striping to a per device round-robin algorithm.
Because of the policy of not attempting to retain previous swap
allocation on page-out, this means that a newly added swap device
almost instantly takes its 1/N share of the I/O load but it takes
somewhat longer for it to assume it's 1/N share of the pages if there
is plenty of space on the other devices.
Change the 8G total swapspace limitation to 8G per device instead
by using a per device blist rather than one global blist. This
reduces the memory footprint by 75% (typically a couple hundred
kilobytes) for the common case with one swapdevice but NSWAPDEV=4.
Remove the compile time constant limit of number of swap devices,
there is no limit now. Instead of a fixed size array, store the
per swapdev structure in a TAILQ.
Total swap space is still addressed by a 32 bit page number and
therefore the upper limit is now 2^42 bytes = 16TB (for i386).
We still do not allocate the first page of each device in order to
give some amount of protection to any bsdlabel at the start of the
device.
A new device is appended after the existing devices in the swap space,
no attempt is made to fill in holes left behind by swapoff (this can
trivially be changed should it ever become a problem).
The sysctl vm.nswapdev now reflects the number of currently configured
swap devices.
Rename vm_swap_size to swap_pager_avail for consistency with other
exported names.
Change argument type for vm_proc_swapin_all() and swap_pager_isswapped()
to be a struct swdevt pointer rather than an index.
Not changed: we are still using blists to manage the free space,
but since the swapspace is no longer fragmented by the striping
different resource managers might fare better.
- Change vm_pageout_object_deactivate_pages()'s first parameter from a
vm_map_t to a pmap_t.
- Change vm_pageout_object_deactivate_pages()'s and
vm_pageout_map_deactivate_pages()'s last parameter from a vm_pindex_t
to a long. Since the number of pages in an address space doesn't
require 64 bits on an i386, vm_pindex_t is overkill.
processes in the first pass. Among other things, this will give
us a chance to launder vnode-backed pages before concluding that
we need more swap. This is particularly useful for systems that
have no swap.
While here, update a comment and remove some long-unused code.
Reported by: Lucky Green <shamrock@cypherpunks.to>
Suggested by: dillon
Approved by: re (rwatson)
- Add a parameter to vm_pageout_flush() that tells vm_pageout_flush()
whether its caller has locked the vm_object. (This is a temporary
measure to bootstrap vm_object locking.)
critical and should not be killed when pageout is looking for more
memory pages in all the wrong places.
Reviewed by: arch@
Sponsored by: St. Bernard Software
process to kill, don't block on a map lock while holding the
process lock. Instead, skip processes whose map locks are held
and find something else to kill.
- Add vm_map_trylock_read() to support the above.
Reviewed by: alc, mike (mentor)
intended to be used by significant memory consumers so that they may drain
some of their caches.
Inspired by: phk
Approved by: re
Tested on: x86, alpha
to reflect its new location, and add page queue and flag locking.
Notes: (1) alpha, i386, and ia64 had identical implementations
of pmap_collect() in terms of machine-independent interfaces;
(2) sparc64 doesn't require it; (3) powerpc had it as a TODO.
indirectly through vm_page_protect(). The one remaining page flag that
is updated by vm_page_protect() is already being updated by our various
pmap implementations.
Note: A later commit will similarly change the VM_PROT_READ case and
eliminate vm_page_protect().
- Begin moving scheduler specific functionality into sched_4bsd.c
- Replace direct manipulation of scheduler data with hooks provided by the
new api.
- Remove KSE specific state modifications and single runq assumptions from
kern_switch.c
Reviewed by: -arch
constants VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS, USRSTACK and PS_STRINGS.
This is mainly so that they can be variable even for the native abi, based
on different machine types. Get stack protections from the sysentvec too.
This makes it trivial to map the stack non-executable for certain abis, on
machines that support it.
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..
best path forward now is likely to change the lockmgr locks to simple
sleep mutexes, then see if any extra contention it generates is greater
than removed overhead of managing local locking state information,
cost of extra calls into lockmgr, etc.
