happen in interrupt context; 1) sleep locks, and 2) malloc/free
calls.
1) is fixed by using spin locks instead.
2) is fixed by preallocating a FIFO (implemented with a STAILQ)
and using elements from this FIFO instead. This turns out
to be rather fast.
OK'ed by: re (scottl)
Thanks to: peter, jhb, rwatson, jake
Apologies to: *
- This is heavily derived from John Baldwin's apic/pci cleanup on i386.
- I have completely rewritten or drastically cleaned up some other parts.
(in particular, bootstrap)
- This is still a WIP. It seems that there are some highly bogus bioses
on nVidia nForce3-150 boards. I can't stress how broken these boards
are. I have a workaround in mind, but right now the Asus SK8N is broken.
The Gigabyte K8NPro (nVidia based) is also mind-numbingly hosed.
- Most of my testing has been with SCHED_ULE. SCHED_4BSD works.
- the apic and acpi components are 'standard'.
- If you have an nVidia nForce3-150 board, you are stuck with 'device
atpic' in addition, because they somehow managed to forget to connect the
8254 timer to the apic, even though its in the same silicon! ARGH!
This directly violates the ACPI spec.
turnstiles to implement blocking isntead of implementing a thread queue
directly. These turnstiles are somewhat similar to those used in Solaris 7
as described in Solaris Internals but are also different.
Turnstiles do not come out of a fixed-sized pool. Rather, each thread is
assigned a turnstile when it is created that it frees when it is destroyed.
When a thread blocks on a lock, it donates its turnstile to that lock to
serve as queue of blocked threads. The queue associated with a given lock
is found by a lookup in a simple hash table. The turnstile itself is
protected by a lock associated with its entry in the hash table. This
means that sched_lock is no longer needed to contest on a mutex. Instead,
sched_lock is only used when manipulating run queues or thread priorities.
Turnstiles also implement priority propagation inherently.
Currently turnstiles only support mutexes. Eventually, however, turnstiles
may grow two queue's to support a non-sleepable reader/writer lock
implementation. For more details, see the comments in sys/turnstile.h and
kern/subr_turnstile.c.
The two primary advantages from the turnstile code include: 1) the size
of struct mutex shrinks by four pointers as it no longer stores the
thread queue linkages directly, and 2) less contention on sched_lock in
SMP systems including the ability for multiple CPUs to contend on different
locks simultaneously (not that this last detail is necessarily that much of
a big win). Note that 1) means that this commit is a kernel ABI breaker,
so don't mix old modules with a new kernel and vice versa.
Tested on: i386 SMP, sparc64 SMP, alpha SMP
another thread. We use the td_oncpu member of the other field to locate
it's associated CPU and then search the that CPU's list of spin locks
contained in its per-CPU data. This is not always safe and may in fact
panic or just not work, but it is useful in at least one case.
by allprison_mtx), a unique prison/jail identifier field, two path
fields (pr_path for reporting and pr_root vnode instance) to store
the chroot() point of each jail.
o Add jail_attach(2) to allow a process to bind to an existing jail.
o Add change_root() to perform the chroot operation on a specified
vnode.
o Generalize change_dir() to accept a vnode, and move namei() calls
to callers of change_dir().
o Add a new sysctl (security.jail.list) which is a group of
struct xprison instances that represent a snapshot of active jails.
Reviewed by: rwatson, tjr
as it could be and can do with some more cleanup. Currently its under
options LAZY_SWITCH. What this does is avoid %cr3 reloads for short
context switches that do not involve another user process. ie: we can
take an interrupt, switch to a kthread and return to the user without
explicitly flushing the tlb. However, this isn't as exciting as it could
be, the interrupt overhead is still high and too much blocks on Giant
still. There are some debug sysctls, for stats and for an on/off switch.
The main problem with doing this has been "what if the process that you're
running on exits while we're borrowing its address space?" - in this case
we use an IPI to give it a kick when we're about to reclaim the pmap.
