This is repository of my old source code which isn't updated any more. Go to git.rot13.org for current projects!
ViewVC logotype

Contents of /trunk/TODO

Parent Directory Parent Directory | Revision Log Revision Log

Revision 28 - (show annotations)
Mon Oct 8 16:20:26 2007 UTC (13 years, 3 months ago) by dpavlin
File size: 16846 byte(s)
++ trunk/HISTORY	(local)
$Id: HISTORY,v 1.1298 2006/07/22 11:27:46 debug Exp $
20060626	Continuing on SPARC emulation (beginning on the 'save'
		instruction, register windows, etc).
20060629	Planning statistics gathering (new -s command line option),
		and renaming speed_tricks to allow_instruction_combinations.
20060630	Some minor manual page updates.
		Various cleanups.
		Implementing the -s command line option.
20060701	FINALLY found the bug which prevented Linux and Ultrix from
		running without the ugly hack in the R2000/R3000 cache isol
		code; it was the phystranslation hint array which was buggy.
		Removing the phystranslation hint code completely, for now.
20060702	Minor dyntrans cleanups; invalidation of physpages now only
		invalidate those parts of a page that have actually been
		translated. (32 parts per page.)
		Some MIPS non-R3000 speed fixes.
		Experimenting with MIPS instruction combination for some
		addiu+bne+sw loops, and sw+sw+sw.
		Adding support (again) for larger-than-4KB pages in MIPS tlbw*.
		Continuing on SPARC emulation: adding load/store instructions.
20060704	Fixing a virtual vs physical page shift bug in the new tlbw*
		implementation. Problem noticed by Jakub Jermar. (Many thanks.)
		Moving rfe and eret to cpu_mips_instr.c, since that is the
		only place that uses them nowadays.
20060705	Removing the BSD license from the "testmachine" include files,
		placing them in the public domain instead; this enables the
		testmachine stuff to be used from projects which are
		incompatible with the BSD license for some reason.
20060707	Adding instruction combinations for the R2000/R3000 L1
		I-cache invalidation code used by NetBSD/pmax 3.0, lui+addiu,
		various branches followed by addiu or nop, and jr ra followed
		by addiu. The time it takes to perform a full NetBSD/pmax R3000
		install on the laptop has dropped from 573 seconds to 539. :-)
20060708	Adding a framebuffer controller device (dev_fbctrl), which so
		far can be used to change the fb resolution during runtime, but
		in the future will also be useful for accelerated block fill/
		copy, and possibly also simplified character output.
		Adding an instruction combination for NetBSD/pmax' strlen.
20060709	Minor fixes: reading raw files in src/file.c wasn't memblock
		aligned, removing buggy multi_sw MIPS instruction combination,
20060711	Adding a machine_qemu.c, which contains a "qemu_mips" machine.
		(It mimics QEMU's MIPS machine mode, so that a test kernel
		made for QEMU_MIPS also can run in GXemul... at least to some
		extent.)  Adding a short section about how to run this mode to
20060714	Misc. minor code cleanups.
20060715	Applying a patch which adds getchar() to promemul/yamon.c
		(from Oleksandr Tymoshenko).
		Adding yamon.h from NetBSD, and rewriting yamon.c to use it
		(instead of ugly hardcoded numbers) + some cleanup.
20060716	Found and fixed the bug which broke single-stepping of 64-bit
		programs between 0.4.0 and (caused by too quick
		refactoring and no testing). Hopefully this fix will not
		break too many other things.
20060718	Continuing on the 8253 PIT; it now works with Linux/QEMU_MIPS.
		Re-adding the sw+sw+sw instr comb (the problem was that I had
		ignored endian issues); however, it doesn't seem to give any
		big performance gain.
20060720	Adding a dummy Transputer mode (T414, T800 etc) skeleton (only
		the 'j' and 'ldc' instructions are implemented so far). :-}
20060721	Adding gtreg.h from NetBSD, updating dev_gt.c to use it, plus
		misc. other updates to get Linux 2.6 for evbmips/malta working
		(thanks to Alec Voropay for the details).
		FINALLY found and fixed the bug which made tlbw* for non-R3000
		buggy; it was a reference count problem in the dyntrans core.
