The zone allocator's locks should be leaflocks, meaning that they
should never be held when entering into another subsystem, however
the sysctl grabs the zone global mutex and individual zone mutexes
while holding the lock it calls SYSCTL_OUT which recurses into the
VM subsystem in order to wire user memory to do a safe copy. This
can block and cause lock order reversals.
To fix this:
lock zone global.
get a count of the number of zones.
unlock global.
allocate temporary storage.
format and SYSCTL_OUT the banner.
lock global.
traverse list.
make sure we haven't looped more than the initial count taken
to avoid overflowing the allocated buffer.
lock each nodes.
read values and format into buffer.
unlock individual node.
unlock global.
format and SYSCTL_OUT the rest of the data.
free storage.
return.
Other problems included not checking for errors when doing sysctl out
of the column header. Fixed.
Inconsistant termination of the copied string. Fixed.
Objected to by: des (for not using sbuf)
Since the output is not variable length and I'm actually over
allocating signifigantly and I'd like to get this fixed now, I'll
work on the sbuf convertion at a later date. I would not object
to someone else taking it upon themselves to convert it to sbuf.
I hold no MAINTIANER rights to this code (for now).
Protect pager object list manipulation with a mutex.
It doesn't look possible to combine them under a single sx lock because
creation may block and we can't have the object list manipulation block
on anything other than a mutex because of interrupt requests.
VOP_BWRITE() was a hack which made it possible for NFS client
side to use struct buf with non-bio backing.
This patch takes a more general approach and adds a bp->b_op
vector where more methods can be added.
The success of this patch depends on bp->b_op being initialized
all relevant places for some value of "relevant" which is not
easy to determine. For now the buffers have grown a b_magic
element which will make such issues a tiny bit easier to debug.
programs. There is a case during a fork() which can cause a deadlock.
From Tor -
The workaround that consists of setting a flag in the vm map that
indicates that a fork is in progress and using that mark in the page
fault handling to force a revalidation failure. That change will only
affect (pessimize) page fault handling during fork for threaded
(linuxthreads style) applications and applications using aio_*().
Submited by: tegge
call is correct, but it interferes with the massive hack called
vm_map_growstack(). The call will be returned after our stack handling
code is fixed.
Reported by: tegge
reference count was transferred to the new object, but both the
new and the old map entries had pointers to the new object.
Correct this by transferring the second reference.
This fixes a panic that can occur when mmap(2) is used with the
MAP_INHERIT flag.
PR: i386/25603
Reviewed by: dillon, alc
supported architectures such as the alpha. This allows us to save
on kernel virtual address space, TLB entries, and (on the ia64) VHPT
entries. pmap_map() now modifies the passed in virtual address on
architectures that do not support direct-mapped segments to point to
the next available virtual address. It also returns the actual
address that the request was mapped to.
- On the IA64 don't use a special zone of PV entries needed for early
calls to pmap_kenter() during pmap_init(). This gets us in trouble
because we end up trying to use the zone allocator before it is
initialized. Instead, with the pmap_map() change, the number of needed
PV entries is small enough that we can get by with a static pool that is
used until pmap_init() is complete.
Submitted by: dfr
Debugging help: peter
Tested by: me
of memory, rather than from the start.
This fixes problems allocating bouncebuffers on alphas where there is only
1 chunk of memory (unlike PCs where there is generally at least one small
chunk and a large chunk). Having 1 chunk had been fatal, because these
structures take over 13MB on a machine with 1GB of ram. This doesn't leave
much room for other structures and bounce buffers if they're at the front.
Reviewed by: dfr, anderson@cs.duke.edu, silence on -arch
Tested by: Yoriaki FUJIMORI <fujimori@grafin.fujimori.cache.waseda.ac.jp>
make sure that PG_NOSYNC is properly set. Previously we only set it
for a write-fault, but this can occur on a read-fault too.
(will be MFCd prior to 4.3 freeze)
this information via the vm.nswapdev sysctl (number of swap areas)
and vm.swapdevX nodes (where X is the device), which contain the MIBs
dev, blocks, used, and flags. These changes are required to allow
top and other userland swap-monitoring utilities to run without
setgid kmem.
Submitted by: Thomas Moestl <tmoestl@gmx.net>
Reviewed by: freebsd-audit
- All processes go into the same array of queues, with different
scheduling classes using different portions of the array. This
allows user processes to have their priorities propogated up into
interrupt thread range if need be.
- I chose 64 run queues as an arbitrary number that is greater than
32. We used to have 4 separate arrays of 32 queues each, so this
may not be optimal. The new run queue code was written with this
in mind; changing the number of run queues only requires changing
constants in runq.h and adjusting the priority levels.
- The new run queue code takes the run queue as a parameter. This
is intended to be used to create per-cpu run queues. Implement
wrappers for compatibility with the old interface which pass in
the global run queue structure.
- Group the priority level, user priority, native priority (before
propogation) and the scheduling class into a struct priority.
- Change any hard coded priority levels that I found to use
symbolic constants (TTIPRI and TTOPRI).
- Remove the curpriority global variable and use that of curproc.
This was used to detect when a process' priority had lowered and
it should yield. We now effectively yield on every interrupt.
- Activate propogate_priority(). It should now have the desired
effect without needing to also propogate the scheduling class.
- Temporarily comment out the call to vm_page_zero_idle() in the
idle loop. It interfered with propogate_priority() because
the idle process needed to do a non-blocking acquire of Giant
and then other processes would try to propogate their priority
onto it. The idle process should not do anything except idle.
vm_page_zero_idle() will return in the form of an idle priority
kernel thread which is woken up at apprioriate times by the vm
system.
- Update struct kinfo_proc to the new priority interface. Deliberately
change its size by adjusting the spare fields. It remained the same
size, but the layout has changed, so userland processes that use it
would parse the data incorrectly. The size constraint should really
be changed to an arbitrary version number. Also add a debug.sizeof
sysctl node for struct kinfo_proc.
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)
by myself. It solves a serious vm_map corruption problem that can occur
with the buffer cache when block sizes > 64K are used. This code has been
heavily tested in -stable but only tested somewhat on -current. An MFC
will occur in a few days. My additions include the vm_map_simplify_entry()
and minor buffer cache boundry case fix.
Make the buffer cache use a system map for buffer cache KVM rather then a
normal map.
Ensure that VM objects are not allocated for system maps. There were cases
where a buffer map could wind up with a backing VM object -- normally
harmless, but this could also result in the buffer cache blocking in places
where it assumes no blocking will occur, possibly resulting in corrupted
maps.
Fix a minor boundry case in the buffer cache size limit is reached that
could result in non-optimal code.
Add vm_map_simplify_entry() calls to prevent 'creeping proliferation'
of vm_map_entry's in the buffer cache's vm_map. Previously only a simple
linear optimization was made. (The buffer vm_map typically has only a
handful of vm_map_entry's. This stabilizes it at that level permanently).
PR: 20609
Submitted by: (Tor Egge) tegge
- Proc locking in a few places.
- faultin() now must be called with the proc lock held.
- Split up swappable() into a couple of tests so that it can be locke in
swapout_procs().
- Use queue macros.
- replace the simplelock in struct vm_zone with a mutex.
- use a proper SLIST rather than a hand-rolled job for the zone list.
- add a subsystem lock that protects the zone list and the statistics
counters.
- merge _zalloc() into zalloc() and _zfree() into zfree(), and
move them below _zget() so there's no need for a prototype.
- add two initialization functions: one which initializes the
subsystem mutex and the zone list, and one that currently doesn't
do anything.
- zap zerror(); use KASSERTs instead.
- dike out half of sysctl_vm_zone(), which was mostly trying to do
manually what the snprintf() call could do better.
Reviewed by: jhb, jasone
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.
The fix works by reverting the ordering of free memory so that the
chances of contig_malloc() succeeding increases.
PR: 23291
Submitted by: Andrew Atrens <atrens@nortel.ca>
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.
Backout the previous delta (rev 1.4), it didn't make any difference.
If the requested handle is NULL then don't add it to the list of
objects, to be found by handle.
The problem is that when asking for a NULL handle you are implying
you want a new object. Because objects with NULL handles were
being added to the list, any further requests for phys backed
objects with NULL handles would return a reference to the initial
NULL handle object after finding it on the list.
Basically one couldn't have more than one phys backed object without
a handle in the entire system without this fix. If you did more
than one shared memory allocation using the phys pager it would
give you your initial allocation again.
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>
Pre-rfork code assumed inherent locking of a process's file descriptor
array. However, with the advent of rfork() the file descriptor table
could be shared between processes. This patch closes over a dozen
serious race conditions related to one thread manipulating the table
(e.g. closing or dup()ing a descriptor) while another is blocked in
an open(), close(), fcntl(), read(), write(), etc...
PR: kern/11629
Discussed with: Alexander Viro <viro@math.psu.edu>
the offending inline function (BUF_KERNPROC) on it being #included
already.
I'm not sure BUF_KERNPROC() is even the right thing to do or in the
right place or implemented the right way (inline vs normal function).
Remove consequently unneeded #includes of <sys/proc.h>
- Close a small race condition. The sched_lock mutex protects
p->p_stat as well as the run queues. Another CPU could change p_stat
of the process while we are waiting for the lock, and we would end up
scheduling a process that isn't runnable.
write caching is disabled on both SCSI and IDE disks where large
memory dumps could take up to an hour to complete.
Taking an i386 scsi based system with 512MB of ram and timing (in
seconds) how long it took to complete a dump, the following results
were obtained:
Before: After:
WCE TIME WCE TIME
------------------ ------------------
1 141.820972 1 15.600111
0 797.265072 0 65.480465
Obtained from: Yahoo!
Reviewed by: peter
in the face of non-stripe-aligned swap areas. The bug could cause a
panic during boot.
Refuse to configure a swap area that is too large (67 GB or so)
Properly document the power-of-2 requirement for SWB_NPAGES.
The patch is slightly different then the one Tor enclosed in the P.R.,
but accomplishes the same thing.
PR: kern/20273
Submitted by: Tor.Egge@fast.no
and initialized during boot. This avoids bloating sizeof(struct lock).
As a side effect, it is no longer necessary to enforce the assumtion that
lockinit()/lockdestroy() calls are paired, so the LK_VALID flag has been
removed.
Idea taken from: BSD/OS.
it to lower its memory usage. This was mentioned on the mailing
lists ages ago, and I've lost the name of the person who brought
it up.
Reviewed by: alc
Add lockdestroy() and appropriate invocations, which corresponds to
lockinit() and must be called to clean up after a lockmgr lock is no
longer needed.
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 SF_IMMUTABLE flag to prevent writing. Instead put in explicit
checking for the SF_SNAPSHOT flag in the appropriate places. With
this change, it is now possible to rename and link to snapshot files.
It is also possible to set or clear any of the owner, group, or
other read bits on the file, though none of the write or execute
bits can be set. There is also an explicit test to prevent the
setting or clearing of the SF_SNAPSHOT flag via chflags() or
fchflags(). Note also that the modify time cannot be changed as
it needs to accurately reflect the time that the snapshot was taken.
Submitted by: Robert Watson <rwatson@FreeBSD.org>
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).
SYSCTL_LONG macro to be consistent with other integer sysctl variables
and require an initial value instead of assuming 0. Update several
sysctl variables to use the unsigned types.
PR: 15251
Submitted by: Kelly Yancey <kbyanc@posi.net>
set equal to the number of kilobytes in your cache. The old options are
still supported for backwards compatibility.
Submitted by: Kelly Yancey <kbyanc@posi.net>
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
<sys/bio.h>.
<sys/bio.h> is now a prerequisite for <sys/buf.h> but it shall
not be made a nested include according to bdes teachings on the
subject of nested includes.
Diskdrivers and similar stuff below specfs::strategy() should no
longer need to include <sys/buf.> unless they need caching of data.
Still a few bogus uses of struct buf to track down.
Repocopy by: peter
a struct buf. Don't try to examine B_ASYNC, it is a layering violation
to do so. The only current user of this interface is vn(4) which, since
it emulates a disk interface, operates on struct bio already.
shared memory objects are regular files; the shm_open(3) routine
uses fcntl(2) to set a flag on the descriptor which tells mmap(2)
to automatically apply MAP_NOSYNC.
Not objected to by: bde, dillon, dufault, jasone
Exceptions:
Vinum untouched. This means that it cannot be compiled.
Greg Lehey is on the case.
CCD not converted yet, casts to struct buf (still safe)
atapi-cd casts to struct buf to examine B_PHYS
(Much of this done by script)
Move B_ORDERED flag to b_ioflags and call it BIO_ORDERED.
Move b_pblkno and b_iodone_chain to struct bio while we transition, they
will be obsoleted once bio structs chain/stack.
Add bio_queue field for struct bio aware disksort.
Address a lot of stylistic issues brought up by bde.
substitute BUF_WRITE(foo) for VOP_BWRITE(foo->b_vp, foo)
substitute BUF_STRATEGY(foo) for VOP_STRATEGY(foo->b_vp, foo)
This patch is machine generated except for the ccd.c and buf.h parts.
field in struct buf: b_iocmd. The b_iocmd is enforced to have
exactly one bit set.
B_WRITE was bogusly defined as zero giving rise to obvious coding
mistakes.
Also eliminate the redundant struct buf flag B_CALL, it can just
as efficiently be done by comparing b_iodone to NULL.
Should you get a panic or drop into the debugger, complaining about
"b_iocmd", don't continue. It is likely to write on your disk
where it should have been reading.
This change is a step in the direction towards a stackable BIO capability.
A lot of this patch were machine generated (Thanks to style(9) compliance!)
Vinum users: Greg has not had time to test this yet, be careful.
This
This feature allows you to specify if mmap'd data is included in
an application's corefile.