Additionally, making the vm_map lock a mutex and respecting it properly
will put us much closer to not needing Giant magic in vm.
While doing this, move it earlier in the sysinit boot process so that the
VM system can use it.
After that, the system is now able to use sx locks instead of lockmgr
locks in the VM system. To accomplish this, some of the more
questionable uses of the locks (such as testing whether they are
owned or not, as well as allowing shared+exclusive recursion) are
removed, and simpler logic throughout is used so locks should also be
easier to understand.
This has been tested on my laptop for months, and has not shown any
problems on SMP systems, either, so appears quite safe. One more
user of lockmgr down, many more to go :)
style(9)
- Minor space adjustment in cases where we have "( ", " )", if(), return(),
while(), for(), etc.
- Add /* SYMBOL */ after a few #endifs.
Reviewed by: alc
In order to determine what to page out, the vm_daemon checks
reference bits on all pages belonging to all processes. Unfortunately,
the algorithm used reacted badly with shared pages; each shared page
would be checked once per process sharing it; this caused an O(N^2)
growth of tlb invalidations. The algorithm has been changed so that
each page will be checked only 16 times.
Prior to this change, a fork/sleepbomb of 1300 processes could cause
the vm_daemon to take over 60 seconds to complete, effectively
freezing the system for that time period. With this change
in place, the vm_daemon completes in less than a second. Any system
with hundreds of processes sharing pages should benefit from this change.
Note that the vm_daemon is only run when the system is under extreme
memory pressure. It is likely that many people with loaded systems saw
no symptoms of this problem until they reached the point where swapping
began.
Special thanks go to dillon, peter, and Chuck Cranor, who helped me
get up to speed with vm internals.
PR: 33542, 20393
Reviewed by: dillon
MFC after: 1 week
- Allow the OOM killer to target processes currently locked in
memory. These very often are the ones doing the memory hogging.
- Drop the wakeup priority of processes currently sleeping while
waiting for their page fault to complete. In order for the OOM
killer to work well, the killed process and other system processes
waiting on memory must be allowed to wakeup first.
Reviewed by: dillon
MFC after: 1 week
against VM_WAIT in the pageout code. Both fixes involve adjusting
the lockmgr's timeout capability so locks obtained with timeouts do not
interfere with locks obtained without a timeout.
Hopefully MFC: before the 4.5 release
Note ALL MODULES MUST BE RECOMPILED
make the kernel aware that there are smaller units of scheduling than the
process. (but only allow one thread per process at this time).
This is functionally equivalent to teh previousl -current except
that there is a thread associated with each process.
Sorry john! (your next MFC will be a doosie!)
Reviewed by: peter@freebsd.org, dillon@freebsd.org
X-MFC after: ha ha ha ha
Also removed some spl's and added some VM mutexes, but they are not actually
used yet, so this commit does not really make any operational changes
to the system.
vm_page.c relates to vm_page_t manipulation, including high level deactivation,
activation, etc... vm_pageq.c relates to finding free pages and aquiring
exclusive access to a page queue (exclusivity part not yet implemented).
And the world still builds... :-)
(this commit is just the first stage). Also add various GIANT_ macros to
formalize the removal of Giant, making it easy to test in a more piecemeal
fashion. These macros will allow us to test fine-grained locks to a degree
before removing Giant, and also after, and to remove Giant in a piecemeal
fashion via sysctl's on those subsystems which the authors believe can
operate without Giant.
processes a little earlier to avoid a deadlock. Second, when calculating
the 'largest process' do not just count RSS. Instead count the RSS + SWAP
used by the process. Without this the code tended to kill small
inconsequential processes like, oh, sshd, rather then one of the many
'eatmem 200MB' I run on a whim :-). This fix has been extensively tested on
-stable and somewhat tested on -current and will be MFCd in a few days.
Shamed into fixing this by: ps
- Assert Giant in vm_pageout_scan() for the vnode hacking that it does.