20060722	Testing stuff; things seem stable enough for a new release.

==============  RELEASE 0.4.1  ==============

1 $Id: TODO,v 1.324 2006/07/22 10:23:39 debug Exp $
3 Hm. This file is in random order, and not all parts of it are up-to-date.
5 Code cleanup:
6 x) 64-bit ranges in src/cpus/memory_mips_v2p.c
7 x) Revert the dyntrans page template experiment? Hm.
8 x) Refactor the cpu type detection/initialization/listing.
10 Documentation:
11 x) Rewrite the section about experimental devices, after the
12 framebuffer acceleration has been implemented, and demos
13 written. (Symbolic names instead of numbers; example
14 use cases, etc. Mention demo files that use the various
15 features?)
16 x) "a very simple linear framebuffer device (for graphics output)"
17 under "which machines does gxemul emulate" ==> better
18 description?
19 x) Better description on how to set up a cross compiler?
20 Example for MIPS64.
22 Long-term implementation:
23 x) Testmachine includes:
24 + dev_fb block fill and copy
25 + dev_fb draw characters (from the built-in font)?
26 + dev_fb input device? mouse pointer coordinates and buttons
27 (allow changes in these to cause interrupts as well?)
28 + Redefine the halt() function so that it stops "sometimes
29 soon", i.e. usage in demo code should be:
30 for (;;) {
31 halt();
32 }
33 x) Rewrite the networking stack; make OpenBSD work better as a guest
34 OS, fix the performance problems, make Linux work with DHCP, etc.
35 x) Make the wdc controller work with modern versions of NetBSD!
36 x) Continue on SPARC emulation
37 + Enable it in the configure script as soon as it can
38 run all the demo programs.
39 x) Continue on Alpha emulation (virtual memory, etc). Cleanup.
40 x) Alignment exceptions (MIPS, PPC, ARM?, ...)
42 Long-term design:
43 x) Instruction combination collisions? How to avoid easily...
44 o) Actually use the settings object, better debugger stuff, etc!
45 o) Debugger command for enabling/disabling instruction statistics
46 during runtime. machine.statistics = on|off
48 x) Clock framework! Go through all clock devices, make sure they
49 return correct data, and run at correct speeds!
50 x) Dyntrans with valgrind-inspired memory checker. (In memory_rw,
51 it would be reasonably simple to add; in each individual fast
52 load/store routine = a lot more work, and it would become
53 kludgy very fast.)
54 x) Dyntrans with SMP... lots of work to be done here.
55 x) Dyntrans with cache emulation... lots of work here as well.
56 x) Reimplement the config file parser from scratch.
58 Test:
59 x) Test with more than one Sprite instance on an emulated network!
60 x) NetBSD 4.x, once it is out.
62 -------------------------------------------------------------------------------
64 Simple Valgrind-like checks?
65 o) Mark every address with bits which tell whether or not the address
66 has been written to.
67 o) What should happen when programs are loaded? Text/data, bss (zero
68 filled). But stack space and heap is uninitialized.
69 o) Uninitialized local variables:
70 A load from a place on the stack which has not previously
71 been stored to => warning. Increasing the stack pointer using
72 any available means should reset the memory to uninitialized.
73 o) If calls to malloc() and free() can be intercepted:
74 o) Access to a memory area after free() => warning.
75 o) Memory returned by malloc() is marked as not-initialized.
76 o) Non-passive, but good to have: Change the argument
77 given to malloc, to return a slightly larger memory
78 area, i.e. margin_before + size + margin_after,
79 and return the pointer + margin_before.
80 Any access to the margin_before or _after space results
81 in warnings. (free() must be modified to free the
82 actually allocated address.)
84 SMP:
85 o) dev_mp doesn't work well with dyntrans yet
86 o) In general, IPIs, CAS, LL/SC etc must be made to work with dyntrans
88 MIPS:
89 +) Some more work on opcodes.
90 x) MIPS64 revision 2.
91 o) Find out which actual CPUs implement the rev2 ISA!
92 x) _MAYBE_ TX79 and R5900 actually differ in their
93 opcodes? Check this carefully!
94 o) Dyntrans: Count register updates are probably not 100% correct yet.
95 o) Refactor code for performance and readability/maintainability.
96 o) Instruction combinations? Possible candidates (but profile first!):
97 o) R2000/R3000 cache cleaner!
98 o) strlen, memset loops etc
99 o) multiple loads/stores in a row, e.g. relative to
100 the stack pointer
101 o) lui + or, lui + add, and 64-bit variants
102 o) jr ra + addiu to the v0 register?
