and uncontrolled growth.
The key fix is from Dan Carasone, who noticed that buf_canfree() was
counting in _bytes_ but freeing in _buffers_, which caused the instant
drop to lowater observed by some users.
We now control the rate of growth; the probability of getting a new
allocation is inversely proportional to the current size of the
cache. This idea is from a long-ago conversation with Kirk McKusick
and, if memory serves, was used for the file-system cache in some
other BSD variant at some point in history.
With growth and shrinkage more or less dealt with, we return the
default maximum cache size to 15%. The default _minimum_ cache size
is raised from 1/16 of the maximum cache size to 1/8, since 1/16 was
chosen when the maximum size was 30% of memory.
Finally, after observing the behaviour of the pagedaemon and the
buffer cache drainer under pathological workloads (e.g. a benchmark
that steps through 75% of available memory backwards) I have moved
the call to buf_drain() to the beginning of the pagedaemon from the
end; if the pagedaemon bogs down, it still won't get run as often
as it should, but at least this way it will see the state of the
free count and free target _before_ the scan step does its thing.
process context ('reaper').
From within the exiting process context:
* deactivate pmap and free vmspace while we can still block
* introduce MD cpu_lwp_free() - this cleans all MD-specific context (such
as FPU state), and is the last potentially blocking operation;
all of cpu_wait(), and most of cpu_exit(), is now folded into cpu_lwp_free()
* process is now immediatelly marked as zombie and made available for pickup
by parent; the remaining last lwp continues the exit as fully detached
* MI (rather than MD) code bumps uvmexp.swtch, cpu_exit() is now same
for both 'process' and 'lwp' exit
uvm_lwp_exit() is modified to never block; the u-area memory is now
always just linked to the list of available u-areas. Introduce (blocking)
uvm_uarea_drain(), which is called to release the excessive u-area memory;
this is called by parent within wait4(), or by pagedaemon on memory shortage.
uvm_uarea_free() is now private function within uvm_glue.c.
MD process/lwp exit code now always calls lwp_exit2() immediatelly after
switching away from the exiting lwp.
g/c now unneeded routines and variables, including the reaper kernel thread
virtual memory reservation and a private pool of memory pages -- by a scheme
based on memory pools.
This allows better utilization of memory because buffers can now be allocated
with a granularity finer than the system's native page size (useful for
filesystems with e.g. 1k or 2k fragment sizes). It also avoids fragmentation
of virtual to physical memory mappings (due to the former fixed virtual
address reservation) resulting in better utilization of MMU resources on some
platforms. Finally, the scheme is more flexible by allowing run-time decisions
on the amount of memory to be used for buffers.
On the other hand, the effectiveness of the LRU queue for buffer recycling
may be somewhat reduced compared to the traditional method since, due to the
nature of the pool based memory allocation, the actual least recently used
buffer may release its memory to a pool different from the one needed by a
newly allocated buffer. However, this effect will kick in only if the
system is under memory pressure.
the `# swap page in use' and `# swap page only' counters. However, at the
time of swap device removal we can no longer figure out how many of the
bad swap pages are actually also `swap only' pages.
So, on swap I/O errors arrange things to not include the bad swap pages in
the `swpgonly' counter as follows: uvm_swap_markbad() decrements `swpgonly'
by the number of bad pages, and the various VM object deallocation routines
do not decrement `swpgonly' for swap slots marked as SWSLOT_BAD.
first step towards per-device MAXPHYS, and has the beneficial side effect
of allowing clustering to MAXPHYS even on systems that need to run with
a reduced MAXBSIZE to get more metadata buffers.
uobject and uanon pointers rather than at the PQ_ANON flag to determine
which lock to hold, since PQ_ANON can be clear even when the anon's lock
is the one which we should hold (if the page was loaned from an object
and then freed by the object).
will be allocated for the respective usage types when there is contention
for memory.
replace "vnode" and "vtext" with "file" and "exec" in uvmexp field names
and sysctl names.
- fix the loaned case in uvm_pagefree().
- redo uvmexp.swpgonly accounting to work with page loaning.
add an assertion before each place we adjust uvmexp.swpgonly.
- fix uvm_km_pgremove() to always free any swap space associated with
the range being removed.
- get rid of UVM_LOAN_WIRED flag. instead, we just make sure that
pages loaned to the kernel are never on the page queues.
this allows us to assert that pages are not loaned and wired
at the same time.
- add yet more assertions.
- remove special treatment of pager_map mappings in pmaps. this is
required now, since I've removed the globals that expose the address range.
pager_map now uses pmap_kenter_pa() instead of pmap_enter(), so there's
no longer any need to special-case it.
- eliminate struct uvm_vnode by moving its fields into struct vnode.
- rewrite the pageout path. the pager is now responsible for handling the
high-level requests instead of only getting control after a bunch of work
has already been done on its behalf. this will allow us to UBCify LFS,
which needs tighter control over its pages than other filesystems do.
writing a page to disk no longer requires making it read-only, which
allows us to write wired pages without causing all kinds of havoc.
- use a new PG_PAGEOUT flag to indicate that a page should be freed
on behalf of the pagedaemon when it's unlocked. this flag is very similar
to PG_RELEASED, but unlike PG_RELEASED, PG_PAGEOUT can be cleared if the
pageout fails due to eg. an indirect-block buffer being locked.
this allows us to remove the "version" field from struct vm_page,
and together with shrinking "loan_count" from 32 bits to 16,
struct vm_page is now 4 bytes smaller.