Change the type of eflags in struct vm_map_entry from u_char to
vm_eflags_t (an unsigned int).
Reviewed by: dillon,jdp,alfred
Approved by: jkh
run out of KVM through a mmap()/fork() bomb that allocates hundreds
of thousands of vm_map_entry structures.
Add panic to make null-pointer dereference crash a little more verbose.
Add a new sysctl, vm.max_proc_mmap, which specifies the maximum number
of mmap()'d spaces (discrete vm_map_entry's in the process). The value
defaults to around 9000 for a 128MB machine. The test is scaled for the
number of processes sharing a vmspace (aka linux threads). Setting
the value to 0 disables the feature.
PR: kern/16573
Approved by: jkh
also broke diskless swapping. Moving the swapdev_vp initialization
to more commonly run code solves the problem.
PR: kern/16165
Additional testing by: David Gilbert <dgilbert@velocet.ca>
invalidation code cannot wait for paging to complete while holding a
vnode lock, so we don't wait. Instead we simply allow the lower level
code to simply block on any busy pages it encounters. I think Yahoo
may be the only entity in the entire world that actually uses this
msync feature :-).
Bug reported by: Paul Saab <paul@mu.org>
This is necessary for vmware: it does not use an anonymous mmap for
the memory of the virtual system. In stead it creates a temp file an
unlinks it. For a 50 MB file, this results in a ot of syncing
every 30 seconds.
Reviewed by: Matthew Dillon <dillon@backplane.com>
is an application space macro and the applications are supposed to be free
to use it as they please (but cannot). This is consistant with the other
BSD's who made this change quite some time ago. More commits to come.
swap_pager.c and related commits.
Essentially swap_pager.c is backed out to before the changes, but
swapdev_vp is converted into a real vnode with just VOP_STRATEGY().
It no longer abuses specfs vnops and no longer needs a dev_t and
/dev/drum (or /dev/swapdev) for the intermediate layer.
This essentially restores the vnode interface as the interface to the
bottom of the swap pager, and vm_swap.c provides a clean vnode interface.
This will need to be revisited when we swap to files (vnodes) - which
is the other reason for keeping the vnode interface between the swap pager
and the swap devices.
OK'ed by: dillon
madvise().
This feature prevents the update daemon from gratuitously flushing
dirty pages associated with a mapped file-backed region of memory. The
system pager will still page the memory as necessary and the VM system
will still be fully coherent with the filesystem. Modifications made
by other means to the same area of memory, for example by write(), are
unaffected. The feature works on a page-granularity basis.
MAP_NOSYNC allows one to use mmap() to share memory between processes
without incuring any significant filesystem overhead, putting it in
the same performance category as SysV Shared memory and anonymous memory.
Reviewed by: julian, alc, dg
* 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
from vm_map_pageable(). At the point they called, vm_map_pageable()
holds a read (or shared) lock on the map. The purpose
of vm_map_{clear,set}_recursive() is to disable/enable repeated
write (or exclusive) lock requests by the same process.
vm_map always failed because vm_map_lookup() looked at
"vm_map_entry->wired_count" instead of "(vm_map_entry->eflags &
MAP_ENTRY_USER_WIRED)". The effect was that many page
wiring operations by sysctl were (silently) failing.
multiplexed underlying swap devices (/dev/drum). The only thing it did
was to allow root to open /dev/drum, but not do anything with it.
Various utilities used to grovel around in here, but Matt has written
a much nicer (and clean) front-end to this for libkvm, and nothing uses
the old system any more.
The VM system was calling VOP_STRATEGY() on the vp of the first underlying
swap device (not the /dev/drum one, the first real device), and using
the VOP system to indirectly (and only) call swstrategy() to choose
an underlying device and enqueue it on that device. I have changed it
to avoid diverting through the VOP system and to call the only possible
target directly, saving a little bit of time and some complexity.
In all, nothing much changes, except some scaffolding to support the
roundabout way of calling swstrategy() is gone.
Matt gave me the ok to do this some time ago, and I apologize for taking
so long to get around to it.
instead of duplicating the code. (2) If a wired page is passed
to vm_page_free_toq, panic instead of printing a friendly warning.
(If we don't panic here, we'll just panic later in vm_page_unwire
obscuring the problem.)
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
"rw" argument, rather than hijacking B_{READ|WRITE}.
Fix two bugs (physio & cam) resulting by the confusion caused by this.
Submitted by: Tor.Egge@fast.no
Reviewed by: alc, ken (partly)
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.
hexdump -C < /dev/drum
by simply refusing to do I/O from userland.
a panic. I'm not sure we even need /dev/drum anymore, it seems
to have been broken for a long time thi
have been there in the first place. A GENERIC kernel shrinks almost 1k.
Add a slightly different safetybelt under nostop for tty drivers.
Add some missing FreeBSD tags
clustering issues (replacing code that used to be in
ufs/ufs/ufs_readwrite.c). vm_fault also now uses the new VM page counter
inlines.
This completes the changeover from vnode->v_lastr to vm_entry_t->v_lastr
for VM, and fp->f_nextread and fp->f_seqcount (which have been in the
tree for a while). Determination of the I/O strategy (sequential, random,
and so forth) is now handled on a descriptor-by-descriptor basis for
base I/O calls, and on a memory-region-by-memory-region and
process-by-process basis for VM faults.
Reviewed by: David Greenman <dg@root.com>, Alan Cox <alc@cs.rice.edu>
spaces which cross a segment boundry in the page table. pmap_kextract()
is not designed for access to the user space portion of the page
table and cannot handle the null-page-directory-entry case.
The fix is to have vm_fault_quick() return a success or failure which
is then used to avoid calling pmap_kextract().
syncs the entire underlying file rather then just the requested range,
resulting in huge inefficiencies when the VM system is articulated in
a certain way. The VOP_FSYNC was also found to massively reduce NFS
performance in certain cases.
Change MADV_DONTNEED and MADV_FREE to call vm_page_dontneed() instead
of vm_page_deactivate(). Using vm_page_deactivate() causes all
inactive and cache pages to be recycled before the dontneed/free page
is recycled, effectively flushing our entire VM inactive & cache
queues continuously even if only a few pages are being actively MADV
free'd and reused (such as occurs with a sequential scan of a
memory-mapped file).
Reviewed by: Alan Cox <alc@cs.rice.edu>, David Greenman <dg@root.com>
from the vnode. (The changeover is undergoing final testing and
will be committed soon).
Reviewed by: Alan Cox <alc@cs.rice.edu>, David Greenman <dg@root.com>
underlying physical sector size when aligning I/O transfer sizes.
It cannot assume 512 bytes.
We assume the underlying sector size is a power of 2. If it isn't,
mmap() will break badly anyway (in the same way mmap broke with NFS
when NFS tried to cache piecemeal write ranges in buffers, before
we enforced read-buffer-before-write-piecemeal for NFS).
Reviewed by: Alan Cox <alc@cs.rice.edu>, David Greenman <dg@root.com>
Swap space can be freed from an interrupt and so swap reservation and
freeing must occur at splvm.
Add swap_pager_reserve() code to support a new swap pre-reservation
capability for the VN device.
Generally cleanup the swap code by simplifying the swp_pager_meta_build()
static function and consolidating the SWAPBLK_NONE test from a bit test
to an absolute compare. The bit test was left over from a rejected
swap allocation scheme that was not ultimately committed. A few other
minor cleanups were also made.
Reorganize the swap strategy code, again for VN support, to not
reallocate swap when writing as this messes up pre-reservation and
can fragment I/O unnecessarily as VN-baesd disk is messed around with.
Reviewed by: Alan Cox <alc@cs.rice.edu>, David Greenman <dg@root.com>
current process from the exclusive lock prior to initiating I/O.
This fixes a panic related to swap-backed VN disks
Reviewed by: Alan Cox <alc@cs.rice.edu>, David Greenman <dg@root.com>
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.
Make the alias list a SLIST.
Drop the "fast recycling" optimization of vnodes (including
the returning of a prexisting but stale vnode from checkalias).
It doesn't buy us anything now that we don't hardlimit
vnodes anymore.
Rename checkalias2() and checkalias() to addalias() and
addaliasu() - which takes dev_t and udev_t arg respectively.
Make the revoke syscalls use vcount() instead of VALIASED.
Remove VALIASED flag, we don't need it now and it is faster
to traverse the much shorter lists than to maintain the
flag.
vfs_mountedon() can check the dev_t directly, all the vnodes
point to the same one.
Print the devicename in specfs/vprint().
Remove a couple of stale LFS vnode flags.
Remove unimplemented/unused LK_DRAINED;
creation of /dev/drum via calling swapon. However, the make_dev has a
bogus (insofar that it hasn't been added yet) cdevsw, so later we end
up crashing with a null pointer dereference on the swap vp's specinfo.
The specinfo points to a dev_t with a major of 254 (uninitialized), and
we get a crash on its d_strategy being called.
The simple solution to this is to call cdevsw_add before the make_dev
is ever used. This fixes the panic which occurred upon swapping.
Diskslice/label code not yet handled.
Vinum, i4b, alpha, pc98 not dealt with (left to respective Maintainers)
Add the correct hook for devfs to kern_conf.c
The net result of this excercise is that a lot less files depends on DEVFS,
and devtoname() gets more sensible output in many cases.
A few drivers had minor additional cleanups performed relating to cdevsw
registration.
A few drivers don't register a cdevsw{} anymore, but only use make_dev().
The lock structure cannot be the first element of the vm_map
because this can result in livelock between two or more system
processes trying to kmem_alloc_wait.
Remove semicolons or add "do { } while (0)" as necessary
to enable the use of these macros in arbitrary statements.
(There are no functional changes.)
Submitted by: dillon
A complete rewrite by dillon and myself to separate
the implementation of behaviors that effect the vm_map_entry
from those that effect the vm_object.
A result of this change is that madvise(..., MADV_FREE);
is much cheaper.
This setting is also acceptable for Celerons and Pentium Pros
with less than 1MB L2 caches.
Note: PQ_L2_SIZE is a misnomer. The correct number of colors is
a function of the cache's degree of associativity as well as its size.
Submitted by: bde and alc
Now that behaviors are stored in the vm_map_entry rather than
the vm_object, it's no longer necessary to instantiate a vm_object
just to hold the behavior.
Reviewed by: dillon
Remove the initialization of PQ_NONE's cnt and lcnt. They aren't
used.
vm_page_insert:
Remove an unnecessary dereference.
vm_page_wire:
Remove the one and only (and thus pointless) reference
to PQ_NONE's lcnt.
When creating new processes (or performing exec), the new page
directory is initialized too early. The kernel might grow before
p_vmspace is initialized for the new process. Since pmap_growkernel
doesn't yet know about the new page directory, it isn't updated, and
subsequent use causes a failure.
The fix is (1) to clear p_vmspace early, to stop pmap_growkernel
from stomping on memory, and (2) to defer part of the initialization
of new page directories until p_vmspace is initialized.
PR: kern/12378
Submitted by: tegge
Reviewed by: dfr
vm_map.c:
Don't set OBJ_ONEMAPPING on arbitrary vm objects. Only default
and swap type vm objects should have it set. vm_object_deallocate
already handles these cases.
vm_object.c:
If OBJ_ONEMAPPING isn't already clear in vm_object_shadow,
we are in trouble. Instead of clearing it, make it
an assertion that it is already clear.
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>
SYSINIT_KT() etc (which is a static, compile-time procedure), use a
NetBSD-style kthread_create() interface. kproc_start is still available
as a SYSINIT() hook. This allowed simplification of chunks of the
sysinit code in the process. This kthread_create() is our old kproc_start
internals, with the SYSINIT_KT fork hooks grafted in and tweaked to work
the same as the NetBSD one.
One thing I'd like to do shortly is get rid of nfsiod as a user initiated
process. It makes sense for the nfs client code to create them on the
fly as needed up to a user settable limit. This means that nfsiod
doesn't need to be in /sbin and is always "available". This is a fair bit
easier to do outside of the SYSINIT_KT() framework.
lockmgr locks. This commit should be functionally equivalent to the old
semantics. That is, all buffer locking is done with LK_EXCLUSIVE
requests. Changes to take advantage of LK_SHARED and LK_RECURSIVE will
be done in future commits.
creating a new entry. vm_map_stack and vm_map_growstack can panic when
a new entry isn't created. Fixed vm_map_stack and vm_map_growstack.
Also, when extending the stack, always set the protection to VM_PROT_ALL.
Insure that device mappings get MAP_PREFAULT(_PARTIAL) set,
so that 4M page mappings are used when possible.
Reviewed by: Luoqi Chen <luoqi@watermarkgroup.com>
The cdevsw_add() function now finds the major number(s) in the
struct cdevsw passed to it. cdevsw_add_generic() is no longer
needed, cdevsw_add() does the same thing.
cdevsw_add() will print an message if the d_maj field looks bogus.
Remove nblkdev and nchrdev variables. Most places they were used
bogusly. Instead check a dev_t for validity by seeing if devsw()
or bdevsw() returns NULL.
Move bdevsw() and devsw() functions to kern/kern_conf.c
Bump __FreeBSD_version to 400006
This commit removes:
72 bogus makedev() calls
26 bogus SYSINIT functions
if_xe.c bogusly accessed cdevsw[], author/maintainer please fix.
I4b and vinum not changed. Patches emailed to authors. LINT
probably broken until they catch up.
Reformat and initialize correctly all "struct cdevsw".
Initialize the d_maj and d_bmaj fields.
The d_reset field was not removed, although it is never used.
I used a program to do most of this, so all the files now use the
same consistent format. Please keep it that way.
Vinum and i4b not modified, patches emailed to respective authors.
Remove a useless argument from vm_map_madvise's interface (vm_map.c,
vm_map.h, and vm_mmap.c).