103 o) compare + branch
104 o) DROTR32 and similar MIPS64 rev 2 instructions, which have
105 a rotation bit which differs from previous ISAs.
106 o) EI and DI instructions for MIPS64/32 rev 2. NOTE: These are
107 _NOT_ the same as for R5900!
108 o) (Re)implement 128-bit loads/stores for R5900.
109 o) R4000 and others:
110 x) watchhi/watchlo exceptions, and other exception
111 handling details
112 o) R10000 and others: (R12000, R14000 ?)
113 x) memory space, exceptions, ...
114 x) use cop0 framemask for tlb lookups
115 (http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi/hdwr/bks/SGI_Developer/books/R10K_UM/sgi_html/t5.Ver.2.0.book_284.html)
117 Dyntrans:
118 x) Redesign/rethink the delay slot mechanism used for e.g. MIPS,
119 so that it caches a translation (that is, an instruction
120 word and the instr_call it was translated to the last
121 time), so that it doesn't need to do slow
122 to_be_translated for each end of page?
123 x) Program Counter statistics:
124 Per machine? What about SMP? All data to the same file?
125 A debugger command should be possible to use to enable/
126 disable statistics gathering.
127 Configuration file option!
128 x) Common fatal_abort() function, which drops into the debugger
129 without continuing.
130 x) INVALIDATION should cause translations in _all_ cpus to be
131 invalidated, e.g. on a write to a write-protected page
132 (containing code)
133 x) Call/return hints?
134 x) 16-bit encodings? (MIPS16, ARM Thumb, SH3, ...)
135 x) H8?
136 x) Lots of other stuff: see src/cpus/README_DYNTRANS
137 x) true recompilation backend? think carefully about this,
138 experiment in a separate project (not in GXemul)
139 o) First test would be to just implement a simple
140 instruction such as MIPS' addiu or lui, on AMD64
141 hosts...
142 x) Idle loop detection? (Depends on target.) Could be turned
143 into usleep(1) or similar on the host... except when doing
144 e.g. SMP emulation. Then it becomes trickier.
146 Alpha:
147 o) Virtual memory (tlbs etc)
148 o) Get {NetBSD,OpenBSD,Linux}/alpha booting. :)
150 SPARC:
151 o) Add all registers (floating point, control regs etc)
152 o) Save/restore register windows etc!
153 o) Load/stores!
154 o) Disassemly of some more instructions?
155 o) Are sll etc 32-bit sign-extending or zero-extending?
156 o) Finish the cmp (subcc) flag computation code.
157 o) Finish the GDB register stuff.
158 o) SPARC v8, v7 etc?
160 Debugger:
161 o) How does SMP debugging work? Does it simply use "threads"?
162 What if the guest OS (running on an emulated SMP machine)
163 has a usertask running, with userland threads?
164 o) Try to make the debugger more modular and, if possible, reentrant!
165 o) Remove the emul command? (But show network info if showing
166 machines?)
167 o) Generalize the expression evaluator. (debugger_expr.c?)
168 settable variables ("show nr of instructions on average")
169 emul[x] defaults to current emul
170 machine[x] defaults to current machine
171 cpu[x] defaults to currently focused cpu
172 registers cpu arch dependent (#-prefix)
173 symbols @-prefix
174 numeric constants decimal, hex, and octal ($-prefix)
175 boolean yes,no, true,false
176 operators (+ - * / % & | ^ !)
177 parentheses for grouping subexpressions
178 NOTE: the change from % to # for register prefix!
179 examples:
180 emul[0].machine[2].cpu[0].pc
181 machine[test2].cpu[1].ra = main
182 settings.show_trace_tree = yes
184 Settings:
185 o) Remove a setting.
186 o) Read/write a setting given a name. (Read as
187 string and/or int64_t simultaneously?)
188 o) Warnings when exiting the emulator, if the
189 settings have not been removed exactly in
190 the same way as they were added? This would
191 improve code cleanliness in the long term.
192 (I.e. require a corresponding _destroy()
193 function for all _new functions... machine_
194 cpu_ etc.)
196 Help command should have subsections! One for "expressions",
197 mirrored in the documentation, but the internal help should
198 be the one that should be considered correct.
199 o) see src/debugger.c for more
201 POWER/PowerPC:
202 x) PPC optimizations; instr combs
203 x) 64-bit stuff: either Linux on G5, or perhaps some hobbyist
204 version of AIX? (if there exists such a thing)
205 x) find and fix the bug which causes NetBSD/macppc to fail after
206 an install!
