with GCC) align them with the GCC special alignment attribute.
Since there was then one available field, I split the protection
attributes and native host pointers into their own fields.
Before, with 3 dwords per TLB entry, some entries (about 3/8)
were spanning two processor cache lines (assuming a 32-byte
cache line). Now, they all fit within one cache line.
Knocked about 1.4% off Win95 boot time, probably more off normal
software runs.
All the EFLAGS bits used to be cached in separate fields. I left
a few of them in separate fields for now - might remove them
at some point also. When the arithmetic fields are known
(ie they're not in lazy mode), they are all cached in a
32-bit EFLAGS image, just like the x86 EFLAGS register expects.
All other eflags are store in the 32-bit register also, with
a few also mirrored in separate fields for now.
The reason I did this, was so that on x86 hosts, asm() statements
can be #ifdef'd in to do the calculation and get the native
eflags results very cheaply. Just to test that it works, I
coded ADD_EdId() and ADD_EwIw() with some conditionally compiled
asm()s for accelerated eflags processing and it works.
-Kevin
it can decide how to proceed. Some of those bits are necessary
to make TLB invalidation decisions. INVLPG doesn't cause
a whole TLB flush anymore, just one page. Some of the
current CPU behaviours model the P6, especially on CR0
reloads. Earlier processors kept some pre-change pre-fetched
instructions until a branch. We could probably model that
by setting a flag, and letting the revalidate_prefetch_q
function cause serialization.
The TLB flush code only invalidates entries which are not
already invalidated for the case where the TLB invalidation
ID trick is not in use.
Read-Modify-Write instructions. The first read phase stores
the host pointer in the "pages" field if a direct use pointer
is available. The Write phase first checks if a pointer was
issued and uses it for a direct write if available.
I chose the "pages" field since it needs to be checked by the
write_RMW_virtual variants anyways and thus needs to be
cached anyways.
Mostly the mods where to access.cc, but I did also macro-ize
the calls to write_RMW_virtual...() in files which use it
and cpu.h. Right now, the macro is just a straight pass-through.
I tried expanding it to a quick initial check for the pointer
availability to do the write in-place, with a function call
as a fall-back. That didn't seemed to matter at all.
Booting is not helped by this really. The upper bound of
the gain is 5 or 6%, and that's only if you have a loop that
looks like:
label:
add [eax], ebx ;; mega read-modify-write instruction
jmp label ;; intensive loop.
Kevin Lawton says he doesn't get a performance benefit.
I'm not sure if I do. Either way, the difference isn't
very large.
This code may get removed if it turns out to be useless.
- bx_gen_reg cannot be declared with BX_SMF or it can't read gen_reg
when static member functions are turned on.
- use "BX_CPU_C_PREFIX" instead of "BX_CPU_C::" for get_segment_base.
- the SMF (static member function) tricks are just plain wierd. The only way to
really be sure that you're not breaking something is to try compiling it with
SMF on and with SMF off. e.g. "configure && make" and
"configure --enable-processors=2 && make".
mode uses the notion of the guest-to-host TLB. This has the
benefit of allowing more uniform and streamlined acceleration
code in access.cc which does not have to check if CR0.PG
is set, eliminating a few instructions per guest access.
Shaved just a little off execution time, as expected.
Also, access_linear now breaks accesses which span two pages,
into two calls the the physical memory routines, when paging
is off, just like it always has for paging on. Besides
being more uniform, this allows the physical memory access
routines to known the complete data item is contained
within a single physical page, and stop reapplying the
A20ADDR() macro to pointers as it increments them.
Perhaps things can be optimized a little more now there too...
I renamed the routines to {read,write}PhysicalPage() as
a reminder that these routines now operate on data
solely within one page.
I also added a little code so that the paging module is
notified when the A20 line is tweaked, so it can dump
whatever mappings it wants to.
I have not tested these functions, but they model the format and
acceleration principals of the byte/word/dword functions. Give them
a try on both little/big endian machines.
so that a compare of the current access could be done more
efficiently against the cached values, both in the normal
paging routines, and in the accelerated code in access.cc.
This cut down the amount of code path needed to get to
direct use of a host address nicely, and speed definitely
got a boost as a result, especially if you use the
--enable-guest2host-tlb option.
The CR0.WP flag was a real pain, because it imparts
a complication on the way protections work. Fortunately
it's not a high-change flag, so I just base the new
cached info on the current CR0.WP value, and dump
the TLB cache when it changes.
access routines in access.cc, completing the upgrade of
those routines. You do need '--enable-guest2host-tlb', before
you get the speedups for now. The guest2host mods seem pretty
solid, though I do need to see what effects the A20 line has
on this cache and the paging TLB in general.
added --enable-repeat-speedups with default to disabled.
Reconfigure/recompile and the speedup code will be #ifdef'd
out for now. It manifested as junk written to the VGA screen
while booting/running Windows.
Also made some more mods to the main cpu loop. Moved the
handling of EXT/errorno outside the main loop, much like
the extra EIP/ESP commits were moved, for a little better
performance.
I changed the fetch_ptr/bytesleft method of fetching to
a slightly different model, which calculates a window
for which EIP will be valid (land on the current page),
and a bias which when applied to EIP will be from
0..upper_page_limit. Speed is about the same for either
method, but a pseudo-op/threaded-interpreter will plug
in better with this and be faster.
- Paging code rehash. You must now use --enable-4meg-pages to
use 4Meg pages, with the default of disabled, since we don't well
support 4Meg pages yet. Paging table walks model a real CPU
more closely now, and I fixed some bugs in the old logic.
- Segment check redundancy elimination. After a segment is loaded,
reads and writes are marked when a segment type check succeeds, and
they are skipped thereafter, when possible.