Remove a redundant test in vm_uiomove (vm_map.c).
Make two changes to vm_object_coalesce:
1. Determine whether the new range of pages actually overlaps
the existing object's range of pages before calling vm_object_page_remove.
(Prior to this change almost 90% of the calls to vm_object_page_remove
were to remove pages that were beyond the end of the object.)
2. Free any swap space allocated to removed pages.
It never makes sense to specify MAP_COPY_NEEDED without also specifying
MAP_COPY_ON_WRITE, and vice versa. Thus, MAP_COPY_ON_WRITE suffices.
Reviewed by: David Greenman <dg@root.com>
udev_t in the kernel but still called dev_t in userland.
Provide functions to manipulate both types:
major() umajor()
minor() uminor()
makedev() umakedev()
dev2udev() udev2dev()
For now they're functions, they will become in-line functions
after one of the next two steps in this process.
Return major/minor/makedev to macro-hood for userland.
Register a name in cdevsw[] for the "filedescriptor" driver.
In the kernel the udev_t appears in places where we have the
major/minor number combination, (ie: a potential device: we
may not have the driver nor the device), like in inodes, vattr,
cdevsw registration and so on, whereas the dev_t appears where
we carry around a reference to a actual device.
In the future the cdevsw and the aliased-from vnode will be hung
directly from the dev_t, along with up to two softc pointers for
the device driver and a few houskeeping bits. This will essentially
replace the current "alias" check code (same buck, bigger bang).
A little stunt has been provided to try to catch places where the
wrong type is being used (dev_t vs udev_t), if you see something
not working, #undef DEVT_FASCIST in kern/kern_conf.c and see if
it makes a difference. If it does, please try to track it down
(many hands make light work) or at least try to reproduce it
as simply as possible, and describe how to do that.
Without DEVT_FASCIST I belive this patch is a no-op.
Stylistic/posixoid comments about the userland view of the <sys/*.h>
files welcome now, from userland they now contain the end result.
Next planned step: make all dev_t's refer to the same devsw[] which
means convert BLK's to CHR's at the perimeter of the vnodes and
other places where they enter the game (bootdev, mknod, sysctl).
Made a new (inline) function devsw(dev_t dev) and substituted it.
Changed to the BDEV variant to this format as well: bdevsw(dev_t dev)
DEVFS will eventually benefit from this change too.
Virtualize bdevsw[] from cdevsw. bdevsw() is now an (inline)
function.
Join CDEV_MODULE and BDEV_MODULE to DEV_MODULE (please pay attention
to the order of the cmaj/bmaj arguments!)
Join CDEV_DRIVER_MODULE and BDEV_DRIVER_MODULE to DEV_DRIVER_MODULE
(ditto!)
(Next step will be to convert all bdev dev_t's to cdev dev_t's
before they get to do any damage^H^H^H^H^H^Hwork in the kernel.)
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1:
s/suser/suser_xxx/
2:
Add new function: suser(struct proc *), prototyped in <sys/proc.h>.
3:
s/suser_xxx(\([a-zA-Z0-9_]*\)->p_ucred, \&\1->p_acflag)/suser(\1)/
The remaining suser_xxx() calls will be scrutinized and dealt with
later.
There may be some unneeded #include <sys/cred.h>, but they are left
as an exercise for Bruce.
More changes to the suser() API will come along with the "jail" code.
unallocated parts of the last page when the file ended on a frag
but not a page boundary.
Delimitted by tags PRE_MATT_MMAP_EOF and POST_MATT_MMAP_EOF,
in files alpha/alpha/pmap.c i386/i386/pmap.c nfs/nfs_bio.c vm/pmap.h
vm/vm_page.c vm/vm_page.h vm/vnode_pager.c miscfs/specfs/spec_vnops.c
ufs/ufs/ufs_readwrite.c kern/vfs_bio.c
Submitted by: Matt Dillon <dillon@freebsd.org>
Reviewed by: Alan Cox <alc@freebsd.org>
1. Don't bother checking object->ref_count == 1 in order to set
OBJ_ONEMAPPING. It's a waste of time. If object->ref_count == 1,
vm_map_entry_delete will "run-down" the object and its pages.
2. If object->ref_count == 1, ignore OBJ_ONEMAPPING. Wait for
vm_map_entry_delete to "run-down" the object and its pages.
Otherwise, we're calling two different procedures to delete
the object's pages.
Note: "vmstat -s" will once again show a non-zero value
for "pages freed by exiting processes".
Remove more (redundant) map timestamp increments from properly
synchronized routines. (Changed: vm_map_entry_link, vm_map_entry_unlink,
and vm_map_pageable.)
Micro-optimize vm_map_entry_link and vm_map_entry_unlink, eliminating
unnecessary dereferences. At the same time, converted them from macros
to inline functions.
address) so that the first 16MB of physical memory is allocated
last rather than first. On large-memory machines, this avoids
the exhaustion of low physical memory before isa_dmainit has run.
block (VM_WAIT) holding the map lock. This is bad. For example, a subsequent
kmem_malloc by an interrupt handler on the same map may find the lock held
and panic in the lockmgr.
In general, vm_map_simplify_entry should be performed INSIDE
the loop that traverses the map, not outside. (Changed:
vm_map_inherit, vm_map_pageable.)
vm_fault_unwire doesn't acquire the map lock (or block holding
it). Thus, vm_map_set/clear_recursive shouldn't be called.
(Changed: vm_map_user_pageable, vm_map_pageable.)
The old VN device broke in -4.x when the definition of B_PAGING
changed. This patch fixes this plus implements additional capabilities.
The new VN device can be backed by a file ( as per normal ), or it can
be directly backed by swap.
Due to dependencies in VM include files (on opt_xxx options) the new
vn device cannot be a module yet. This will be fixed in a later commit.
This commit delimitted by tags {PRE,POST}_MATT_VNDEV
1. The size of vm_object::memq is vm_object::resident_page_count,
not vm_object::size.
2. The "size > 4" test sometimes results in the traversal of a ~1000 page
memq in order to locate ~10 pages.
lock) until it actually needs to modify the vm_map.
Note: it is legal to modify vm_map::hint without holding a write lock.
Submitted by: "Richard Seaman, Jr." <dick@tar.com> with minor changes
by myself.
the read lock around the subyte operations in mincore. After the lock is
reacquired, use the map's timestamp to determine if we need to restart
the scan.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
To prevent a deadlock, if we are extremely low on memory, force synchronous
operation by the VOP_PUTPAGES in vnode_pager_putpages.
Fix bug where an object's OBJ_WRITEABLE/OBJ_MIGHTBEDIRTY flags do
not get set under certain circumstances ( page rename case ).
Reviewed by: Alan Cox <alc@cs.rice.edu>, John Dyson
is the preparation step for moving pmap storage out of vmspace proper.
Reviewed by: Alan Cox <alc@cs.rice.edu>
Matthew Dillion <dillon@apollo.backplane.com>
be in progress at any given moment.
Add two swap tuneables to sysctl:
vm.swap_async_max: 4
vm.swap_cluster_max: 16
Recommended values are a cluster size of 8 or 16 pages. async_max is
about right for 1-4 swap devices. Reduce to 2 if swap is eating too much
bandwidth, or even 1 if swap is both eating too much bandwidth and sitting
on a slow network (10BaseT).
The defaults work well across a broad range of configurations and should
normally be left alone.
Unlock vnode before messing with map to avoid deadlock between map and
vnode ( e.g. with exec_map and underlying program binary vnode ). Solves
a deadlock that most often occurs during a large -j# buildworld reported
by three people.
been made but the code has been reorganized and documented to make
it more readable, reduce the size of the code, and optimize the branch
path caching capabilities that most modern processors have.
free swap space out from under a busy page. This is not legal because
the swap may be reallocated and I/O issued while I/O is still in
progress on the same swap page from the madvise()'d object. This bug
could only occur under extreme paging conditions but might not cause
an error until much later. As a side-benefit, madvise() is now even
smaller.
possible without actually unmapping it from the process.
As of now, I declare madvise() on OBJT_DEFAULT/OBJT_SWAP objects to be
'working and complete'.
OBJ_ONEMAPPING in the case where an object is extended by an
additional vm_map_entry must be allocated.
In vm_object_madvise(), remove calll to vm_page_cache() in MADV_FREE
case in order to avoid a page fault on page reuse. However, we still
mark the page as clean and destroy any swap backing store.
Submitted by: Alan Cox <alc@cs.rice.edu>
because there was a concensus on current in regards to leaving bss r+w+x
instead of r+w. This is in order to maintain reasonable compatibility
with existing JIT compilers (e.g. kaffe) and possibly other programs.
no major operational changes were made. The three core object->memq loops
were moved into a single inline procedure and various operational
characteristics of the collapse function were documented.
PQ_FREE. There is little operational difference other then the kernel
being a few kilobytes smaller and the code being more readable.
* vm_page_select_free() has been *greatly* simplified.
* The PQ_ZERO page queue and supporting structures have been removed
* vm_page_zero_idle() revamped (see below)
PG_ZERO setting and clearing has been migrated from vm_page_alloc()
to vm_page_free[_zero]() and will eventually be guarenteed to remain
tracked throughout a page's life ( if it isn't already ).
When a page is freed, PG_ZERO pages are appended to the appropriate
tailq in the PQ_FREE queue while non-PG_ZERO pages are prepended.
When locating a new free page, PG_ZERO selection operates from within
vm_page_list_find() ( get page from end of queue instead of beginning
of queue ) and then only occurs in the nominal critical path case. If
the nominal case misses, both normal and zero-page allocation devolves
into the same _vm_page_list_find() select code without any specific
zero-page optimizations.
Additionally, vm_page_zero_idle() has been revamped. Hysteresis has been
added and zero-page tracking adjusted to conform with the other changes.
Currently hysteresis is set at 1/3 (lo) and 1/2 (hi) the number of free
pages. We may wish to increase both parameters as time permits. The
hysteresis is designed to avoid silly zeroing in borderline allocation/free
situations.
attempt to optimize forks but were essentially given-up on due to
problems and replaced with an explicit dup of the vm_map_entry structure.
Prior to the removal, they were entirely unused.
rather then VM_PROT_ALL. obreak, on the otherhand, uses VM_PROT_ALL.
This prevents vm_map_insert() from being able to coalesce the heap and
creates an extra map entry. Since current architectures ignore
VM_PROT_EXECUTE anyway, and since not having VM_PROT_EXECUTE on data/bss
may provide protection in the future, obreak now uses read+write rather
then all (r+w+x).
This is an optimization, not a bug fix.
Submitted by: Alan Cox <alc@cs.rice.edu>
Since paging is in progress, page scan in vm_page_qcollapse() must be
protected at atleast splbio() to prevent pages from being ripped out from
under the scan.
The vm_map_insert()/vm_object_coalesce() optimization has been extended
to include OBJT_SWAP objects as well as OBJT_DEFAULT objects. This is
possible because it costs nothing to extend an OBJT_SWAP object with
the new swapper. We can't do this with the old swapper. The old swapper
used a linear array that would have had to have been reallocated, costing
time as well as a potential low-memory deadlock.
in vm_map_simplify_entry. Basically, once you've verified that
the objects in the adjacent vm_map_entry's are the same, either
NULL or the same vm_object, there's no point in checking that the
objects have the same behavior.
Obtained from: Alan Cox <alc@cs.rice.edu>
Checked by: "Richard Seaman, Jr." <dick@tar.com>
Fix the following problem:
As the code stands now, growing any stack, and not just the process's
main stack, modifies vm->vm_ssize. This is inconsistent with the code
earlier in the same procedure.
This changes the definitions of a few items so that structures are the
same whether or not the option itself is enabled. This allows
people to enable and disable the option without recompilng the world.
As the author says:
|I ran into a problem pulling out the VM_STACK option. I was aware of this
|when I first did the work, but then forgot about it. The VM_STACK stuff
|has some code changes in the i386 branch. There need to be corresponding
|changes in the alpha branch before it can come out completely.
what is done:
|
|1) Pull the VM_STACK option out of the header files it appears in. This
|really shouldn't affect anything that executes with or without the rest
|of the VM_STACK patches. The vm_map_entry will then always have one
|extra element (avail_ssize). It just won't be used if the VM_STACK
|option is not turned on.
|
|I've also pulled the option out of vm_map.c. This shouldn't harm anything,
|since the routines that are enabled as a result are not called unless
|the VM_STACK option is enabled elsewhere.
|
|2) Add what appears to be appropriate code the the alpha branch, still
|protected behind the VM_STACK switch. I don't have an alpha machine,
|so we would need to get some testers with alpha machines to try it out.
|
|Once there is some testing, we can consider making the change permanent
|for both i386 and alpha.
|
[..]
|
|Once the alpha code is adequately tested, we can pull VM_STACK out
|everywhere.
|
Submitted by: "Richard Seaman, Jr." <dick@tar.com>
This takes the conditionals out of the code that has been tested by
various people for a while.
ps and friends (libkvm) will need a recompile as some proc structure
changes are made.
Submitted by: "Richard Seaman, Jr." <dick@tar.com>
vm_page_rename(), but never pulled the page off PQ_CACHE if it was on
PQ_CACHE. Dirty pages in PQ_CACHE are not allowed and a KASSERT was
added in -4.x to test for this... and got hit.
In -4.x, vm_page_rename() automatically dirties the page. This commit
also has it deal with the PQ_CACHE case, deactivating the page in that
case.
values. The 'int' return value for the procedure was never used and
not well defined in any case when there are mixed errors on pages, so
it has been removed. vm_pager_put_pages() and associated vm_pager
functions now return void.
swap blocks are now in PAGE_SIZE'd increments instead of DEV_BSIZE'd
increments. We still convert to DEV_BSIZE'd increments for the
backing store I/O, but everything else is in PAGE_SIZE increments.
vm_pager.h
Added argument to getpbuf() and relpbuf() to allow each subsystem to
specify a different hard limit on the number of simultanious physical
bufferes that said subsystem may allocate. Without this feature, one
subsystem ( e.g. the vfs clustering code ) could hog *ALL* the pbufs,
causing a deadlock in the pager in a low memory situation.