207 x) macppc: adb controller; keyboard (for framebuffer mode)
208 x) make OpenBSD/macppc work (PCI controller stuff)
210 Algor:
211 PCI and ISA and LOCAL interrupts! --> wdc could start working
212 Add interrupt controller in dev_algor.c.
214 ARM:
215 o) try to get netbsd/evbarm 3.x running (iq80321)
216 o) make the xscale counter registers (ccnt) work
217 o) make the ata controller usable for FreeBSD!
218 o) zaurus for openbsd...
219 o) debian/cats crashes because of unimplemented coproc stuff.
220 fix this?
222 Cache simulation:
223 o) Command line flags for:
224 o) CPU endianness?
225 o) Cache sizes? (multiple levels)
226 o) Separate from the CPU concept, so that multi-core CPUs sharing
227 e.g. a L2 cache can be simulated (?)
228 o) Instruction cache emulation is easiest (if separate from the
229 data cache); similar hack as the S;I; hack in cpu_dyntrans.c.
230 NOTE: if the architecture has a delay slot, then an instruction
231 slot can actually be executed as 2 instructions.
232 o) Data cache emulation = harder; each arch's load/store routines
233 must include support? running one instruction at a time and
234 having a cpu-dependant lookup function for each instruction
235 is another option (easier to implement, but very very slow).
237 Documentation:
238 o) machines, cpus, devices.
239 o) Automagic documentation generation:
240 x) REMEMBER that several machines/devices can be in
241 the same source file!
242 o) Try to rewrite the install instructions for those machines
243 that use 3MAX into using CATS? (To remove the need to a raw
244 ffs partition using up all of the disk image.)
246 More generic out_of_memory error reporting, and check everywhere!
247 Causes: OpenBSD has low default limits for normal users.
248 Host is 32-bit? (32-bit hosts are limited to 4 GB or less
249 of userspace memory.)
250 You are actually low on RAM. (As trivial as this might sound,
251 Unix systems usually allow processes to allocate virtual
252 memory beyond the amount of RAM in the machine.)
254 Breakpoints: 32-bit vs 64-bit sign extension for MIPS, warnings, etc.
255 Use the debugger's symbolic name stuff. (which will have to be
256 extended soon to support stuff like "2*x + symbol + y" etc. cool
257 stuff)
259 The Device subsystem:
260 x) allow devices to be moved and/or changed in size (down to a
261 minimum size, etc, or up to a max size)
262 x) keep track of interrupts and busses? actually, allowing any device
263 to be a bus might be a nice idea.
264 x) turn interrupt controllers into devices? :-)
265 x) refactor various clocks/nvram/cmos into one device?
267 Clocks:
268 x) General framework for automagic clock adjustment for _all_
269 kinds of clocks and timers. (Which should be possible to turn
270 off, of course, like the way DECstation emulation works now.)
272 PCI:
273 x) last write was ffffffff ==> fix this, it should be used
274 together with a mask to get the correct bits. also, not ALL
275 bits are size bits! (lowest 4 vs lowest 2?)
276 x) add support for address fixups
277 x) generalize the interrupt routing stuff (lines etc). this should
278 be per machine? or per bus, that's better
279 x) add a "pcn" NIC (AMD PCnet32 Lance 79c970 (PCI 1022:2000)),
280 could be useful for several machine modes (Malta, Algor, evbarm,
281 hp700?, macppc, etc.)
283 Network layer:
284 o) DHCP (for Debian and BSD installers :-)
285 o) increase performance
286 o) don't rely on NetBSD-ish usage
287 o) Multiple networks per emulation, and let different
288 NICs in machines connect to different networks.
289 o) many other issues: see src/net.c
291 Busses:
292 o) Redesign the entire "mainbus" concept!
293 o) Busses should be placed in a hierarchical tree!
294 o) Easily configurable interrupt routing in SMP systems.
295 o) Specific clock/bus speeds, cpu speeds etc.
296 o) Synchronization over network? or at least in dyntrans within
297 one emulated machine
298 o) dev->bus: TurboChannel, PCMCIA, ADB?
300 Config file parser:
301 o) Rewrite it from scratch!