Its not compiled in unless you add the LAZY_SWITCH option. I want to fix a
few more things and get some more feedback before turning it on by default.
This is NOT a replacement for Bosko's lazy interrupt stuff. This was more
meant for the kthread case, while his was for interrupts. Mine helps a
little for interrupts, but his helps a lot more.
The stats are enabled with options SWTCH_OPTIM_STATS - this has been a
pseudo-option for years, I just added a bunch of stuff to it.
One non-trivial change was to select a new thread before calling
cpu_switch() in the first place. This allows us to catch the silly
case of doing a cpu_switch() to the current process. This happens
uncomfortably often. This simplifies a bit of the asm code in cpu_switch
(no longer have to call choosethread() in the middle). This has been
implemented on i386 and (thanks to jake) sparc64. The others will come
soon. This is actually seperate to the lazy switch stuff.
Glanced at by: jake, jhb
is set to 0, it now has the same affect as setting witness_dead used to
have.
- Added a sysctl handler that allows root to change witness_watch from a
non-zero value to zero to disable witness at runtime. Note that you
can't turn witness back on once it is off. You can only turn it off as
a one-way switch.
- Added a comment describing the possible values of witness_watch.
a parameter instead of using the level of a given witness. When
recursing, pass an indent level of indent + 1.
- Make use of the information witness_levelall() provides in
witness_display_list() to use an O(n) algorithm instead of an O(n^2)
algo to decide which witnesses to display hierarchies from. Basically,
we only display a hierarchy for witnesses with a level of 0.
- Add a new per-witness flag that is reset at the start of
witness_display() for all witness's and is set the first time a witness
is displayed in witness_displaydescendants(). If a witness is
encountered more than once in the lock order tree (which happens often),
witness_displaydescendants() marks the later occurrences with the string
"(already displayed)" and doesn't display the subtree under that
witness. This avoids duplicating large amounts of the lock order tree
in the 'show witness' output in DDB.
All these changes serve to make 'show witness' a lot more readable and
useful than it was previously.
adds a witness to the child list of a parent witness. rebalancetree()
runs through the entire tree removing direct descendants of witnesses
who already have said child witness as an indirect descendant through
another direct descendant. itismychild() now calls insertchild()
followed by rebalancetree() and no longer needs the evil hack of
having static recursed variable.
- Add a function reparentchildren() that adds all the direct descendants
of one witness as direct descendants of another witness.
- Change the return value of itismychild() and similar functions so that
they return 0 in the case of failure due to lack of resources instead
of 1. This makes the return value more intuitive.
- Check the return value of itismychild() when defining the static lock
order in witness_initialize().
- Don't try to setup a lock instance in witness_lock() if itismychild()
fails. Witness is hosed anyways so no need to do any more witness
related activity at that point. It also makes the code flow easier to
understand.
- Add a new depart() function as the opposite of enroll(). When the
reference count of a witness drops to 0 in witness_destroy(), this
function is called on that witness. First, it runs through the
lock order tree using reparentchildren() to reparent direct descendants
of the departing witness to each of the witness' parents in the tree.
Next, it releases it's own child list and other associated resources.
Finally it calls rebalanacetree() to rebalance the lock order tree.
- Sort function prototypes into something closer to alphabetical order.
As a result of these changes, there should no longer be 'dead' witnesses
in the order tree, and repeatedly loading and unloading a module should no
longer exhaust witness of its internal resources.
Inspired by: gallatin
recursing on a lock instead of before. This fixes a bug where WITNESS
could get a little confused if you did an sx_tryslock() on a sx lock that
you already had an slock on. WITNESS would still function correctly but
it could result in weirdness in the output of 'show locks'. This also
makes it possible for mtx_trylock() to recurse on a lock.
ddb 'show locks' command. Thus, move witness_list() to the #ifdef DDB
section and remove extra checks for calling this function outside of
DDB. Also, witness_list() now returns void instead of returning an int.
