranging in hardclock() to avoid excess locking.
- Protect the p_timer fields with the per-process spinlock.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Use thread_lock() rather than sched_lock for per-thread scheduling
sychronization.
- Use the per-process spinlock rather than the sched_lock for per-process
scheduling synchronization.
- Move some common code into thread_suspend_switch() to handle the
mechanics of suspending a thread. The locking here is incredibly
convoluted and should be simplified.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Add a per-turnstile spinlock to solve potential priority propagation
deadlocks that are possible with thread_lock().
- The turnstile lock order is defined as the exact opposite of the
lock order used with the sleep locks they represent. This allows us
to walk in reverse order in priority_propagate and this is the only
place we wish to multiply acquire turnstile locks.
- Use the turnstile_chain lock to protect assigning mutexes to turnstiles.
- Change the turnstile interface to pass back turnstile pointers to the
consumers. This allows us to reduce some locking and makes it easier
to cancel turnstile assignment while the turnstile chain lock is held.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Adapt sleepqueues to the new thread_lock() mechanism.
- Delay assigning the sleep queue spinlock as the thread lock until after
we've checked for signals. It is illegal for a thread to return in
mi_switch() with any lock assigned to td_lock other than the scheduler
locks.
- Change sleepq_catch_signals() to do the switch if necessary to simplify
the callers.
- Simplify timeout handling now that locking a sleeping thread has the
side-effect of locking the sleepqueue. Some previous races are no
longer possible.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
- Move all scheduler locking into the schedulers utilizing a technique
similar to solaris's container locking.
- A per-process spinlock is now used to protect the queue of threads,
thread count, suspension count, p_sflags, and other process
related scheduling fields.
- The new thread lock is actually a pointer to a spinlock for the
container that the thread is currently owned by. The container may
be a turnstile, sleepqueue, or run queue.
- thread_lock() is now used to protect access to thread related scheduling
fields. thread_unlock() unlocks the lock and thread_set_lock()
implements the transition from one lock to another.
- A new "blocked_lock" is used in cases where it is not safe to hold the
actual thread's lock yet we must prevent access to the thread.
- sched_throw() and sched_fork_exit() are introduced to allow the
schedulers to fix-up locking at these points.
- Add some minor infrastructure for optionally exporting scheduler
statistics that were invaluable in solving performance problems with
this patch. Generally these statistics allow you to differentiate
between different causes of context switches.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
Now, we assume no more sched_lock protection for some of them and use the
distribuited loads method for vmmeter (distribuited through CPUs).
Reviewed by: alc, bde
Approved by: jeff (mentor)
- Rename PCPU_LAZY_INC into PCPU_INC
- Add the PCPU_ADD interface which just does an add on the pcpu member
given a specific value.
Note that for most architectures PCPU_INC and PCPU_ADD are not safe.
This is a point that needs some discussions/work in the next days.
Reviewed by: alc, bde
Approved by: jeff (mentor)
sysctl_handle_int is not sizeof the int type you want to export.
The type must always be an int or an unsigned int.
Remove the instances where a sizeof(variable) is passed to stop
people accidently cut and pasting these examples.
In a few places this was sysctl_handle_int was being used on 64 bit
types, which would truncate the value to be exported. In these
cases use sysctl_handle_quad to export them and change the format
to Q so that sysctl(1) can still print them.
- In the ioctl path let command get queued up and return
when complete _without_ blocking the driving waiting for
the response. This way the driver doesn't "lock up" for
~30s during a flash command. Submitted by scottl.
- Add a guard so that if a DCMD of 0 is sent down the ioctl
path don't send it to the controller. Return with a
status of OK. This is a little strange since MegaCli
doesn't seem to like something and will issue some DCMD
of 0. This doesn't happen under Linux. So the emulation
needs to be improved but I'm not sure what. Another strange
thing is that when a DCMD of 0 gets issued under i386 the
controller returns OK but in amd64 the context is messed
up.
- Add a guard so the context has to be with-in the legal
limit so we get a reasonable error assertion versus random
panic.
It's going to be a challenge to figure out why MegaCli is not totally
happy and then sends some bogus commands. This means that flashing
firmware via the Linux tool won't work since it generates a DCMD of
0 when it should be opening the firmware for a flash update. Without
this problem flashing works fine. This means there is no publicly
available tool to upgrade the RAID firmware under FreeBSD right now.
I plan to MFC all of the mfi changes to 6.X shortly. This might not
include the SCSI pass-through changes.
Submitted by: scottl
Reviewed by: scottl
MFC after: 3 days
1. Pass locking flags to VFS_ROOT().
2. Check v_mountedhere while the vnode is locked.
3. Always return locked vnode on success.
Change 1 fixes problem reported by Stephen M. Rumble - after
zfs_vfsops.c,1.9 change, zfs_root() no longer locks the vnode
unconditionally and traverse() didn't pass right lock type to
VFS_ROOT(). The result was that kernel paniced when .zfs/ directory
was accessed via NFS.
playtone() so that it uses times of 1/100ths of a second.
Now 'time echo T60ABC >/dev/speaker' takes ~3 seconds.