- Repeated IO and memory string copy acceleration. Only some variants
of instructions are available on all platforms, word and dword
variants only on x86 for the moment due to alignment and endian issues.
This is compiled in currently with no option - I should add a configure
option.
- Added a guest linear address to host TLB. Actually, I just stick
the host address (mem.vector[addr] address) in the upper 29 bits
of the field 'combined_access' since they are unused. Convenient
for now. I'm only storing page frame addresses. This was the
simplest for of such a TLB. We can likely enhance this. Also,
I only accelerated the normal read/write routines in access.cc.
Could also modify the read-modify-write versions too. You must
use --enable-guest2host-tlb, to try this out. Currently speeds
up Win95 boot time by about 3.5% for me. More ground to cover...
- Minor mods to CPUI/MOV_CdRd for CMOV.
- Integrated enhancements from Volker to getHostMemAddr() for PCI
being enabled.
Specific changes from the patch:
1.) renamed fdcache_eip to fdcache_ip, as it is using
the RIP instead of the EIP.
2.) added a Boolean array fdcache_is32 which uses is32
to determine icache hits. Otherwise we could run 32-bit
code as 16-bit or vice versa.
Modified Files:
config.h.in cpu/cpu.cc cpu/cpu.h memory/memory.cc
[ #433759 ] virtual address checks can overflow
> Bochs has been crashing in some cases when you try to access data which
> overlaps the segment limit, when the segment limit is near the 32-bit
> boundary. The example that came up a few times is reading/writing 4 bytes
> starting at 0xffffffff when the segment limit was 0xffffffff. The
> condition used to compare offset+length-1 with the limit, but
> offset+length-1 was overflowing so the comparison went wrong. This patch
> changes the condition so that it supports all segment limits except for
> sizes 0,1,2,3 bytes. Dave and I figured that these sizes would not be
> needed, while size 0xffffffff is used quite a lot.
in an output format similar to gdb (when you do info all-registers).
Also, if you do "info all" you get the CPU registers and the FPU
registers.
- added bx_cpu_c method called fpu_print_regs, which is implemented
in wmFPUemu_glue.cc
BX_SUPPORT_APIC were used. To follow the pattern used by other
names like this, I changed them all to BX_SUPPORT_APIC.
Thanks to Tom Lindström for chasing this down!
BX_CPU_C bx_cpu;
BX_MEM_C bx_mem;
and when more than one processor, use
BX_CPU_C *bx_cpu_array[BX_SMP_PROCESSORS];
BX_MEM_C *bx_mem_array[BX_ADDRESS_SPACES];
The changeover is controlled by BX_SMP_PROCESSORS, but there are only
a few code changes since nearly all code uses the BX_CPU(n) and BX_MEM(n)
macros.
- This turns out to make a 10% speed difference! With this revision,
the CVS version now gets 95% of the performance of the 3/25/2000
snapshot, which I've been using as my baseline.
tries to fix it. The shortcuts to register names such as AX and DL are
#defines in cpu/cpu.h, and they are defined in terms of BX_CPU_THIS_PTR.
When BX_USE_CPU_SMF=1, this works fine. (This is what bochs used for
a long time, and nobody used the SMF=0 mode at all.) To make SMP bochs
work, I had to get SMF=0 mode working for the CPU so that there could
be an array of cpus.
When SMF=0 for the CPU, BX_CPU_THIS_PTR is defined to be "this->" which
only works within methods of BX_CPU_C. Code outside of BX_CPU_C must
reference BX_CPU(num) instead.
- to try to enforce the correct use of AL/AX/DL/etc. shortcuts, they are
now only #defined when "NEED_CPU_REG_SHORTCUTS" is #defined. This is
only done in the cpu/*.cc code.
in BRANCH-smp-bochs revisions.
- The general task was to make multiple CPU's which communicate
through their APICs. So instead of BX_CPU and BX_MEM, we now have
BX_CPU(x) and BX_MEM(y). For an SMP simulation you have several
processors in a shared memory space, so there might be processors
BX_CPU(0..3) but only one memory space BX_MEM(0). For cosimulation,
you could have BX_CPU(0) with BX_MEM(0), then BX_CPU(1) with
BX_MEM(1). WARNING: Cosimulation is almost certainly broken by the
SMP changes.
- to simulate multiple CPUs, you have to give each CPU time to execute
in turn. This is currently implemented using debugger guards. The
cpu loop steps one CPU for a few instructions, then steps the
next CPU for a few instructions, etc.
- there is some limited support in the debugger for two CPUs, for
example printing information from each CPU when single stepping.
To see the commit logs for this use either cvsweb or
cvs update -r BRANCH-io-cleanup and then 'cvs log' the various files.
In general this provides a generic interface for logging.
logfunctions:: is a class that is inherited by some classes, and also
. allocated as a standalone global called 'genlog'. All logging uses
. one of the ::info(), ::error(), ::ldebug(), ::panic() methods of this
. class through 'BX_INFO(), BX_ERROR(), BX_DEBUG(), BX_PANIC()' macros
. respectively.
.
. An example usage:
. BX_INFO(("Hello, World!\n"));
iofunctions:: is a class that is allocated once by default, and assigned
as the iofunction of each logfunctions instance. It is this class that
maintains the file descriptor and other output related code, at this
point using vfprintf(). At some future point, someone may choose to
write a gui 'console' for bochs to which messages would be redirected
simply by assigning a different iofunction class to the various logfunctions
objects.
More cleanup is coming, but this works for now. If you want to see alot
of debugging output, in main.cc, change onoff[LOGLEV_DEBUG]=0 to =1.
Comments, bugs, flames, to me: todd@fries.net