302 o) Usage of any expression available through the debugger
303 o) Support for running debugger commands (like the -c
304 command line option)
306 Floating point layer:
307 o) make it common enough to be used by _all_ emulation modes
308 o) implement more stuff
309 o) non-IEEE modes (i.e. x86)?
311 Userland emulation:
312 x) Lots of stuff; freebsd and netbsd (and linux?) syscalls.
313 x) Dynamic linking? Hm.
315 Sound:
316 x) generic sound framework
317 x) add one or more sound cards as devices
319 ASC SCSI controller:
320 x) NetBSD/arc 2.0 uses the ASC controller in a way which GXemul
321 cannot yet handle. (NetBSD 1.6.2 works ok.) (Possibly a problem
322 in NetBSD itself, http://mail-index.netbsd.org/source-changes/
323 2005/11/06/0024.html suggests that.)
325 Caches / memory hierarchies: (this is mostly MIPS-specific)
326 o) src/memory*.c: Implement correct cache emulation for
327 all CPU types. (currently only R2000/R3000 is implemented)
328 (per CPU, multiple levels should be possible, associativity etc!)
329 o) R2000/R3000 isn't _100%_ correct, just almost correct :)
330 o) Move the -S (fill mem with random) functionality into the
331 memory.c subsystem, not machine.c or wherever it is now
332 o) ECC stuff, simulation of memory errors? (Machine dependent)
333 o) More than 4GB of emulated RAM, when run on a 32-bit host?
334 (using manual swap-out of blocks to disk, ugly)
335 o) A global command line option should be used to turn
336 cache emulation on or off. When off, caches should be
337 faked like they are right now. When on, caches and
338 memory latencies should be emulated as correctly as
339 possible.
341 File/disk/symbol handling:
342 o) Remove some of the complexity in file format guessing, for
343 Ultrix kernels that are actually disk images?
344 o) Better handling of tape files
345 o) Read function argument count and types from binaries? (ELF?)
346 o) Better demangling of C++ names. Note: GNU's C++ differs from e.g.
347 Microsoft's C++, so multiple schemes must be possible. See
348 URL at top of src/symbol_demangle.c for more info.
350 Userland ABI emulation:
351 o) see src/useremul.c
353 Terminal/console:
354 o) allow emulated serial ports to be connected to the outside
355 world in a more generic way, or even to other emulated
356 machines(?)
358 Save state of the whole emulated machine, to be able to load it back
359 in later? (Memory, all device's states, all registers and
360 so on. Like taking a snapshot. (SimOS seems to do this,
361 according to its website.))
363 Better framebuffer and X-windows functionality:
364 o) -Yx sometimes causes crashes.
365 o) Simple device access to framebuffer_blockcopyfill() etc,
366 and text output (using the built-in fonts), for dev_fb.
367 o) CLEAN UP the ugly event code
368 o) Mouse clicks can be "missed" in the current system; this is
369 not good. They should be put on a stack of some kind.
370 o) More 2D and 3D framebuffer acceleration.
371 o) Non-resizable windows? Or choose scaledown depending
372 on size (and center the image, with a black border).
373 o) Different scaledown on different windows?
374 o) Switch scaledown during runtime? (Ala CTRL-ALT-plus/minus)
375 o) Bug reported by Elijah Rutschman on MacOS with weird
376 keys (F5 = cursor down?).
377 o) Keyboard and mouse events:
378 x) Do this for more machines than just DECstation
379 x) more X11 cursor keycodes
380 x) Keys like CTRL, ALT, SHIFT do not get through
381 by themselves (these are necessary for example
382 to change the font of an xterm in X in the
383 emulator)
384 o) Generalize the framebuffer stuff by moving _ALL_ X11
385 specific code to src/x11.c!
387 Statistics: (this could be interesting)
388 o) Save to file and show graphics. It should be possible to
389 run gxemul after a simulation to just show the graphics,
390 or convert to a .ppm or .tga or similar.
391 o) memory accesses (to measure cache efficiency and
392 page coloring efficiency)
393 o) nr of simultaneous ASIDs in use in the TLB, for MIPS
394 o) percentage of time spent in different "states", such as
395 running userland code, kernel code, or idling (for CPUs
396 that have such an instruction, or whenever the PC is
397 inside a specific idle-function (address range)).
398 Possible additional state (for example on R3000): caches
399 disabled.
400 o) position of read/write on (SCSI) disks

  ViewVC Help
Powered by ViewVC 1.1.26