Reported by: Steve Ames <steve@energistic.com>
Prodded by: davidxu
witness. Sleepable locks such as sx locks always come before all mutexes
including Giant. However, the static lock order list placed Giant before
the proctree and allproc sx locks. This resulted in witness creating a
cycle in its lock order "tree" (real trees don't have cycles) leading to
infinite recursion and eventually a double fault. To fix, put Giant after
sx locks in the lock order list.
- Add a comment about special lock order rules and Giant near the top of
subr_witness.c. Specifically, this documents and explains the real lock
order relationship between Giant and sleepable locks (i.e. lockmgr locks
and sx locks). Basically, Giant can be safely acquired either before or
after sleepable locks and the case of Giant before a sleepable lock is
exempted as a special case.
- Add a new static function 'witness_list_lock()' that displays a single
line of information about a struct lock_instance. This is used to
make the output of witness messages more consistent and reduce some code
duplication.
- Fixup a few comments in witness_lock().
- Properly handle the Giant-before-sleepable-lock lock order exception in
a more general fashion and remove the no longer needed LI_SLEPT flag.
- Break up the last condition before assuming a reversal a bit to try
and make the logic less confusing in witness_lock().
- Axe WITNESS_SLEEP() now that LI_SLEPT is no longer needed and replace it
with a more general WITNESS_WARN() macro/function combination.
WITNESS_WARN() allows you to output a customized message out to the
console along with a list of held locks. It will optionally drop into
the debugger as well. You can exempt a single lock from the check by
passing it in as the second argument. You can also use flags to specify
if Giant should be exempt from the check, if all sleepable locks should
be exempt from the check, and if witness should panic if any non-exempt
locks are found.
- Make the witness_list() function static. Other areas of the kernel
should use the new WITNESS_WARN() instead.
#if'ed out for a while. Complete the deed and tidy up some other bits.
We need to be able to call this stuff from outer edges of interrupt
handlers for devices that have the ISR bits in pci config space. Making
the bios code mpsafe was just too hairy. We had also stubbed it out some
time ago due to there simply being too much brokenness in too many systems.
This adds a leaf lock so that it is safe to use pci_read_config() and
pci_write_config() from interrupt handlers. We still will use pcibios
to do interrupt routing if there is no acpi.. [yes, I tested this]
Briefly glanced at by: imp
I'm not convinced there is anything major wrong with the patch but
them's the rules..
I am using my "David's mentor" hat to revert this as he's
offline for a while.
data structure called kse_upcall to manage UPCALL. All KSE binding
and loaning code are gone.
A thread owns an upcall can collect all completed syscall contexts in
its ksegrp, turn itself into UPCALL mode, and takes those contexts back
to userland. Any thread without upcall structure has to export their
contexts and exit at user boundary.
Any thread running in user mode owns an upcall structure, when it enters
kernel, if the kse mailbox's current thread pointer is not NULL, then
when the thread is blocked in kernel, a new UPCALL thread is created and
the upcall structure is transfered to the new UPCALL thread. if the kse
mailbox's current thread pointer is NULL, then when a thread is blocked
in kernel, no UPCALL thread will be created.
Each upcall always has an owner thread. Userland can remove an upcall by
calling kse_exit, when all upcalls in ksegrp are removed, the group is
atomatically shutdown. An upcall owner thread also exits when process is
in exiting state. when an owner thread exits, the upcall it owns is also
removed.
KSE is a pure scheduler entity. it represents a virtual cpu. when a thread
is running, it always has a KSE associated with it. scheduler is free to
assign a KSE to thread according thread priority, if thread priority is changed,
KSE can be moved from one thread to another.
When a ksegrp is created, there is always N KSEs created in the group. the
N is the number of physical cpu in the current system. This makes it is
possible that even an userland UTS is single CPU safe, threads in kernel still
can execute on different cpu in parallel. Userland calls kse_create to add more
upcall structures into ksegrp to increase concurrent in userland itself, kernel
is not restricted by number of upcalls userland provides.