- Don't hold vm_mtx around vget() or vput().
- Lock Giant when calling vm_pageout_scan() from the pagedaemon. Also,
lock curproc while setting the P_BUFEXHAUST flag.
- For now we still hold Giant for all of the vm_daemon. When process
limits are locked we will be only need Giant for swapout_procs().
vm_mtx does not recurse and is required for most low level
vm operations.
faults can not be taken without holding Giant.
Memory subsystems can now call the base page allocators safely.
Almost all atomic ops were removed as they are covered under the
vm mutex.
Alpha and ia64 now need to catch up to i386's trap handlers.
FFS and NFS have been tested, other filesystems will need minor
changes (grabbing the vm lock when twiddling page properties).
Reviewed (partially) by: jake, jhb
the process in question locked as soon as we find it and determine it to
be eligible until we actually kill it. To avoid deadlock, we don't block
on the process lock but skip any process that is already locked during our
search.
other "system" header files.
Also help the deprecation of lockmgr.h by making it a sub-include of
sys/lock.h and removing sys/lockmgr.h form kernel .c files.
Sort sys/*.h includes where possible in affected files.
OK'ed by: bde (with reservations)
mtx_enter(lock, type) becomes:
mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks)
mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized)
similarily, for releasing a lock, we now have:
mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN.
We change the caller interface for the two different types of locks
because the semantics are entirely different for each case, and this
makes it explicitly clear and, at the same time, it rids us of the
extra `type' argument.
The enter->lock and exit->unlock change has been made with the idea
that we're "locking data" and not "entering locked code" in mind.
Further, remove all additional "flags" previously passed to the
lock acquire/release routines with the exception of two:
MTX_QUIET and MTX_NOSWITCH
The functionality of these flags is preserved and they can be passed
to the lock/unlock routines by calling the corresponding wrappers:
mtx_{lock, unlock}_flags(lock, flag(s)) and
mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN
locks, respectively.
Re-inline some lock acq/rel code; in the sleep lock case, we only
inline the _obtain_lock()s in order to ensure that the inlined code
fits into a cache line. In the spin lock case, we inline recursion and
actually only perform a function call if we need to spin. This change
has been made with the idea that we generally tend to avoid spin locks
and that also the spin locks that we do have and are heavily used
(i.e. sched_lock) do recurse, and therefore in an effort to reduce
function call overhead for some architectures (such as alpha), we
inline recursion for this case.
Create a new malloc type for the witness code and retire from using
the M_DEV type. The new type is called M_WITNESS and is only declared
if WITNESS is enabled.
Begin cleaning up some machdep/mutex.h code - specifically updated the
"optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN
and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently
need those.
Finally, caught up to the interface changes in all sys code.
Contributors: jake, jhb, jasone (in no particular order)
in 4.2-REL which I ripped out in -stable and -current when implementing the
low-memory handling solution. However, maxlaunder turns out to be the saving
grace in certain very heavily loaded systems (e.g. newsreader box). The new
algorithm limits the number of pages laundered in the first pageout daemon
pass. If that is not sufficient then suceessive will be run without any
limit.
Write I/O is now pipelined using two sysctls, vfs.lorunningspace and
vfs.hirunningspace. This prevents excessive buffered writes in the
disk queues which cause long (multi-second) delays for reads. It leads
to more stable (less jerky) and generally faster I/O streaming to disk
by allowing required read ops (e.g. for indirect blocks and such) to occur
without interrupting the write stream, amoung other things.
NOTE: eventually, filesystem write I/O pipelining needs to be done on a
per-device basis. At the moment it is globalized.
struct swblock entries by dividing the number of the entries by 2
until the swap metadata fits.
- Reject swapon(2) upon failure of swap_zone allocation.
This is just a temporary fix. Better solutions include:
(suggested by: dillon)
o reserving swap in SWAP_META_PAGES chunks, and
o swapping the swblock structures themselves.