MFC after: 2 weeks
Problem noted by: dwmalone
- removed unused structure members
- fixed a minor bug that the ECN code point may not be restored correctly
Approved by: ume (mentor)
MFC after: 1 week
GEMs is unable to discriminate UDP from TCP packets such that
it can generate 0x0000 checksum value for the UDP datagram. So the
UDP checksum offload was disabled by default. You can enable it
by setting link0 flag with ifconfig(8).
o bus_dma(9) clean up. It now correctly set number of required DMA
segments/size and removed incorrect use of BUS_DMA_ALLOCNOW flag
in static allocations done via bus_dmamem_alloc(9).
o Implemented ALTQ(9) support.
o Implemented Tx side bus_dmamap_load_mbuf_sg(9) which can remove
several book keeping chores orginated from call-back mechanism.
Therefore gem_txdma_callback() was removed and its functionality
was reimplemented in gem_load_txmbuf().
o Don't set GEM_TD_START_OF_PACKET flag until all remaining mbuf
chains are set. I think it was a long standing bug and it caused
fluctuating interrupts/CPU usage patterns while netperf test
is in progress. Previously it seems that we race with the device.
Because I don't have a documentation for GEM I'm not sure this is
correct but almost all other documentations I have stated this
implications on setting SOP mark in descriptor ring(e.g. hme(4)).
o Borrowed gem_defrag() from ath(4) which is supposed to be much
faster than m_defrag(9) since it's not need to defrag all
mbuf chains.
o gem_load_txmbuf() was changed to allow passed mbuf chains to free.
Caller of gem_load_txmbuf() correctly handles freed mbuf chains.
o In gem_start_locked(), added checks for availability of Tx
descriptors before trying to load DMA maps which could save CPU
cycles when number of available descriptors are low. Also, simplyfy
IFF_DRV_OACTIVE detection logic.
o Removed hard-coded function names in CTR macros and replaced it
with __func__.
o Moved statistics counter register access to gem_tick() to reduce
number of PCI bus accesses. There is no reason to update statistics
counters in interrupt handler.
o Removed unnecessary call of gem_start_locked() in gem_ioctl().
Reviewed by: grehan (initial version), marius (with improvements and suggestions)
Tested by: grehan (ppc), marius(sparc64)
own entry in the softc. This should allow more of cbb_pci_intr() to
migrate to a new cbb_pci_filt() so that we don't have to run cbb's ISR
in almost every case we get an interrupt. We can't just move
cbb_pci_intr into cbb_pci_filt because it does things that aren't safe
to do from a fast interrupt handler, err I mean from a filter. This is
an important first step.
# I wonder if I need to make cardok volatile or not.
mpt.h:
Add support for reading extended configuration pages.
mpt_cam.c:
Do a top level topology scan on the SAS controller. If any SATA
device are discovered in this scan, send a passthrough FIS to set
the write cache. This is controllable through the following
tunable at boot:
hw.mpt.enable_sata_wc:
-1 = Do not configure, use the controller default
0 = Disable the write cache
1 = Enable the write cache
The default is -1. This tunable is just a hack and may be
deprecated in the future.
Turning on the write cache alleviates the write performance problems with
SATA that many people have observed. It is not recommend for those who
value data reliability! I cannot stress this strongly enough. However,
it is useful in certain circumstances, and it brings the performence in line
with what a generic SATA controller running under the FreeBSD ATA driver
provides (and the ATA driver has had the WC enabled by default for years).
Things can get ugly without it due to uninitialized class. RELENG_6 need
a simmilar, but different treatment as well.
err.. perhaps we should teach devclass_get_maxunit() to return -1 ?
MFC after: 1 day
o If we don't have a filter, also check to make sure the card is there before
calling the scheduled ISR. This is necessary to help old drivers whose
ISRs can't cope with being called with the hardware missing, which sadly
still exist in the tree. This is the main reason why we have an extra
layer of indirection for cardbus interrupts.
o If the card is no longer present, mark the interrupt as 'handled' rather
than 'stray' because this accounts for why the interrupt happened. Stray
isn't all bad, since there are other filters that would claim it...
o Fix some comments
+ Add comment about why we check for CARD_OK and touch the hardware in both
the filter and ISR.
+ add a note about why we don't care about Giant
+ also note that giant can't be taken out in a filter...
+ Some minor formatting nits on very long comments.
While in the suspend path, this means the idle thread will just return
immediately rather than trying to enter C1-n. This helps in the case where
the chipset is powered down before the rest of the system and reads from
the cpu sleep registers begin returning immediately, causing the logic that
catches bad C2/C3 behavior to kick in. Observed on my Panasonic Y4.
MFC after: 3 days
1.50 to help out with the GCC 2 to GCC 3 transition and it became
obsolete when C flags compatible with GCC 3.x became the default.
With GCC 4 in the tree this variable (i.e. GCC3) is beyond bogus
because it causes confusion when looking for the newly introduced
WITH_GCC3 option that helps the GCC 3 -> GCC 4 bump.
(j/i) was being used and it was being incremented, not decremented as before.
Factor out this code into a common function and call it from both the common
and per-CPU case.
MFC after: 1 day
The global lock is a memory region shared with the BIOS and thus
has some strange behavior like the fact that the sleep is 1 ms max.
We use standard mutexes to synchronize with the SCI so acquiring
the global lock after locking the mutex resulted in a witness
warning.
To deal with this for now, acquire the global lock before all other
locks, similar to Giant. This should fix the witness "sleeping
with mutex held" issue on boot that occurred after the last ACPI-CA
import. In the future, we hope to move to the new mutex interface
in ACPI-CA instead of the pseudo-semaphore version we have now.
Reviewed by: jkim