- no longer use PG_RELEASED for swap-backed pages. if the page is busy
because it's being paged out, we can't release the swap slot to be
reallocated until that write is complete, but unlike with vnodes we
don't keep a count of in-progress writes so there's no good way to
know when the write is done. instead, when we need to free a busy
swap-backed page, just sleep until we can get it busy ourselves.
- implement a fast-path for extending writes which allows us to avoid
zeroing new pages. this substantially reduces cpu usage.
- encapsulate the data used by the genfs code in a struct genfs_node,
which must be the first element of the filesystem-specific vnode data
for filesystems which use genfs_{get,put}pages().
- eliminate many of the UVM pagerops, since they aren't needed anymore
now that the pager "put" operation is a higher-level operation.
- enhance the genfs code to allow NFS to use the genfs_{get,put}pages
instead of a modified copy.
- clean up struct vnode by removing all the fields that used to be used by
the vfs_cluster.c code (which we don't use anymore with UBC).
- remove kmem_object and mb_object since they were useless.
instead of allocating pages to these objects, we now just allocate
pages with no object. such pages are mapped in the kernel until they
are freed, so we can use the mapping to find the page to free it.
this allows us to remove splvm() protection in several places.
The sum of all these changes improves write throughput on my
decstation 5000/200 to within 1% of the rate of NetBSD 1.5
and reduces the elapsed time for "make release" of a NetBSD 1.5
source tree on my 128MB pc to 10% less than a 1.5 kernel took.
some time ago. The mistake was to check that the page was not
referenced since the last active scan before moving it to inactive.
Now we just clear reference and move it to inacive (which is where
the second clock hand sweep occurs).
the mapping is:
VM_PAGER_OK 0
VM_PAGER_BAD <unused>
VM_PAGER_FAIL <unused>
VM_PAGER_PEND 0 (see below)
VM_PAGER_ERROR EIO
VM_PAGER_AGAIN EAGAIN
VM_PAGER_UNLOCK EBUSY
VM_PAGER_REFAULT ERESTART
for async i/o requests, it used to be possible for the request to
be convert to sync, and the pager would return VM_PAGER_OK or VM_PAGER_PEND
to indicate whether the caller should perform post-i/o cleanup.
this is no longer allowed; pagers must now return 0 to indicate that
the async i/o was successfully started, and the caller never needs to
worry about doing the post-i/o cleanup.
each of the basic types (anonymous data, executable image, cached files)
and prevent the pagedaemon from reusing a given page if that would reduce
the count of that type of page below a sysctl-setable minimum threshold.
the thresholds are controlled via three new sysctl tunables:
vm.anonmin, vm.vnodemin, and vm.vtextmin. these tunables are the
percentages of pageable memory reserved for each usage, and we do not allow
the sum of the minimums to be more than 95% so that there's always some
memory that can be reused.
Mach VM's now. Specific changes:
- Pages now need not have all of their mappings removed before being
put on the inactive list. They only need to have the "referenced"
attribute cleared. This makes putting pages onto the inactive list
much more efficient. In order to eliminate redundant clearings of
"refrenced", callers of uvm_pagedeactivate() must now do this
themselves.
- When checking the "modified" attribute for a page (for clearing
PG_CLEAN), make sure to only do it if PG_CLEAN is currently set on
the page (saves a potentially expensive pmap operation).
- When scanning the inactive list, if a page is referenced, reactivate
it (this part was actually added in uvm_pdaemon.c,v 1.27). This
now works properly now that pages on the inactive list are allowed to
have mappings.
- When scanning the inactive list and considering a page for freeing,
remove all mappings, and then check the "modified" attribute if the
page is marked PG_CLEAN.
- When scanning the active list, if the page was referenced since its
last sweep by the scanner, don't deactivate it. (This part was
actually added in uvm_pdaemon.c,v 1.28.)
These changes greatly improve interactive performance during
moderate to high memory and I/O load.
When it wasn't (which could happen on a 4Mb machine with 32kb pages),
uvm_pagealloc_strat could refuse to allocate user memory, while the pagedaemon
didn't think it was worth freeing any more, resulting in the system seizing up.
<vm/pglist.h> -> <uvm/uvm_pglist.h>
<vm/vm_inherit.h> -> <uvm/uvm_inherit.h>
<vm/vm_kern.h> -> into <uvm/uvm_extern.h>
<vm/vm_object.h> -> nothing
<vm/vm_pager.h> -> into <uvm/uvm_pager.h>
also includes a bunch of <vm/vm_page.h> include removals (due to redudancy
with <vm/vm.h>), and a scattering of other similar headers.
level directly, instead of making the caller wrap the calls in
splimp()/splx().
- Add a comment documenting that interrupts that cause memory allocation
must be blocked while the free page queue is locked.
Since interrupts must be blocked while this lock is asserted, tying them
together like this helps to prevent mistakes.
numerous pagedaemon improvements were needed to make this useful:
- don't bother waking up procs waiting for memory if there's none to be had.
- start 4 times as many pageouts as we need free pages.
this should reduce latency in low-memory situations.
- in inactive scanning, if we find dirty swap-backed pages when swap space
is full of non-resident pages, reactivate some number of these to flush
less active pages to the inactive queue so we can consider paging them out.
this replaces the previous scheme of inactivating pages beyond the
inactive target when we failed to free anything during inactive scanning.
- during both active and inactive scanning, free any swap resources from
dirty swap-backed pages if swap space is full. this allows other pages
be paged out into that swap space.
chs