Reviewed by: alfred, dillon
of explicit calls to lockmgr. Also provides macros for the flags
pased to specify shared, exclusive or release which map to the
lockmgr flags. This is so that the use of lockmgr can be easily
replaced with optimized reader-writer locks.
- Add some locking that I missed the first time.
cases with file fragments and read-write mmap's can lead to a situation
where a VM page has odd dirty bits, e.g. 0xFC - due to being dirtied by
an mmap and only the fragment (representing a non-page-aligned end of
file) synced via a filesystem buffer. A correct solution that
guarentees consistent m->dirty for the file EOF case is being
worked on. In the mean time we can't be so conservative in the
KASSERT.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
include:
* Mutual exclusion is used instead of spl*(). See mutex(9). (Note: The
alpha port is still in transition and currently uses both.)
* Per-CPU idle processes.
* Interrupts are run in their own separate kernel threads and can be
preempted (i386 only).
Partially contributed by: BSDi (BSD/OS)
Submissions by (at least): cp, dfr, dillon, grog, jake, jhb, sheldonh
the gating of system calls that cause modifications to the underlying
filesystem. The gating can be enabled by any filesystem that needs
to consistently suspend operations by adding the vop_stdgetwritemount
to their set of vnops. Once gating is enabled, the function
vfs_write_suspend stops all new write operations to a filesystem,
allows any filesystem modifying system calls already in progress
to complete, then sync's the filesystem to disk and returns. The
function vfs_write_resume allows the suspended write operations to
begin again. Gating is not added by default for all filesystems as
for SMP systems it adds two extra locks to such critical kernel
paths as the write system call. Thus, gating should only be added
as needed.
Details on the use and current status of snapshots in FFS can be
found in /sys/ufs/ffs/README.snapshot so for brevity and timelyness
is not included here. Unless and until you create a snapshot file,
these changes should have no effect on your system (famous last words).
and sysv shared memory support for it. It implements a new
PG_UNMANAGED flag that has slightly different characteristics
from PG_FICTICIOUS.
A new sysctl, kern.ipc.shm_use_phys has been added to enable the
use of physically-backed sysv shared memory rather then swap-backed.
Physically backed shm segments are not tracked with PV entries,
allowing programs which use a large shm segment as a rendezvous
point to operate without eating an insane amount of KVM in the
PV entry management. Read: Oracle.
Peter's OBJT_PHYS object will also allow us to eventually implement
page-table sharing and/or 4MB physical page support for such segments.
We're half way there.
have pv_entries. This is intended for very special circumstances,
eg: a certain database that has a 1GB shm segment mapped into 300
processes. That would consume 2GB of kvm just to hold the pv_entries
alone. This would not be used on systems unless the physical ram was
available, as it's not pageable.
This is a work-in-progress, but is a useful and functional checkpoint.
Matt has got some more fixes for it that will be committed soon.
Reviewed by: dillon
to various pmap_*() functions instead of looking up the physical address
and passing that. In many cases, the first thing the pmap code was doing
was going to a lot of trouble to get back the original vm_page_t, or
it's shadow pv_table entry.
Inspired by: John Dyson's 1998 patches.
Also:
Eliminate pv_table as a seperate thing and build it into a machine
dependent part of vm_page_t. This eliminates having a seperate set of
structions that shadow each other in a 1:1 fashion that we often went to
a lot of trouble to translate from one to the other. (see above)
This happens to save 4 bytes of physical memory for each page in the
system. (8 bytes on the Alpha).
Eliminate the use of the phys_avail[] array to determine if a page is
managed (ie: it has pv_entries etc). Store this information in a flag.
Things like device_pager set it because they create vm_page_t's on the
fly that do not have pv_entries. This makes it easier to "unmanage" a
page of physical memory (this will be taken advantage of in subsequent
commits).
Add a function to add a new page to the freelist. This could be used
for reclaiming the previously wasted pages left over from preloaded
loader(8) files.