Same for trypbuf().
Removed call to vm_object_collapse(), which can block. This was being
called without the pageout code holding any sort of reference on the
vm_object or vm_page_t structures being manipulated. Since this code
can block, it was possible for other kernel code to shred the state
the pageout code was assuming remained intact.
Fixed potential blocking condition in vm_pageout_page_free() ( which
could cause a deadlock in a low-memory situation ).
Currently there is a hack in-place to deal with clean filesystem meta-data
polluting the inactive page queue. John doesn't like the hack, and neither
do I.
Revamped and commented a portion of the pageout loop.
Added protection against potential memory deadlocks with OBJT_VNODE
when using VOP_ISLOCKED(). The problem is that vp->v_data can be NULL
which causes VOP_ISLOCKED() to return a less informed answer.
remove vm_pager_sync() -- none of the pagers use it any more ( the old
swapper used to. The new one does not ).
reducing the size of vm_page_t.
SWAPBLK_NONE and SWAPBLK_MASK are defined here. These actually are
more generalized then their names imply, but their placement is somewhat
of a legacy issue from a prior test version of this code that put
the swapblk in the vm_page_t structure. That test code was eventually
thrown away. The legacy remains.
Added vm_page_flash() inline. Similar to vm_page_wakeup() except that
it does not clear PG_BUSY ( one assumes that PG_BUSY is already clear ).
Used by a number of routines to wakeup waiters.
Collapsed some of the code in inline calls to make other inline calls.
GCC will optimize this well and it reduces duplication.
vm_page_free() and vm_page_free_zero() inlines added to convert to
the proper vm_page_free_toq() call.
vm_page_sleep_busy() inline added, replacing vm_page_sleep() ( which has
been removed ). This implements a much more optimizable page-waiting
function.
pointers per entry ). The table has been changed to a singly linked
list of vm_page_t pointers. The table has been doubled in size, but
the entries only take half the space so a net-zero change in memory use.
The hash function has been changed, hopefully for the better. The
combination of the larger hash table size of changed function should
keep the chain length down to a reasonable number (0-3, average 1).
vm_object->page_hint has been removed. This 'optimization' was not
only never needed, but costs as much as a hash chain link to implement.
While having page_hint in vm_object might result in better locality
of reference, the cost is not worth the space in vm_object or the
extra instructions in my view.
vm_page_alloc*() functions have been inlined and call a generalized
non-inlined vm_page_alloc_toq() which combines the standard alloc
and zero-page alloc functions together, reducing code size and the L1
cache footprint. Some reordering has been done... not much. The
delinking code should be faster ( because unlinking a doubly-linked list
requires four memory ops and unlinking a singly linked list only requires
two ), and we get a hash consistancy check for free.
vm_page_rename() now automatically sets the page's dirty bits.
vm_page_alloc() does not try to manually inline freeing a cache page.
Instead, it now properly calls vm_page_free(m) ... vm_page_free() is
really too complex to manually inline.
vm_await(), supporting asleep(), has been added.
of most of the swap-pager-specific fields, the removal of the id,
and the removal of paging_offset.
A new inline, vm_object_pip_wakeupn() has been added to subtract an
arbitrary number n from the paging_in_progress count and then wakeup
waiters as necessary. n may be 0, resulting in a 'flash'.
object->paging_offset has been removed - it was used to optimize a
single OBJT_SWAP collapse case yet introduced massive confusion throughout
vm_object.c. The optimization was inconsequential except for the
claim that it didn't have to allocate any memory. The optimization
has been removed.
madvise() has been fixed. The old madvise() could be made to operate
on shared objects which is a big no-no. The new one is much more careful
in what it modifies. MADV_FREE was totally broken and has now been fixed.
vm_page_rename() now automatically dirties a page, so explicit dirtying
of the page prior to calling vm_page_rename() has been removed.
about conversions of objects to OBJT_SWAP, it is done automatically
now.
Replaced manually inserted code with inline calls for busy waiting on
pages, which also incidently fixes a potential PG_BUSY race due to
the code not running at splvm().
vm_objects no longer have a paging_offset field ( see vm/vm_object.c )
instead to properly handle any waiters.
Added comments, added support for M_ASLEEP. Generally treat M_ flags
as flags instead of constants to compare against.
and the swap_pager has been completely replaced.
The new swap pager uses the new blist radix-tree based bitmap allocator
for low level swap allocation and deallocation. The new allocator
is effectively O(5) while the old one was O(N), and the new allocator
allocates all required memory at init time rather then at allocate
memory on the fly at run time.
Swap metadata is allocated in clusters and stored in a hash table,
eliminating linearly allocated structures.
Many, many features have been rewritten or added. Swap space is now
reallocated on the fly providing a poor-mans auto defragmentation of
swap space. Swap space that is no longer needed is freed on a timely
basis so no garbage collection is necessary.
Swap I/O is marked B_ASYNC and NFS has been fixed to do the right
thing with it, so NFS-based paging now has around 10x the performance
as it did before ( previously NFS enforced synchronous I/O for paging ).
changes to the VM system to support the new swapper, VM bug
fixes, several VM optimizations, and some additional revamping of the
VM code. The specific bug fixes will be documented with additional
forced commits. This commit is somewhat rough in regards to code
cleanup issues.
Reviewed by: "John S. Dyson" <root@dyson.iquest.net>, "David Greenman" <dg@root.com>
shared signal handling when there is shared signal handling being
used.
This removes the main objection to making the shared signal handling
a standard ability in rfork() and friends and 'unconditionalising'
this code. (i.e. the allocation of an extra 328 bytes per process).
Signal handling information remains in the U area until such a time as
it's reference count would be incremented to > 1. At that point a new
struct is malloc'd and maintained in KVM so that it can be shared between
the processes (threads) using it.
A function to check the reference count and move the struct back to the U
area when it drops back to 1 is also supplied. Signal information is
therefore now swapable for all processes that are not sharing that
information with other processes. THis should addres the concerns raised
by Garrett and others.
Submitted by: "Richard Seaman, Jr." <dick@tar.com>
downward growing stacks more general.
Add (but don't activate) code to use the new stack facility
when running threads, (specifically the linux threads support).
This allows people to use both linux compiled linuxthreads, and also the
native FreeBSD linux-threads port.
The code is conditional on VM_STACK. Not using this will
produce the old heavily tested system.
Submitted by: Richard Seaman <dick@tar.com>
"dying daemons" problem. (I thought this code was introduced in rev.1.80,
but it just relaxed the condition.)
Also, kill related "suggest more swap space" warning (also introduced in
1.80). It was confusing, to say the least...
Requested by: msmith
Not objected by: dg
Submitted by: "Richard Seaman, Jr." <lists@tar.com>
Obtained from: linux :-)
Code to allow Linux Threads to run under FreeBSD.
By default not enabled
This code is dependent on the conditional
COMPAT_LINUX_THREADS (suggested by Garret)
This is not yet a 'real' option but will be within some number of hours.
for possible buffer overflow problems. Replaced most sprintf()'s
with snprintf(); for others cases, added terminating NUL bytes where
appropriate, replaced constants like "16" with sizeof(), etc.
These changes include several bug fixes, but most changes are for
maintainability's sake. Any instance where it wasn't "immediately
obvious" that a buffer overflow could not occur was made safer.
Reviewed by: Bruce Evans <bde@zeta.org.au>
Reviewed by: Matthew Dillon <dillon@apollo.backplane.com>
Reviewed by: Mike Spengler <mks@networkcs.com>
almost always causes this panic for the curproc != pageproc case.
This case apparently doesn't happen in normal operation, but it
happens when vm_page_alloc_contig() is called when there is a memory
hogging application that hasn't already been paged out.
PR: 8632
Reviewed by: info@opensound.com (Dev Mazumdar), dg
Broken in: rev.1.89 (1998/02/23)
truncated to 32 bits.
* Change the calling convention of the device mmap entry point to
pass a vm_offset_t instead of an int for the offset allowing
devices with a larger memory map than (1<<32) to be supported
on the alpha (/dev/mem is one such).
These changes are required to allow the X server to mmap the various
I/O regions used for device port and memory access on the alpha.
file to a stream socket. sendfile(2) is similar to implementations in
HP-UX, Linux, and other systems, but the API is more extensive and
addresses many of the complaints that the Apache Group and others have
had with those other implementations. Thanks to Marc Slemko of the
Apache Group for helping me work out the best API for this.
Anyway, this has the "net" result of speeding up sends of files over
TCP/IP sockets by about 10X (that is to say, uses 1/10th of the CPU
cycles) when compared to a traditional read/write loop.
when bdevsw[] became sparse. We still depend on magic to avoid having to
check that (v_rdev) device numbers in vnodes are not NODEV.
Removed a redundant `major(dev) < nblkdev' test instead of updating it.
Don't follow a garbage bdevsw pointer for attempts to swap on empty
regular files. This case currently can't happen. Swapping on regular
files is ifdefed out in swapon() and isn't attempted for empty files
in nfs_mountroot().
needs to be called prior to freeing remaining pages in the object so that
the device pager has an opportunity to grab its "fake" pages. Also, in
the case of wired pages, the page must be made busy prior to calling
vm_page_remove. This is a difference from 2.2.x that I overlooked when
I brought these changes forward.
legitimately wired pages. Currently we print a diagnostic when this
happens, but this will be removed soon when it will be common for this
to occur with zero-copy TCP/IP buffers.
1) The vnode pager wasn't properly tracking the file size due to
"size" being page rounded in some cases and not in others.
This sometimes resulted in corrupted files. First noticed by
Terry Lambert.
Fixed by changing the "size" pager_alloc parameter to be a 64bit
byte value (as opposed to a 32bit page index) and changing the
pagers and their callers to deal with this properly.
2) Fixed a bogus type cast in round_page() and trunc_page() that
caused some 64bit offsets and sizes to be scrambled. Removing
the cast required adding casts at a few dozen callers.
There may be problems with other bogus casts in close-by
macros. A quick check seemed to indicate that those were okay,
however.
simple-lock.
The reviewer raises the following caveat: "I believe these changes
open a non-critical race condition when adding memory to the pool
for the zone. I think what will happen is that you could have two
threads that are simultaneously adding additional memory when the
pool runs out. This appears to not be a problem, however, since
the re-aquisition of the lock will protect the list pointers."
The submitter agrees that the race is non-critical, and points out
that it already existed for the non-SMP case. He suggests that
perhaps a sleep lock (using the lock manager) should be used to
close that race. This might be worth revisiting after 3.0 is
released.
Reviewed by: dg (David Greenman)
Submitted by: tegge (Tor Egge)
expected. This bug caused builds of Modula-3 to fail in mysterious
ways on SMP kernels. More precisely, such builds failed on systems
with kern.fast_vfork equal to 0, the default and only supported
value for SMP kernels.
PR: kern/7468
Submitted by: tegge (Tor Egge)
when nfs is an LKM. Declare it in a header file. Don't forget to use
it in non-Lite2 code. Initialize it to -1 instead of to 0, since 0
will soon be the mount type number for the first vfs loaded.
NetBSD uses strcmp() to avoid this ugly global.
Add some overflow checks to read/write (from bde).
Change all modifications to vm_page::flags, vm_page::busy, vm_object::flags
and vm_object::paging_in_progress to use operations which are not
interruptable.
Reviewed by: Bruce Evans <bde@zeta.org.au>
managed to avoid corruption of this variable by luck (the compiler used a
memory read-modify-write instruction which wasn't interruptable) but other
architectures cannot.
With this change, I am now able to 'make buildworld' on the alpha (sfx: the
crowd goes wild...)
code still left in there. The macros it describes disapeared some-
time since 4.4BSD lite.
PR: 7246
Reviewed by: phk
Submitted by: Stefan Eggers <seggers@semyam.dinoco.de>
respectively. Most of the longs should probably have been
u_longs, but this changes is just to prevent warnings about
casts between pointers and integers of different sizes, not
to fix poorly chosen types.
casting them to long, etc. Fixed some nearby printf bogons (sign
errors not warned about by gcc, and style bugs, but not truncation
of vm_ooffset_t's).
casting them to long, etc. Fixed some nearby printf bogons (sign
errors not warned about by gcc, and style bugs, but not truncation
of vm_ooffset_t's).
Use slightly less bogus casts for passing pointers to ddb command
functions.
There is only cdevsw (which should be renamed in a later edit to deventry
or something). cdevsw contains the union of what were in both bdevsw an
cdevsw entries. The bdevsw[] table stiff exists and is a second pointer
to the cdevsw entry of the device. it's major is in d_bmaj rather than
d_maj. some cleanup still to happen (e.g. dsopen now gets two pointers
to the same cdevsw struct instead of one to a bdevsw and one to a cdevsw).
rawread()/rawwrite() went away as part of this though it's not strictly
the same patch, just that it involves all the same lines in the drivers.
cdroms no longer have write() entries (they did have rawwrite (?)).
tapes no longer have support for bdev operations.
Reviewed by: Eivind Eklund and Mike Smith
Changes suggested by eivind.
as the value in b_vp is often not really what you want.
(and needs to be frobbed). more cleanups will follow this.
Reviewed by: Bruce Evans <bde@freebsd.org>
FreeBSD/alpha. The most significant item is to change the command
argument to ioctl functions from int to u_long. This change brings us
inline with various other BSD versions. Driver writers may like to
use (__FreeBSD_version == 300003) to detect this change.
