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
You can now sling 14 devices off of a 274xT. In the process of adding
twin channel support, I removed all evident restrictions on supporting
Wide channeled devices, but I do not have a Wide controller to test them
on.
aic7770_seq.h, the pre-compiled header, is no longer needed since config
handles this dependancy.
timestamps for an atomic operation such as rename() on a local file
system to be identical.
Uniformize yet another idempotency ifdef. The comment nesting was
bogus.
make the sequencer code fully compatible with the aic7870 (ie 294x adaptors).
I've also added to my local mods putting the sequencer into "FASTMODE" clock.
This gives upwards of 2M/sec write preformance improvement in some scenarios.
There haven't been any reports of this causing problems, and I have been
reaping the benifits of it for more than a week now.
This also includes a new version of the pre-generated file <ugh>
Obtained from: John Aycock (aycock@cpsc.ucalgary.ca) and myself
a discussion going on about removing this code from the burden of the
GPL, but it won't happen before Beta, and this code should be tested
before release.
Supports 27/2842 class adaptec cards and is almost capable of supporting
aic7870 based adapters (294X series cards). It does not support Wide
controllers or the second channel on Twin boards although I have work in
progress on getting both channels and running.
I have also added a few performance improvements to this version that give
us approximately a 25% boost over the original driver. These patches have
been submitted to the author.
Obtained from: Linux aic7770 driver (John Aycock - aycock@cpsc.ucalgary.ca)
put_fs_word() to fuword() and suword(). Linux words are 16 bits
but BSD words are 32-bits, at least on vax's. Writing a too-large
word trashed the neighbouring short word.
Bruce
from David Greenman, Bruce Evans and Julian Elischer.
They are:
[vnode pager - David/Bruce]:
The following patch fixes a problem where some data could be lost in a
delayed-write buffer if the cached buffer was larger than a page. This fix was
provided by Bruce Evans and modified slightly by me.
[st.c - Julian]:
My fix for "bad request, must be between 0 and 0"
RTFS if you're interested).
[gnu/fpemul - David/Bruce]:
These changes fix single stepping of emulated FPU instructions.
Previously, the instruction after an emulated instruction was
executed without causing a SIGTRAP ...
The also fix the initial control word being different for the
GPL emulator (it is still wrong for the old emulator) and remove
an unnecessary panic when emulation is not configured (I hope at
least init, sh and reboot will run without floating point. I
remember only df and mkfs being broken by the lack of FP in 0.0).
[Various fixes described below - Bruce/David]:
sys/i386/boot/boot2.S:
Yet another attempt to propagate the correct fix for 16 vs
32-bit mode bugs. [verified]
sys/i386/i386/db_interface.c:
Protect against reentering Debugger().
sys/kern/kern_time.c:
Don't allow 'time.tv_usec == 0' except at clock interrupts.
sys/pcfs/pcfs_fat.c:
Make it compile without -O.
sys/scsi/sd.c:
Fix as posted to some freebsd mailing list.
(changes the order of the assignment of "sectors" because it earlier
value is needed first -DG)
sys/vm/vm_glue.c:
Fix stale comments and verbose code.
sys/vm/vm_mmap.c
Fix off by 1 errors and verbose code.
[From Nate - cosmetic but non-intrusive and useful enough to go in]
sys/i386/isa/isa.c:
Appended you'll find a patch to the NMI error log routine in isa/isa.c.
The below patch just adds some additional information when an NMI occurs
which can help debug the hardware problem.