The code hasn't been tested under SMP by author due to lack of hardware.
Reviewed by: julian
earlier acquired lock with the same witness as the lock currently being
acquired. If we had released several earlier acquired locks after
acquiring enough locks to require another lock_list_entry bucket in the
lock list, then subsequent lock_list_entry buckets could contain only one
lock instance in which case i would be zero.
Reported by: Joel M. Baldwin <qumqats@outel.org>
- Get the initial mode from the prom settings and don't clobber the mode
on open.
- Copy output into an internal ring buffer instead of accessing the tty
outq directly in the interrupt handler. This fixes a problem where
garbage would show up in the output stream.
- Reset the console port completely and reprogram all the parameters
before enabling it. This fixes seemingly random hangs on startup
when using a fast interrupt handler.
- Add minimal locking in place of spls.
- Remove dead code and minor cleanups.
- It actually works this time, honest!
- Fine grained TLB shootdowns for SMP on i386. IPI's are very expensive,
so try and optimize things where possible.
- Introduce ranged shootdowns that can be done as a single IPI.
- PG_G support for i386
- Specific-cpu targeted shootdowns. For example, there is no sense in
globally purging the TLB cache for where we are stealing a page from
the local unshared process on the local cpu. Use pm_active to track
this.
- Add some instrumentation for the tlb shootdown code.
- Rip out SMP code from <machine/cpufunc.h>
- Try and fix some very bogus PG_G and PG_PS interactions that were bad
enough to cause vm86 bios calls to break. vm86 depended on our existing
bugs and this was the cause of the VESA panics last time.
- Fix the silly one-line error that caused the 'panic: bad pte' last time.
- Fix a couple of other silly one-line errors that should have caused more
pain than they did.
Some more work is needed:
- pmap_{zero,copy}_page[_idle]. These can be done without IPI's if we
have a hook in cpu_switch.
- The IPI handlers need some cleanup. I have a bogus %ds load that can
be avoided.
- APTD handling is rather bogus and appears to be a large source of
global TLB IPI shootdowns for no really good reason.
I see speedups of between 1.5% and ~4% on buildworlds in a while 1 loop.
I expect to see a bigger difference when there is significant pageout
activity or the system otherwise has memory shortages.
I have backed out a few optimizations that I had been using over the last
few days in order to be a little more conservative. I'll revisit these
again over the next few days as the dust settles.
New option: DISABLE_PG_G - In case I missed something.
queue lock (revision 1.33 of vm/vm_page.c removed them).
o Make the free queue lock a spin lock because it's sometimes acquired
inside of a critical section.
The ability to schedule multiple threads per process
(one one cpu) by making ALL system calls optionally asynchronous.
to come: ia64 and power-pc patches, patches for gdb, test program (in tools)
Reviewed by: Almost everyone who counts
(at various times, peter, jhb, matt, alfred, mini, bernd,
and a cast of thousands)
NOTE: this is still Beta code, and contains lots of debugging stuff.
expect slight instability in signals..
lock_object by another pointer (though all of lock_object should be
conditional on LOCK_DEBUG anyways) in exchange for an O(1) TAILQ_REMOVE()
in witness_destroy() (called for every mtx_destroy() and sx_destroy())
instead of an O(n) STAILQ_REMOVE. Since WITNESS is so dog slow as it is,
the speed-up is worth the space cost.
Suggested by: iedowse
being created and destroyed without a single long-term one around to ensure
the witness associated with that group of locks stays alive. The pipe
mutexes are an example of this group. For a dead witness we no longer
clear the witness name. Instead, when looking up the witness for a lock,
if a dead witness' (a witness with a refcount of 0) w_name pointer is
identical to the witness name of the lock then we revive that witness
instead of using a new witness for the lock. This results in far fewer
dead witness objects and also better preserves locking orders over the long
term resulting in more correct lock order checking. Note that we can't
ever derefence w_name of a dead witness since we don't know if the string
it is pointing to has been free()'d or kldunload()'d out from under us.