The prototype FreeBSD/alpha machdep will follow in a couple of days
time.
printf() of "Out of mbuf clusters - adjust NMBCLUSTERS or increase
maxusers" so that the message is more informative and so that it will
appear in the kernel message buffer.
unexpectedly do not complete writes even with sync I/O requests.
This should help the behavior of mmaped files when using
softupdates (and perhaps in other circumstances also.)
the page offset. If a large file offset was passed in, a large negative
array index could be generated which could cause page faults etc at worst
and file corruption at the least. (Pages are allocated within file
space on page alignment boundaries, so a file offset being passed in here
is harmless to DTRT. The case where this was happening has already been
fixed though, this is in case it happens again).
Reviewed by: dyson
deallocation cycles. This should provide a measurable improvement
on swap and memory allocation on loaded systems. It is unlikely a
complete solution. Also, provide more map info with procfs.
Chuck Cranor spurred on this improvement.
"time" wasn't a atomic variable, so splfoo() protection were needed
around any access to it, unless you just wanted the seconds part.
Most uses of time.tv_sec now uses the new variable time_second instead.
gettime() changed to getmicrotime(0.
Remove a couple of unneeded splfoo() protections, the new getmicrotime()
is atomic, (until Bruce sets a breakpoint in it).
A couple of places needed random data, so use read_random() instead
of mucking about with time which isn't random.
Add a new nfs_curusec() function.
Mark a couple of bogosities involving the now disappeard time variable.
Update ffs_update() to avoid the weird "== &time" checks, by fixing the
one remaining call that passwd &time as args.
Change profiling in ncr.c to use ticks instead of time. Resolution is
the same.
Add new function "tvtohz()" to avoid the bogus "splfoo(), add time, call
hzto() which subtracts time" sequences.
Reviewed by: bde
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1) When freeing pages, it is a good idea to protect them off.
(This is probably gratuitious, but good form.)
2) Allow collapsing pages in the backing object that are
PQ_CACHE. This will improve memory utilization.
3) Correct the collapse code so that pages that were on the
cache queue are moved to the inactive queue. This is
done when pages are marked dirty (so that those pages
will be properly paged out instead of freed), so that
cached pages will not be paradoxically marked dirty.
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
put alot of it's context into a data structure. This allows
significant shortening of its codepath, and will significantly
decrease it's cache footprint.
Also, add some stats to vmmeter. Note that you'll have to
rebuild/recompile vmstat, systat, etc... Otherwise, you'll
get "very interesting" paging stats.
2) Do not unnecessarily force page blocking when paging
pages out.
3) Further improve swap pager performance and correctness,
including fixing the paging in progress deadlock (except
in severe I/O error conditions.)
4) Enable vfs_ioopt=1 as a default.
5) Fix and enable the page prezeroing in SMP mode.
All in all, SMP systems especially should show a significant
improvement in "snappyness."
These diffs implement the first stage of a VOP_{GET|PUT}PAGES pushdown
for local media FS's.
See ffs_putpages in /sys/ufs/ufs/ufs_readwrite.c for implementation
details for generic *_{get|put}pages for local media FS's. Support
is trivial to add for any FS that formerly relied on the default
behaviour of the vnode_pager in in EOPNOTSUPP cases (just copy the
ffs_getpages() code for the FS in question's *_{get|put}pages).
Obviously, it would be better if each local media FS implemented a
more optimal method, instead of calling an exported interface from
the /sys/vm/vnode_pager.c, but this is a necessary first step in
getting the FS's to a point where they can be supplied with better
implementations on a case-by-case basis.
Obviously, the cd9660_putpages() can be rather trivial (since it
is a read-only FS type 8-)).
A slight (temporary) modification is made to print a diagnostic message
in the case where the underlying filesystem attempts to engage in the
previous behaviour. Failure is likely to be ungraceful.
Submitted by: terry@freebsd.org (Terry Lambert)
improve tuning on larger systems. (A couple of the VM tuning params for
small systems were so badly chosen that the system could hang under load.)
The broken tuning was originaly my fault.
have declined due to code-rot over time. The swap pager rundown code
has been clean-up, and unneeded wakeups removed. Lots of splbio's
are changed to splvm's. Also, set the dynamic tunables for the
pageout daemon to be more sane for larger systems (thereby decreasing
the daemon overheadla.)
in a way identically as before.) I had problems with the system properly
handling the number of vnodes when there is alot of system memory, and the
default VM_KMEM_SIZE. Two new options "VM_KMEM_SIZE_SCALE" and
"VM_KMEM_SIZE_MAX" have been added to support better auto-sizing for systems
with greater than 128MB.
Add some accouting for vm_zone memory allocations, and provide properly
for vm_zone allocations out of the kmem_map. Also move the vm_zone
allocation stats to the VM OID tree from the KERN OID tree.
of the various ad-hoc schemes.
2) When bringing in UPAGES, the pmap code needs to do another vm_page_lookup.
3) When appropriate, set the PG_A or PG_M bits a-priori to both avoid some
processor errata, and to minimize redundant processor updating of page
tables.
4) Modify pmap_protect so that it can only remove permissions (as it
originally supported.) The additional capability is not needed.
5) Streamline read-only to read-write page mappings.
6) For pmap_copy_page, don't enable write mapping for source page.
7) Correct and clean-up pmap_incore.
8) Cluster initial kern_exec pagin.
9) Removal of some minor lint from kern_malloc.
10) Correct some ioopt code.
11) Remove some dead code from the MI swapout routine.
12) Correct vm_object_deallocate (to remove backing_object ref.)
13) Fix dead object handling, that had problems under heavy memory load.
14) Add minor vm_page_lookup improvements.
15) Some pages are not in objects, and make sure that the vm_page.c can
properly support such pages.
16) Add some more page deficit handling.
17) Some minor code readability improvements.
MUST be PG_BUSY. It is bogus to free a page that isn't busy,
because it is in a state of being "unavailable" when being
freed. The additional advantage is that the page_remove code
has a better cross-check that the page should be busy and
unavailable for other use. There were some minor problems
with the collapse code, and this plugs those subtile "holes."
Also, the vfs_bio code wasn't checking correctly for PG_BUSY
pages. I am going to develop a more consistant scheme for
grabbing pages, busy or otherwise. For now, we are stuck
with the current morass.
This introduce an xxxFS_BOOT for each of the rootable filesystems.
(Presently not required, but encouraged to allow a smooth move of option *FS
to opt_dontuse.h later.)
LFS is temporarily disabled, and will be re-enabled tomorrow.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
config option in pmap. Fix a problem with faulting in pages. Clean-up
some loose ends in swap pager memory management.
The system should be much more stable, but all subtile bugs aren't fixed yet.
Fix the UIO optimization code.
Fix an assumption in vm_map_insert regarding allocation of swap pagers.
Fix an spl problem in the collapse handling in vm_object_deallocate.
When pages are freed from vnode objects, and the criteria for putting
the associated vnode onto the free list is reached, either put the
vnode onto the list, or put it onto an interrupt safe version of the
list, for further transfer onto the actual free list.
Some minor syntax changes changing pre-decs, pre-incs to post versions.
Remove a bogus timeout (that I added for debugging) from vn_lock.
PHK will likely still have problems with the vnode list management, and
so do I, but it is better than it was.
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
of vnodes and objects. There are some metadata performance improvements
that come along with this. There are also a few prototypes added when
the need is noticed. Changes include:
1) Cleaning up vref, vget.
2) Removal of the object cache.
3) Nuke vnode_pager_uncache and friends, because they aren't needed anymore.
4) Correct some missing LK_RETRY's in vn_lock.
5) Correct the page range in the code for msync.
Be gentle, and please give me feedback asap.
noticed some major enhancements available for UP situations. The number
of UP TLB flushes is decreased much more than significantly with these
changes. Since a TLB flush appears to cost minimally approx 80 cycles,
this is a "nice" enhancement, equiv to eliminating between 40 and 160
instructions per TLB flush.
Changes include making sure that kernel threads all use the same PTD,
and eliminate unneeded PTD switches at context switch time.
is apparently useful for large shell systems, or systems with long running
idle processes. To enable the feature:
sysctl -w vm.swap_idle_enabled=1
Please note that some of the other vm sysctl variables have been renamed
to be more accurate.
Submitted by: Much of it from Matt Dillon <dillon@best.net>
the maximum pageout rate:
sysctl -w vm.vm_maxlaunder=n
1 < n < inf.
If paging heavily on large systems, it is likely that a performance
improvement can be achieved by increasing the parameter. On a large
system, the parm is 32, but numbers as large as 128 can make a big
difference. If paging is expensive, you might try decreasing the
number to 1-8.
sysctl -w vm.defer_swap_pageouts=1
Causes the system to resist the use of swap space. In low memory
conditions, performance will decrease.
sysctl -w vm.disable_swap_pageouts=1
Causes the system to mostly disable the use of swap space. In
low memory conditions, the system will likely start killing
processes.
it in struct proc instead.
This fixes a boatload of compiler warning, and removes a lot of cruft
from the sources.
I have not removed the /*ARGSUSED*/, they will require some looking at.
libkvm, ps and other userland struct proc frobbing programs will need
recompiled.
Distribute all but the most fundamental malloc types. This time I also
remembered the trick to making things static: Put "static" in front of
them.
A couple of finer points by: bde
plus the previous changes to use the zone allocator decrease the useage
of malloc by half. The Zone allocator will be upgradeable to be able
to use per CPU-pools, and has more intelligent usage of SPLs. Additionally,
it has reasonable stats gathering capabilities, while making most calls
inline.
entry when handling a fault. This is set by procfs whenever it wants
to write to a page, as a means of overriding `r-x COW' entries, but
causes failures in the `rwx' case.
Submitted by: bde
Introduce VFREE which indicates that vnode is on freelist.
Rename vholdrele() to vdrop().
Create vfree() and vbusy() to add/delete vnode from freelist.
Add vfree()/vbusy() to keep (v_holdcnt != 0 || v_usecount != 0)
vnodes off the freelist.
Generalize vhold()/v_holdcnt to mean "do not recycle".
Fix reassignbuf()s lack of use of vhold().
Use vhold() instead of checking v_cache_src list.
Remove vtouch(), the vnodes are always vget'ed soon enough
after for it to have any measuable effect.
Add sysctl debug.freevnodes to keep track of things.
Move cache_purge() up in getnewvnodes to avoid race.
Decrement v_usecount after VOP_INACTIVE(), put a vhold() on
it during VOP_INACTIVE()
Unmacroize vhold()/vdrop()
Print out VDOOMED and VFREE flags (XXX: should use %b)
Reviewed by: dyson
allowed to choose the address, or that the MAP_FIXED address has the same
remainder when modulo PAGE_SIZE as the file offset. Apparently this is
posix1003.1b specified behavior. SVR4 and the other *BSD's allow it too.
It costs us nothing to support and means we don't get EINVAL on some mmap
code that works perfectly elsewhere.
Obtained from: NetBSD
vm_inherit_t. These types are smaller than ints, so the prototypes
should have used the promoted type (int) to match the old-style function
definitions. They use just vm_prot_t and/or vm_inherit_t. This depends
on gcc features to work. I fixed the definitions since this is easiest.
The correct fix may be to change the small types to u_int, to optimize
for time instead of space.
free list problem. Also, the vnode age flag is no longer used by the
vnode pager. (It is actually incorrect to use then.) Constructive
feedback welcome -- just be kind.
VM systems usage of the kernel lock (lockmgr) code. This is a first
pass implementation, and is expected to evolve as needed. The API
for the lock manager code has not changed, but the underlying implementation
has changed significantly. This change should not materially affect
our current SMP or UP code without non-standard parameters being used.
the system is out of memory. The daemon does a minimal amount of work that
increases as the system becomes more likely to run out of memory and page in/out.
The default tuning is fairly low in background CPU usage, and sysctl variables
have been added to enable flexable operation. This is an experimental feature
that will likely be changed and improved over time.
of the kernel, and also most of the dynamic parts of the kernel. Additionally,
4MB pages will be allocated for display buffers as appropriate (only.)
The 4MB support for SMP isn't complete, but doesn't interfere with operation
either.
flag wasn't being respected during vref(), et. al. Note that this
isn't the eventual fix for the locking problem. Fine grained SMP
in the VM and VFS code will require (lots) more work.
and b_validend. The changes to vfs_bio.c are a bit ugly but hopefully
can be tidied up later by a slight redesign.
PR: kern/2573, kern/2754, kern/3046 (possibly)
Reviewed by: dyson
the pageout daemon wasn't always being waken up appropriately when the
(cache + free) queues were depleted.
Submitted by: David S. Miller <davem@jenolan.rutgers.edu>
There are various options documented in i386/conf/LINT, there is more to
come over the next few days.
The kernel should run pretty much "as before" without the options to
activate SMP mode.
There are a handful of known "loose ends" that need to be fixed, but
have been put off since the SMP kernel is in a moderately good condition
at the moment.
This commit is the result of the tinkering and testing over the last 14
months by many people. A special thanks to Steve Passe for implementing
the APIC code!
fork. (On my machine, fork is about 240usecs, vfork is 78usecs.)
Implement rfork(!RFPROC !RFMEM), which allows a thread to divorce its memory
from the other threads of a group.
Implement rfork(!RFPROC RFCFDG), which closes all file descriptors, eliminating
possible existing shares with other threads/processes.
Implement rfork(!RFPROC RFFDG), which divorces the file descriptors for a
thread from the rest of the group.
Fix the case where a thread does an exec. It is almost nonsense for a thread
to modify the other threads address space by an exec, so we
now automatically divorce the address space before modifying it.
space. (!)
Have each process use the kernel stack and pcb in the kvm space. Since
the stacks are at a different address, we cannot copy the stack at fork()
and allow the child to return up through the function call tree to return
to user mode - create a new execution context and have the new process
begin executing from cpu_switch() and go to user mode directly.