Reviewed by: dillon
* lockstatus() and VOP_ISLOCKED() gets a new process argument and a new
return value: LK_EXCLOTHER, when the lock is held exclusively by another
process.
* The ASSERT_VOP_(UN)LOCKED family is extended to use what this gives them
* Extend the vnode_if.src format to allow more exact specification than
locked/unlocked.
This commit should not do any semantic changes unless you are using
DEBUG_VFS_LOCKS.
Discussed with: grog, mch, peter, phk
Reviewed by: peter
eliminate an extra (useless) level of indirection in half of the page
queue accesses and (2) to use a single name for each queue throughout,
instead of, e.g., "vm_page_queue_active" in some places and
"vm_page_queues[PQ_ACTIVE]" in others.
Reviewed by: dillon
Merge the contents (less some trivial bordering the silly comments)
of <vm/vm_prot.h> and <vm/vm_inherit.h> into <vm/vm.h>. This puts
the #defines for the vm_inherit_t and vm_prot_t types next to their
typedefs.
This paves the road for the commit to follow shortly: change
useracc() to use VM_PROT_{READ|WRITE} rather than B_{READ|WRITE}
as argument.
Replace various VM related page count calculations strewn over the
VM code with inlines to aid in readability and to reduce fragility
in the code where modules depend on the same test being performed
to properly sleep and wakeup.
Split out a portion of the page deactivation code into an inline
in vm_page.c to support vm_page_dontneed().
add vm_page_dontneed(), which handles the madvise MADV_DONTNEED
feature in a related commit coming up for vm_map.c/vm_object.c. This
code prevents degenerate cases where an essentially active page may
be rotated through a subset of the paging lists, resulting in premature
disposal.
QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN,
QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean
and dirty buffers have been separated. Empty buffers with KVM
assignments have been separated from truely empty buffers. getnewbuf()
has been rewritten and now operates in a 100% optimal fashion. That is,
it is able to find precisely the right kind of buffer it needs to
allocate a new buffer, defragment KVM, or to free-up an existing buffer
when the buffer cache is full (which is a steady-state situation for
the buffer cache).
Buffer flushing has been reorganized. Previously buffers were flushed
in the context of whatever process hit the conditions forcing buffer
flushing to occur. This resulted in processes blocking on conditions
unrelated to what they were doing. This also resulted in inappropriate
VFS stacking chains due to multiple processes getting stuck trying to
flush dirty buffers or due to a single process getting into a situation
where it might attempt to flush buffers recursively - a situation that
was only partially fixed in prior commits. We have added a new daemon
called the buf_daemon which is responsible for flushing dirty buffers
when the number of dirty buffers exceeds the vfs.hidirtybuffers limit.
This daemon attempts to dynamically adjust the rate at which dirty buffers
are flushed such that getnewbuf() calls (almost) never block.
The number of nbufs and amount of buffer space is now scaled past the
8MB limit that was previously imposed for systems with over 64MB of
memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed
somewhat. The number of physical buffers has been increased with the
intention that we will manage physical I/O differently in the future.
reassignbuf previously attempted to keep the dirtyblkhd list sorted which
could result in non-deterministic operation under certain conditions,
such as when a large number of dirty buffers are being managed. This
algorithm has been changed. reassignbuf now keeps buffers locally sorted
if it can do so cheaply, and otherwise gives up and adds buffers to
the head of the dirtyblkhd list. The new algorithm is deterministic but
not perfect. The new algorithm greatly reduces problems that previously
occured when write_behind was turned off in the system.
The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive
P_BUFEXHAUST bit. This bit allows processes working with filesystem
buffers to use available emergency reserves. Normal processes do not set
this bit and are not allowed to dig into emergency reserves. The purpose
of this bit is to avoid low-memory deadlocks.
A small race condition was fixed in getpbuf() in vm/vm_pager.c.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
Reviewed by: Kirk McKusick <mckusick@mckusick.com>