In theory this should speed up fork a bit.
Context switch the tss_esp0 pointer in the common tss. This is a lot
simpler since than swithching the gdt[GPROC0_SEL].sd.sd_base pointer
to each process's tss since the esp0 pointer is a 32 bit pointer, and the
sd_base setting is split into three different bit sections at non-aligned
boundaries and requires a lot of twiddling to reset.
The 8K of memory at the top of the process space is now empty, and unmapped
(and unmappable, it's higher than VM_MAXUSER_ADDRESS).
Simplity the pmap code to manage process contexts, we no longer have to
double map the UPAGES, this simplifies and should measuably speed up fork().
The following parts came from John Dyson:
Set PG_G on the UPAGES that are now in kernel context, and invalidate
them when swapping them out.
Move the upages object (upobj) from the vmspace to the proc structure.
Now that the UPAGES (pcb and kernel stack) are out of user space, make
rfork(..RFMEM..) do what was intended by sharing the vmspace
entirely via reference counting rather than simply inheriting the mappings.
The typo was detected once apon a time with the -Wunused compile option.
The result was that a block of code for implementing
madvise(.. MADV_SEQUENTIAL..) behavior was "dead" and unused, probably
negating the effect of activating the option.
Reviewed by: dyson
by Alan Cox <alc@cs.rice.edu>, and his description of the problem.
The bug was primarily in procfs_mem, but the mistake likely happened
due to the lack of vm system support for the operation. I added
better support for selective marking of page dirty flags so that
vm_map_pageable(wiring) will not cause this problem again.
The code in procfs_mem is now less bogus (but maybe still a little
so.)
centric rather than VM-centric to fix a problem with errors not being
detectable when the header is read.
Killed exech_map as a result of these changes.
There appears to be no performance difference with this change.
changes, so don't expect to be able to run the kernel as-is (very well)
without the appropriate Lite/2 userland changes.
The system boots and can mount UFS filesystems.
Untested: ext2fs, msdosfs, NFS
Known problems: Incorrect Berkeley ID strings in some files.
Mount_std mounts will not work until the getfsent
library routine is changed.
Reviewed by: various people
Submitted by: Jeffery Hsu <hsu@freebsd.org>
the page to be unbusy, and it caused some algorithmic problems
as a result. There were some other problems with it also, so
this is a general cleanup of the code.
Submitted by: Douglas Crosher <dtc@scrooge.ee.swin.oz.au> and myself.
when allocating memory for network buffers at interrupt time. This is due
to inadequate checking for the new mcl_map. Fixed by merging mb_map and
mcl_map into a single mb_map.
Reviewed by: wollman
This will make a number of things easier in the future, as well as (finally!)
avoiding the Id-smashing problem which has plagued developers for so long.
Boy, I'm glad we're not using sup anymore. This update would have been
insane otherwise.
anymore with the "full" collapse fix that we added about 1yr ago!!! The
code has been removed by optioning it out for now, so we can put it back
in ASAP if any problems are found.
and objects. Previously, "fancy" memory management techniques
such as that used by the M3 RTS would have the tendancy of chopping
up processes allocated memory into lots of little objects. Alan
has come up with some improvements to migtigate the sitution to
the point where even the M3 RTS only has one object for bss and
it's managed memory (when running CVSUP.) (There are still cases where the
situation isn't improved when the system pages -- but this is much much
better for the vast majority of cases.) The system will now be able
to much more effectively merge map entries.
Submitted by: Alan Cox <alc@cs.rice.edu>
also implies VM_PROT_EXEC. We support it that way for now,
since the break system call by default gives VM_PROT_ALL. Now
we have a better chance of coalesing map entries when mixing
mmap/break type operations. This was contributing to excessive
numbers of map entries on the modula-3 runtime system. The
problem is still not "solved", but the situation makes more
sense.
Eventually, when we work on architectures where VM_PROT_READ
is orthogonal to VM_PROT_EXEC, we will have to visit this
issue carefully (esp. regarding security issues.)
maps. Additionally, eliminate the map->hint distortion
associated with useracc. That may/may-not be the "right"
thing to do -- but time will tell.
Submitted by: Partially by Alan Cox <alc@cs.rice.edu>
vm_map_simplify and vm_map_simplify_entry. Make vm_map_simplify_entry
handle wired maps so that we can get rid of vm_map_simplify. Modify
the callers of vm_map_simplify to properly use vm_map_simplify_entry.
Submitted by: Alan Cox <alc@cs.rice.edu>
has the negative effect of disabling some map optimizations. This
patch defers the creation of the object until it needs to be at fault time.
Submitted by: Alan Cox <alc@cs.rice.edu>
that we do allow mlock to span unallocated regions (of course, not
mlocking them.) We also allow mlocking of RO regions (which the old
code couldn't.) The restriction there is that once a RO region is
wired (mlocked), it cannot be debugged (or EVER written to.)
Under normal usage, the new mlock code will be a significant improvement
over our old stuff.
that map entries are coalesced when appropriate. Also, conditionalize
some code that is currently not used in vm_map_insert. This mod
has been added to eliminate unnecessary map entries in buffer map.
Additionally, there were some cases where map coalescing could be done
when it shouldn't. That problem has been resolved.
scheme. Additionally, add the capability for checking for unexpected
kernel page faults. The maximum amount of kva space for buffers hasn't
been decreased from where it is, but it will now be possible to do so.
This scheme manages the kva space similar to the buffers themselves. If
there isn't enough kva space because of usage or fragementation, buffers
will be reclaimed until a buffer allocation is successful. This scheme
should be very resistant to fragmentation problems until/if the LFS code
is fixed and uses the bogus buffer locking scheme -- but a 'fixed' LFS
is not likely to use such a scheme.
Now there should be NO problem allocating buffers up to MAXPHYS.
problem of allocating contiguous buffer memory in general, but
make it much more likely to work at boot-up time. The best
chance for an LKM-type load of a sound driver is immediately
after the mount of the root filesystem.
This appears to work for a 64K allocation on an 8MB system.
it breaks in the DEVFS_ROOT case. replicate a bit too much of bdevvp()
in here to circumvent the problem. The real problem is the magic that
lives in bdevsw[1].
the one place that depended on it. wakeup() is now prototyped in
<sys/systm.h> so that it is normally visible.
Added nested include of <sys/queue.h> in <vm/vm_object.h>. The queue
macros are a more fundamental prerequisite for <vm/vm_object.h> than
the wakeup prototype and previously happened to be included by
namespace pollution from <sys/proc.h> or elsewhere.
64K. The change has essentially neutral effect on those machines with
little or no cache, and has a positive effect on "normal" machines
with 256K or more cache.
`show vmopag', `show page' and `show pageq'. Moved all vm ddb stuff
to the ends of the vm source files.
Changed printf() to db_printf(), `indent' to db_indent, and iprintf()
to db_iprintf() in ddb commands. Moved db_indent and db_iprintf()
from vm to ddb.
vm_page.c:
Don't use __pure. Staticized.
db_output.c:
Reduced page width from 80 to 79 to inhibit double spacing for long
lines (there are still some problems if words are printed across
column 79).
The default level works with minimal overhead, but one can also enable
full, efficient use of a 512K cache. (Parameters can be generated
to support arbitrary cache sizes also.)
The interface into the "VMIO" system has changed to be more consistant
and robust. Essentially, it is now no longer necessary to call vn_open
to get merged VM/Buffer cache operation, and exceptional conditions
such as merged operation of VBLK devices is simpler and more correct.
This code corrects a potentially large set of problems including the
problems with ktrace output and loaded systems, file create/deletes,
etc.
Most of the changes to NFS are cosmetic and name changes, eliminating
a layer of subroutine calls. The direct calls to vput/vrele have
been re-instituted for better cross platform compatibility.
Reviewed by: davidg
problem with the 'shell scripts' was found, but there was a 'strange'
problem found with a 486 laptop that we could not find. This commit
backs the code back to 25-jul, and will be re-entered after the snapshot
in smaller (more easily tested) chunks.
is little or no reason to create a swap pager for small mmap's. The
vm_map_insert code will automatically create a swap pager if the object
becomes too large. This fix, per a request from phk.
performance issues.
1) The pmap module has had too many inlines, and so the
object file is simply bigger than it needs to be.
Some common code is also merged into subroutines.
2) Removal of some *evil* PHYS_TO_VM_PAGE macro calls.
Unfortunately, a few have needed to be added also.
The removal caused the need for more vm_page_lookups.
I added lookup hints to minimize the need for the
page table lookup operations.
3) Removal of some bogus performance improvements, that
mostly made the code more complex (tracking individual
page table page updates unnecessarily). Those improvements
actually hurt 386 processors perf (not that people who
worry about perf use 386 processors anymore :-)).
4) Changed pv queue manipulations/structures to be TAILQ's.
5) The pv queue code has had some performance problems since
day one. Some significant scalability issues are resolved
by threading the pv entries from the pmap AND the physical
address instead of just the physical address. This makes
certain pmap operations run much faster. This does
not affect most micro-benchmarks, but should help loaded system
performance *significantly*. DG helped and came up with most
of the solution for this one.
6) Most if not all pmap bit operations follow the pattern:
pmap_test_bit();
pmap_clear_bit();
That made for twice the necessary pv list traversal. The
pmap interface now supports only pmap_tc_bit type operations:
pmap_[test/clear]_modified, pmap_[test/clear]_referenced.
Additionally, the modified routine now takes a vm_page_t arg
instead of a phys address. This eliminates a PHYS_TO_VM_PAGE
operation.
7) Several rewrites of routines that contain redundant code to
use common routines, so that there is a greater likelihood of
keeping the cache footprint smaller.
performance to LRU or worse when RSS limiting takes effect. Also,
make an end condition in the active queue scan more efficient in the
case where pages are removed from the active queue as a side effect
of a pmap operation.
sleeps before being swapped out. The time is increased from 4 secs to
10 secs. Originally I had decreased it from 20 to 4, but that is a bit
severe. 20 is too long though.
Re-enables the RSS limiting, and the routine is now tail-recursive,
making it much more safe (eliminates the possiblity of kernel stack
overflow.) Also, the RSS limiting is a little more intelligent about
finding the likely objects that are pushing the process over the limit.
Added some sysctls that help with VM system tuning.
New sysctl features:
1) Enable/disable lru pageout algorithm.
vm.pageout_algorithm = 0, default algorithm that works
well, especially using X windows and heavy
memory loading. Can have adverse effects,
sometimes slowing down program loading.
vm.pageout_algorithm = 1, close to true LRU. Works much
better than clock, etc. Does not work as well as
the default algorithm in general. Certain memory
"malloc" type benchmarks work a little better with
this setting.
Please give me feedback on the performance results
associated with these.
2) Enable/disable swapping.
vm.swapping_enabled = 1, default.
vm.swapping_enabled = 0, useful for cases where swapping
degrades performance.
The config option "NO_SWAPPING" is still operative, and
takes precedence over the sysctl. If "NO_SWAPPING" is
specified, the sysctl still exists, but "vm.swapping_enabled"
is hard-wired to "0".
Each of these can be changed "on the fly."
1) Make it much less likely to miss a wakeup in vm_page_free_wakeup
2) Create a new entry point into pmap: pmap_ts_referenced, eliminates
the need to scan the pv lists twice in many cases. Perhaps there
is alot more to do here to work on minimizing pv list manipulation
3) Minor improvements to vm_pageout including the use of pmap_ts_ref.
4) Major changes and code improvement to pmap. This code has had
several serious bugs in page table page manipulation. In order
to simplify the problem, and hopefully solve it for once and all,
page table pages are no longer "managed" with the pv list stuff.
Page table pages are only (mapped and held/wired) or
(free and unused) now. Page table pages are never inactive,
active or cached. These changes have probably fixed the
hold count problems, but if they haven't, then the code is
simpler anyway for future bugfixing.
5) The pmap code has been sorely in need of re-organization, and I
have taken a first (of probably many) steps. Please tell me
if you have any ideas.
1) Remove potential race conditions on waking up in vm_page_free_wakeup
by making sure that it is at splvm().
2) Fix another bug in vm_map_simplify_entry.
3) Be more complete about converting from default to swap pager
when an object grows to be large enough that there can be
a problem with data structure allocation under low memory
conditions.
4) Make some madvise code more efficient.
5) Added some comments.
code improvements include significant simplification of the reservation
of the swap pager control blocks for reads. Add a panic for an inconsistent
swap pager control block count.
reserving "cached" pages before waking up the pageout daemon. This will reserve
the faulted page, and keep the system from thrashing itself to death given
this condition.
queue in vm_fault.
Move the PG_BUSY in vm_fault to the correct place.
Remove redundant/unnecessary code in pmap.c.
Properly block on rundown of page table pages, if they are busy.
I think that the VM system is in pretty good shape now, and the following
individuals (among others, in no particular order) have helped with this
recent bunch of bugs, thanks! If I left anyone out, I apologize!
Stephen McKay, Stephen Hocking, Eric J. Chet, Dan O'Brien, James Raynard,
Marc Fournier.
some problems with the page-table page management code, since it can't
deal with the notion of page-table pages being paged out or in transit.
Also, clean up some stylistic issues per some suggestions from
Stephen McKay.
operations don't work with FICTITIOUS pages.) Also, close a window
between PG_MANAGED and pmap_enter that can mess up the accounting of
the managed flag. This problem could likely cause a hold_count error
for page table pages.
queue corruption problems, and to apply Gary Palmer's code cleanups.
David Greenman helped with these problems also. There is still
a hang problem using X in small memory machines.
case where blocking can occur, thereby giving other process's a chance
to modify the queue where a page resides. This could cause numerous
process and system failures.
that the datastructures needed to support the swap pager can take
enough space to fully deplete system memory, and cause a deadlock.
This change keeps large objects from being filled with dirty pages
without the appropriate swap pager datastructures. Right now,
default objects greater than 1/4 the size of available system memory
are converted to swap objects, thereby eliminating the risk of deadlock.
a condition when blocking can occur, and the daemon did not check properly
for a page remaining on the expected queue. Additionally, the inactive
target was being set much too large for small memory machines. It is now
being calculated based upon the amount of user memory available on every
pageout daemon run. Another problem was that if memory was very low, the
pageout daemon could fail repeatedly to traverse the inactive queue.
problem. BY MISTAKE, the vm_page_unqueue (or equiv) was removed from the
vm_fault code. Really bad things appear to happen if a page is on a queue
while it is being faulted.
contributions or ideas from Stephen McKay <syssgm@devetir.qld.gov.au>,
Alan Cox <alc@cs.rice.edu>, David Greenman <davidg@freebsd.org> and me:
More usage of the TAILQ macros. Additional minor fix to queue.h.
Performance enhancements to the pageout daemon.
Addition of a wait in the case that the pageout daemon
has to run immediately.
Slightly modify the pageout algorithm.
Significant revamp of the pmap/fork code:
1) PTE's and UPAGES's are NO LONGER in the process's map.
2) PTE's and UPAGES's reside in their own objects.
3) TOTAL elimination of recursive page table pagefaults.
4) The page directory now resides in the PTE object.
5) Implemented pmap_copy, thereby speeding up fork time.
6) Changed the pv entries so that the head is a pointer
and not an entire entry.
7) Significant cleanup of pmap_protect, and pmap_remove.
8) Removed significant amounts of machine dependent
fork code from vm_glue. Pushed much of that code into
the machine dependent pmap module.
9) Support more completely the reuse of already zeroed
pages (Page table pages and page directories) as being
already zeroed.
Performance and code cleanups in vm_map:
1) Improved and simplified allocation of map entries.
2) Improved vm_map_copy code.
3) Corrected some minor problems in the simplify code.
Implemented splvm (combo of splbio and splimp.) The VM code now
seldom uses splhigh.
Improved the speed of and simplified kmem_malloc.
Minor mod to vm_fault to avoid using pre-zeroed pages in the case
of objects with backing objects along with the already
existant condition of having a vnode. (If there is a backing
object, there will likely be a COW... With a COW, it isn't
necessary to start with a pre-zeroed page.)
Minor reorg of source to perhaps improve locality of ref.
is needed because of the vm_fault used to bring the page table page
for the kernel stack (UPAGES) back in. The consequence of the
previous incorrect change was a system hang.
in a suboptimal manner. I had also noticed some panics that appeared
to be at least superficially caused by this problem. Also, included
are some minor mods to support more general handling of page table page
faulting. More details in a future commit.
device have reference count problems. We mark the underlying object
ono-persistent, and account for the reference count that the VM system
maintainsfor the special device close. This should fix the removable
device problem.
regarding the "real" problem with maps that we have been having
over the last few weeks. He noted that the first_free pointer was
left dangling in certain circumstances -- and he was right!!! This
should fix the map problems that we were having, and also give us the
advantage of being able to simplify maps more aggressively.
The pmap_remove in vm_map_clean incorrectly unmapped the entire
map entry.
The new vm_map_simplify_entry code had an error (the offset
of the combined map entry was not set correctly.)
Submitted by: Alan Cox <alc@cs.rice.edu>
were paged in under low swap space conditions to both loose their
backing store and their dirty bits. This would cause pages to
be demand zeroed under certain conditions in low VM space conditions
and consequential sig-11's or sig-10's. This situation was made
worse lately when the level for swap space reclaim threshold was
increased.
on in the FreeBSD development, I had made a global lock around the
rlist code. This was bogus, and now the lock is maintained on a
per resource list basis. This now allows the rlist code to be used for
almost any non-interrupt level application.
that Bruce asked for.
These still are not quite perfect, and in particular, it can get
upset on extreme boundary cases (addr = 0xfff, len = 0xffffffff,
which would end up mapping a single page rather than failing), but
this is better code that I committed before.
(note, the VM system does not (apparently) support single mmap segment
sizes above 0x80000000 anyway)
kern_fork.c: add the tiny bit of code for rfork operation.
kern/sysv_*: shmfork() takes one less arg, it was never used.
sys/shm.h: drop "isvfork" arg from shmfork() prototype
sys/param.h: declare rfork args.. (this is where OpenBSD put it..)
sys/filedesc.h: protos for fdshare/fdcopy.
vm/vm_mmap.c: add minherit code, add rounding to mmap() type args where
it makes sense.
vm/*: drop unused isvfork arg.
Note: this rfork() implementation copies the address space mappings,
it does not connect the mappings together. ie: once the two processes
have split, the pages may be shared, but the address space is not. If one
does a mmap() etc, it does not appear in the other. This makes it not
useful for pthreads, but it is useful in it's own right for having
light-weight threads in a static shared address space.
Obtained from: Original by Ron Minnich, extended by OpenBSD
to help diagnose a problem on wcarchive (where the kernel stack was
sometimes not present), but is useful in its own right since swapping
actually reduces performance on some systems (such as wcarchive).
Note: swapping in this context means making the U pages pageable and has
nothing to do with generic VM paging, which is unaffected by this option.
Reviewed by: <dyson>
way to avoid crossing a 64K DMA boundary was to specify an alignment
greater than the size even when the alignment didn't matter, and for
sizes larger than a page, this reduced the chance of finding enough
contiguous pages. E.g., allocations of 8K not crossing a 64K boundary
previously had to be allocated on 8K boundaries; now they can be
allocated on any 4K boundary except (64 * n + 60)K.
Fixed bugs in vm_alloc_page_contig():
- the last page wasn't allocated for sizes smaller than a page.
- failures of kmem_alloc_pageable() weren't handled.
Mutated vm_page_alloc_contig() to create a more convenient interface
named contigmalloc(). This is the same as the one in 1.1.5 except
it has `low' and `high' args, and the `alignment' and `boundary'
args are multipliers instead of masks.
Speed up for vfs_bio -- addition of a routine bqrelse to greatly diminish
overhead for merged cache.
Efficiency improvement for vfs_cluster. It used to do alot of redundant
calls to cluster_rbuild.
Correct the ordering for vrele of .text and release of credentials.
Use the selective tlb update for 486/586/P6.
Numerous fixes to the size of objects allocated for files. Additionally,
fixes in the various pagers.
Fixes for proper positioning of vnode_pager_setsize in msdosfs and ext2fs.
Fixes in the swap pager for exhausted resources. The pageout code
will not as readily thrash.
Change the page queue flags (PG_ACTIVE, PG_INACTIVE, PG_FREE, PG_CACHE) into
page queue indices (PQ_ACTIVE, PQ_INACTIVE, PQ_FREE, PQ_CACHE),
thereby improving efficiency of several routines.
Eliminate even more unnecessary vm_page_protect operations.
Significantly speed up process forks.
Make vm_object_page_clean more efficient, thereby eliminating the pause
that happens every 30seconds.
Make sequential clustered writes B_ASYNC instead of B_DELWRI even in the
case of filesystems mounted async.
Fix a panic with busy pages when write clustering is done for non-VMIO
buffers.
redistribute a few last routines to beter places and shoot the file
I haven't act actually 'deleted' the file yet togive people time
to
have done a config.. I.e. they are likely to have done one in a week or so
so I'll remove it then..
it's now empty.
makes the question of a USL copyright rather moot.
seems to work hre just fine though I can't check every file
that changed due to limmited h/w, however I've checked enught to be petty
happy withe hte code..
WARNING... struct lkm[mumble] has changed
so it might be an idea to recompile any lkm related programs
up. The effect of this was that msync with a size would generally sync
1 page less than it should. This problem was brought to my attention
by Darrel Herbst <dherbst@gradin.cis.upenn.edu> and Ron Minnich
<rminnich@sarnoff.com>.
<sys/types.h> (if KERNEL is defined). This allows removing bogus
dependencies on vm stuff in several places (e.g., ddb) and stops
<vm_param.h> from depending on <vm_param.h>
Added declaration of boolean_t to <vm/vm.h> (if KERNEL is not
defined). It never belonged in <vm/vm_param.h>. Unfortunately,
it is required for some vm headers that are included by applications.
Deleted declarations of TRUE and FALSE from <vm/vm_param.h>. They
are defined in <sys/param.h> if KERNEL is defined and we'll soon
find out if any applications depend on them being defined in a vm
header.
most devsw referenced functions are now static, as they are
in the same file as their devsw structure. I've also added DEVFS
support for nearly every device in the system, however
many of the devices have 'incorrect' names under DEVFS
because I couldn't quickly work out the correct naming conventions.
(but devfs won't be coming on line for a month or so anyhow so that doesn't
matter)
If you "OWN" a device which would normally have an entry in /dev
then search for the devfs_add_devsw() entries and munge to make them right..
check out similar devices to see what I might have done in them in you
can't see what's going on..
for a laugh compare conf.c conf.h defore and after... :)
I have not doen DEVFS entries for any DISKSLICE devices yet as that will be
a much more complicated job.. (pass 5 :)
pass 4 will be to make the devsw tables of type (cdevsw * )
rather than (cdevsw)
seems to work here..
complaints to the usual places.. :)
Move a lot of variables home to their own code (In good time before xmas :-)
Introduce the string descrition of format.
Add a couple more functions to poke into these marvels, while I try to
decide what the correct interface should look like.
Next is adding vars on the fly, and sysctl looking at them too.
Removed a tine bit of defunct and #ifdefed notused code in swapgeneric.
Staticized some functions.
__purified some functions. Some functions were bogusly declared as
returning `const'. This hasn't done anything since gcc-2.5. For
later versions of gcc, the equivalent is __attribute__((const)) at
the end of function declarations.
That's EVERY SINGLE driver that has an entry in conf.c..
my next trick will be to define cdevsw[] and bdevsw[]
as empty arrays and remove all those DAMNED defines as well..
Each of these drivers has a SYSINIT linker set entry
that comes in very early.. and asks teh driver to add it's own
entry to the two devsw[] tables.
some slight reworking of the commits from yesterday (added the SYSINIT
stuff and some usually wrong but token DEVFS entries to all these
devices.
BTW does anyone know where the 'ata' entries in conf.c actually reside?
seems we don't actually have a 'ataopen() etc...
If you want to add a new device in conf.c
please make sure I know
so I can keep it up to date too..
as before, this is all dependent on #if defined(JREMOD)
(and #ifdef DEVFS in parts)
This closes a probably rare but nonetheless real window that would result
in a process hanging or the system panicing.
Reviewed by: dyson, davidg
Submitted by: kato@eclogite.eps.nagoya-u.ac.jp (KATO Takenori)
structs and prototypes for syscalls.
Ifdefed duplicated decentralized declarations of args structs. It's
convenient to have this visible but they are hard to maintain. Some
are already different from the central declarations. 4.4lite2 puts
them in comments in the function headers but I wanted to avoid the
large changes for that.
These functions went away:
enosys (hasn't been used for some time)
enxio
enodev
enoioctl (was used only once, actually for a vop)
if_tun.c:
Continued cleaning up...
conf.h:
Probably fixed the type of d_reset_t. It is hard to tell the correct
type because there are no non-dummy device reset functions.
Removed last vestige of ambiguous sleep message strings.
functions. The application header <sys/user.h> includes <vm/vm.h>
which includes <vm/lock.h>...
vm.h:
Don't include <machine/cpufunc.h>. It is already included by
<sys/systm.h> in the kernel and isn't designed to be included by
applications (the 2.1 version causes a syntax error in C++ and the
current version has initializers that are invalid in strict C++).
lock.h:
Only declare kernel functions if KERNEL is defined.
is swapped in. Also, remove unnecessary map locking/unlocking during
selection of processes to be swapped out.
This code might afford proper panics as opposed to spontaneous reboots
on certain systems. This should allow us to debug these problems better.
David Greenman, it has been determined that the more sophisticated code
only made a very minor difference in fault performance. Therefore, this
code eliminates some of the complication of the fault code, decreasing
the amount of CPU used to scan shadow chains.
UPAGES and associated page table page. Panic on error. This is less than
optimial and will be fixed in the future, but is better than the old
behavior of panicing with a "kernel page directory invalid" in pmap_enter.
Submitted by: terry (terry lambert)
This is a composite of 3 patch sets submitted by terry.
they are:
New low-level init code that supports loadbal modules better
some cleanups in the namei code to help terry in 16-bit character support
some changes to the mount-root code to make it a little more
modular..
NOTE: mounting root off cdrom or NFS MIGHT be broken as I haven't been able
to test those cases..
certainly mounting root of disk still works just fine..
mfs should work but is untested. (tomorrows task)
The low level init stuff includes a total rewrite of init_main.c
to make it possible for new modules to have an init phase by simply
adding an entry to a TEXT_SET (or is it DATA_SET) list. thus a new module can
be added to the kernel without editing any other files other than the
'files' file.
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
that call vnode_pager_alloc() so that a failure return can be dealt with.
This fixes a panic seen on NFS clients when a file being opened is deleted
on the server before the open completes.
2) Removed unnecessary vm_object_lookup()/pager_cache(object, TRUE) pairs
after vnode_pager_alloc() calls - the object is already guaranteed to be
persistent.
3) Removed some gratuitous casts.
1) Files weren't properly synced on filesystems other than UFS. In some
cases, this lead to lost data. Most likely would be noticed on NFS.
The fix is to make the VM page sync/object_clean general rather than
in each filesystem.
2) Mixing regular and mmaped file I/O on NFS was very broken. It caused
chunks of files to end up as zeroes rather than the intended contents.
The fix was to fix several race conditions and to kludge up the
"b_dirtyoff" and "b_dirtyend" that NFS relies upon - paying attention
to page modifications that occurred via the mmapping.
Reviewed by: David Greenman
Submitted by: John Dyson
These changes solve the problem in a general way by moving the
initialization out of the individual fs_mountroot's and into swaponvp().
Submitted by: Poul-Henning Kamp
changes. The check for nswap was bogus, but the code was so convoluted
that it was difficult to tell. It's better now. :-)
Reviewed by: David Greenman (extensively), and John Dyson
Submitted by: Poul-Henning Kamp, w/tweaks by me.
inconsistencies in the VM system that eventually lead to a panic. These
changes fix the behavior to conform to the behavior in SunOS, which is
to deny faults to pages beyond the EOF (returning SIGBUS). Internally,
this is implemented by requiring faults to be within the object size
boundaries. These changes exposed another bug, namely that passing in
an offset to mmap when trying to map an unnamed anonymous region also
results in internal inconsistencies. In this case, the offset is forced
to zero.
Reviewed by: John Dyson and others
require specific partitions be mentioned in the kernel config
file ("swap on foo" is now obsolete).
From Poul-Henning:
The visible effect is this:
As default, unless
options "NSWAPDEV=23"
is in your config, you will have four swap-devices.
You can swapon(2) any block device you feel like, it doesn't have
to be in the kernel config.
There is a performance/resource win available by getting the NSWAPDEV right
(but only if you have just one swap-device ??), but using that as default
would be too restrictive.
The invisible effect is that:
Swap-handling disappears from the $arch part of the kernel.
It gets a lot simpler (-145 lines) and cleaner.
Reviewed by: John Dyson, David Greenman
Submitted by: Poul-Henning Kamp, with minor changes by me.
with davidg about it, I hereby kill two undocumented misfeatures:
The code to skip a miniroot in the swapdev is not particular useful, and
if we need it we need it to be done properly, ie size the fs and skip all
of it not some hardcoded size, and subtract what we skip from the length
in the first place.
The SEQSWAP dies too. It's not the way to do it, it doesn't work, and
nobody have expressed any great desire for it to work. The way to
implement it correctly would be a second argument to swapon(2) to give
a priority/policy information. Low priority swapdevs can be made so
by adding them at a far offset (0x80000000 kind of thing), with almost no
modification to the strategy routine (in particular a offset per swapdev).
But until the need is obvious, it will not be done.
to emit spurious page outside of object type messages. It is not
a fatal condition anyway, so the message will be omitted for
release. Also, the code that "clips" the allocation size, associated
with the above problem, was fixed.
space for the hash list buckets and is a little faster. The features
of tailq aren't needed. Increased the size of the object hash table
to improve performance. In the future, this will be changed so that
the table is sized dynamically.
Headers should always use `__inline' for inline functions to avoid
syntax errors when modules that don't even use the offending functions
are compiled with `gcc -ansi'.
pages that are in FS buffers. This fixes the (believed to already have been
fixed) problem with msync() not doing it's job...in other words, the
stuff that Andrew has continuously been complaining about.
Submitted by: John Dyson, w/minor changes by me.
Fixed remaining known bugs in the buffer IO and VM system.
vfs_bio.c:
Fixed some race conditions and locking bugs. Improved performance
by removing some (now) unnecessary code and fixing some broken
logic.
Fixed process accounting of # of FS outputs.
Properly handle NFS interrupts (B_EINTR).
(various)
Replaced calls to clrbuf() with calls to an optimized routine
called vfs_bio_clrbuf().
(various FS sync)
Sync out modified vnode_pager backed pages.
ffs_vnops.c:
Do two passes: Sync out file data first, then indirect blocks.
vm_fault.c:
Fixed deadly embrace caused by acquiring locks in the wrong order.
vnode_pager.c:
Changed to use buffer I/O system for writing out modified pages. This
should fix the problem with the modification date previous not getting
updated. Also dramatically simplifies the code. Note that this is
going to change in the future and be implemented via VOP_PUTPAGES().
vm_object.c:
Fixed a pile of bugs related to cleaning (vnode) objects. The performance
of vm_object_page_clean() is terrible when dealing with huge objects,
but this will change when we implement a binary tree to keep the object
pages sorted.
vm_pageout.c:
Fixed broken clustering of pageouts. Fixed race conditions and other
lockup style bugs in the scanning of pages. Improved performance.
to accurately track this. It isn't an indicator of resource consumption
anyway.
Removed cnt.v_kernel_pages: We don't implement this and doing so accurately
would be very difficult (and ambiguous - since process pages are often
double mapped in the kernel and the process address spaces).
VTEXT not always getting cleared when it is supposed to. Added check to
make sure that vm_object_remove() isn't called with a NULL pager or for
a pager for an OBJ_INTERNAL object (neither of which will be on the hash
list). Clear OBJ_CANPERSIST if we decide to terminate it because of no
resident pages.
was the wrong size. This is the likely cause of panics reported by
Lars Fredriksen and Paul Richards related to a -1 blkno when paging
via the swap_pager.
Submitted by: John Dyson
routines (and merged local changes). The changed vnode_pager_uncache
gets rids of the bogosity that you can call the routine without
having the vnode locked. The changed vnode_pager_umount properly locks
the vnode before calling vnode_pager_uncache.
now returns NULL and sets a global 'mb_map_full' when the map is full.
m_clalloc() has further been taught to expect this and do the right thing.
This should fix the "mb_map full" panics that several people have reported.
object has fallen off the end of the cached list - this is likely the
last reference to the vnode and it should be reused before non file
vnodes that are already on the free list (VDIR mostly).
New functions create - vm_object_pip_wakeup and pagedaemon_wakeup that
are used to reduce the actual number of wakeups.
New function vm_page_protect which is used in conjuction with some new
page flags to reduce the number of calls to pmap_page_protect.
Minor changes to reduce unnecessary spl nesting.
Rewrote vm_page_alloc() to improve readability.
Various other mostly cosmetic changes.
that 9 bits aren't lost in the conversion. Changed all callers to expect
this. This allows paging on large (>2GB) filesystems.
Submitted by: John Dyson
Use request==VM_ALLOC_NORMAL rather than object!=kmem_object in deciding
if the caller is "important" in vm_page_alloc(). Also established a new
low threshold for non-interrupt allocations via cnt.v_interrupt_free_min.
vm_pageout.c:
Various algorithmic cleanup. Some calculations simplified. Initialize
cnt.v_interrupt_free_min to 2 pages.
Submitted by: John Dyson
pager(). Almost completely rewrote vm_mmap(); when John gets done with
the bottom half, it will be a complete rewrite. Deprecated most use of
vm_object_setpager(). Removed side effect of setting object persist
in vm_object_enter and moved this into the pager(s). A few other
cosmetic changes.
Slight change to reverse collapsing so that vm_object_deallocate doesn't
have to be called recursively.
Removed half of a previous fix - the renamed page during a collapse doesn't
need to be marked dirty because the pager backing store pointers are copied
- thus preserving the page's data. This assumes that pages without backing
store are always dirty (except perhaps for when they are first zeroed, but
this doesn't matter).
Switch order of two lines of code so that the correct pager is removed
from the hash list. The previous code bogusly passed a NULL pointer to
vm_object_remove(). The call to vm_object_remove() should be unnecessary
if named anonymous objects were being dealt with correctly. They are
currently marked as OBJ_INTERNAL, which really screws up things (such as
this).
2) bump reference counts by 2 instead of 1 so that an object deallocate
doesn't try to recursively collapse the object.
3) mark pages renamed during the collapse as dirty so that their contents
are preserved.
Submitted by: John and me.
Fixed long standing bug in freeing swap space during object collapses.
Fixed 'out of space' messages from printing out too often.
Modified to use new kmem_malloc() calling convention.
Implemented an additional stat in the swap pager struct to count the
amount of space allocated to that pager. This may be removed at some
point in the future.
Minimized unnecessary wakeups.
vm_fault.c:
Don't try to collect fault stats on 'swapped' processes - there aren't
any upages to store the stats in.
Changed read-ahead policy (again!).
vm_glue.c:
Be sure to gain a reference to the process's map before swapping.
Be sure to lose it when done.
kern_malloc.c:
Added the ability to specify if allocations are at interrupt time or
are 'safe'; this affects what types of pages can be allocated.
vm_map.c:
Fixed a variety of map lock problems; there's still a lurking bug that
will eventually bite.
vm_object.c:
Explicitly initialize the object fields rather than bzeroing the struct.
Eliminated the 'rcollapse' code and folded it's functionality into the
"real" collapse routine.
Moved an object_unlock() so that the backing_object is protected in
the qcollapse routine.
Make sure nobody fools with the backing_object when we're destroying it.
Added some diagnostic code which can be called from the debugger that
looks through all the internal objects and makes certain that they
all belong to someone.
vm_page.c:
Fixed a rather serious logic bug that would result in random system
crashes. Changed pagedaemon wakeup policy (again!).
vm_pageout.c:
Removed unnecessary page rotations on the inactive queue.
Changed the number of pages to explicitly free to just free_reserved
level.
Submitted by: John Dyson
There is similar bogusness in the pageout daemon that will be fixed soon.
This fixes a panic pointed out to me by Bruce Evans that occurs when
/dev/mem is used to map managed memory.
Added hook for pmap_prefault() and use symbolic constant for new third
argument to vm_page_alloc() (vm_fault.c, various)
Changed the way that upages and page tables are held. (vm_glue.c)
Fixed architectural flaw in allocating pages at interrupt time that was
introduced with the merged cache changes. (vm_page.c, various)
Adjusted some algorithms to acheive better paging performance and to
accomodate the fix for the architectural flaw mentioned above. (vm_pageout.c)
Fixed pbuf handling problem, changed policy on handling read-behind page.
(vnode_pager.c)
Submitted by: John Dyson
need to be moved into the qcollapse and rcollapse routines, but I don't
have time at the moment to make all the required changes...this will do
for now.
being cleared in some cases for vnode backed objects; we now do this in
vnode_pager_alloc proper to guarantee it. Also be more careful in the
rcollapse code about messing with busy/bmapped pages.
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1. The pageout daemon used to block under certain
circumstances, and we needed to add new functionality
that would cause the pageout daemon to block more often.
Now, the pageout daemon mostly just gets rid of pages
and kills processes when the system is out of swap.
The swapping, rss limiting and object cache trimming
have been folded into a new daemon called "vmdaemon".
This new daemon does things that need to be done for
the VM system, but can block. For example, if the
vmdaemon blocks for memory, the pageout daemon
can take care of it. If the pageout daemon had
blocked for memory, it was difficult to handle
the situation correctly (and in some cases, was
impossible).
2. The collapse problem has now been entirely fixed.
It now appears to be impossible to accumulate unnecessary
vm objects. The object collapsing now occurs when ref counts
drop to one (where it is more likely to be more simple anyway
because less pages would be out on disk.) The original
fixes were incomplete in that pathological circumstances
could still be contrived to cause uncontrolled growth
of swap. Also, the old code still, under steady state
conditions, used more swap space than necessary. When
using the new code, users will generally notice a
significant decrease in swap space usage, and theoretically,
the system should be leaving fewer unused pages around
competing for memory.
Submitted by: John Dyson
stages of debugging LFS:
* if we can't bmap, use old VOP code
*/
! if (/* (vp->v_mount && vp->v_mount->mnt_stat.f_type == MOUNT_LFS) || */
! VOP_BMAP(vp, foff, &dp, 0, 0)) {
for (i = 0; i < count; i++) {
if (i != reqpage) {
vnode_pager_freepage(m[i]);
--- 804,810 ----
/*
* if we can't bmap, use old VOP code
*/
! if (VOP_BMAP(vp, foff, &dp, 0, 0)) {
Reviewed by: gibbs
Submitted by: John Dyson
Disable the bogus declaration of pmap_bootstrap(). Since its arg list
is machine-dependent, it must be declared in a machine-dependent header.
vm_page.h:
Change `inline' to `__inline' and old-style function parameter lists for
inlined functions to new-style.
`inline' and old-style function parameter lists should never be used in
system headers, even in very machine-dependent ones, because they cause
warnings from gcc -Wreally-all.
fault was at offset 0 in the object. This resulted in more overhead but
was othewise benign. Added incore() check in vnode_pager_has_page()
to work around a problem with LFS...other than slightly higher overhead,
this change has no affect on UFS.
Enabled via REL2_1.
Added support for doing object collapses "on the fly". Enabled via REL2_1a.
Improved object collapses so that they can happen in more cases. Improved
sensing of modified pages to fix an apparant race condition and improve
clustered pageout opportunities. Fixed an "oops" with not restarting page
scan after a potential block in vm_pageout_clean() (not doing this can result
in strange behavior in some cases).
Submitted by: John Dyson & David Greenman
that this is intended for use only in floppy situations and is done at
the sacrifice of performance in that case (in ther words, this is not the
best solution, but works okay for this exceptional situation).
Submitted by: John Dyson
From now on, >all< swapdevices must be activated with "swapon".
If you havn't got it, add this line to /etc/fstab:
/dev/wd0b none swap sw 0 0
ne sec
Reason:
We want our GENERIC* kernels to have a large selection of swap-devices, but
on the other hand, we don't want to use a wd0b as swap when we boot of a
floppy. This way, we will never use a unexpected swapdevice. Nothing else
has changed.
scheme of things, so I've changed them to be more appropriate. page in/ous
are now associated with the pager that did them. Nuked v_fault as the
only fault of interest that wouldn't be already counted in v_trap is a VM
fault, and this is counted seperately.
2) Implemented most of the remaining counters and corrected the counting of
some that were done wrong. They are all almost correct now...just a few
minor ones left to fix.
