/[gxemul]/trunk/TODO

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

Diff of /trunk/TODO

Parent Directory | Revision Log | View Patch Patch

-revision 22 by dpavlin,
Mon Oct  8 16:19:37 2007 UTC
+revision 28 by dpavlin,
Mon Oct  8 16:20:26 2007 UTC
 Line 1
- $Id: TODO,v 1.209 2006/02/18 21:03:09 debug Exp $
+ $Id: TODO,v 1.324 2006/07/22 10:23:39 debug Exp $
  Hm. This file is in random order, and not all parts of it are up-to-date.
- Algor:  PCI and ISA and LOCAL interrupts! --> wdc could start working
+ Code cleanup:
-         Add interrupt controller in dev_algor.c.
+         x)  64-bit ranges in src/cpus/memory_mips_v2p.c
+         x)  Revert the dyntrans page template experiment? Hm.
+         x)  Refactor the cpu type detection/initialization/listing.
+ Documentation:
+         x)  Rewrite the section about experimental devices, after the
+             framebuffer acceleration has been implemented, and demos
+             written. (Symbolic names instead of numbers; example
+             use cases, etc. Mention demo files that use the various
+             features?)
+         x)  "a very simple linear framebuffer device (for graphics output)"
+             under "which machines does gxemul emulate" ==> better
+             description?
+         x)  Better description on how to set up a cross compiler?
+             Example for MIPS64.
+ Long-term implementation:
+         x)  Testmachine includes:
+                 + dev_fb block fill and copy
+                 + dev_fb draw characters (from the built-in font)?
+                 + dev_fb input device? mouse pointer coordinates and buttons
+                         (allow changes in these to cause interrupts as well?)
+                 + Redefine the halt() function so that it stops "sometimes
+                   soon", i.e. usage in demo code should be:
+                         for (;;) {
+                                 halt();
+                         }
+         x)  Rewrite the networking stack; make OpenBSD work better as a guest
+             OS, fix the performance problems, make Linux work with DHCP, etc.
+         x)  Make the wdc controller work with modern versions of NetBSD!
+         x)  Continue on SPARC emulation
+                 + Enable it in the configure script as soon as it can
+                   run all the demo programs.
+         x)  Continue on Alpha emulation  (virtual memory, etc). Cleanup.
+         x)  Alignment exceptions (MIPS, PPC, ARM?, ...)
+ Long-term design:
+         x)  Instruction combination collisions? How to avoid easily...
+         o)  Actually use the settings object, better debugger stuff, etc!
+         o)  Debugger command for enabling/disabling instruction statistics
+             during runtime.   machine.statistics = on|off
+         x)  MAINBUS REDESIGN!
+         x)  Clock framework! Go through all clock devices, make sure they
+             return correct data, and run at correct speeds!
+         x)  Dyntrans with valgrind-inspired memory checker. (In memory_rw,
+             it would be reasonably simple to add; in each individual fast
+             load/store routine = a lot more work, and it would become
+             kludgy very fast.)
+         x)  Dyntrans with SMP... lots of work to be done here.
+         x)  Dyntrans with cache emulation... lots of work here as well.
+         x)  Reimplement the config file parser from scratch.
+ Test:
+         x)  Test with more than one Sprite instance on an emulated network!
+         x)  NetBSD 4.x, once it is out.
+ -------------------------------------------------------------------------------
+ Simple Valgrind-like checks?
+         o)  Mark every address with bits which tell whether or not the address
+             has been written to.
+         o)  What should happen when programs are loaded?  Text/data, bss (zero
+             filled). But stack space and heap is uninitialized.
+         o)  Uninitialized local variables:
+                 A load from a place on the stack which has not previously
+                 been stored to => warning. Increasing the stack pointer using
+                 any available means should reset the memory to uninitialized.
+         o)  If calls to malloc() and free() can be intercepted:
+                 o)  Access to a memory area after free() => warning.
+                 o)  Memory returned by malloc() is marked as not-initialized.
+                 o)  Non-passive, but good to have: Change the argument
+                     given to malloc, to return a slightly larger memory
+                     area, i.e.  margin_before + size + margin_after,
+                     and return the pointer  + margin_before.
+                     Any access to the margin_before or _after space results
+                     in warnings. (free() must be modified to free the
+                     actually allocated address.)
+ SMP:
+         o)  dev_mp doesn't work well with dyntrans yet
+         o)  In general, IPIs, CAS, LL/SC etc must be made to work with dyntrans
+ MIPS:
+         +)  Some more work on opcodes.
+                 x) MIPS64 revision 2.
+                         o)  Find out which actual CPUs implement the rev2 ISA!
+                 x) _MAYBE_ TX79 and R5900 actually differ in their
+                    opcodes? Check this carefully!
+         o)  Dyntrans: Count register updates are probably not 100% correct yet.
+         o)  Refactor code for performance and readability/maintainability.
+         o)  Instruction combinations? Possible candidates (but profile first!):
+                 o)  R2000/R3000 cache cleaner!
+                 o)  strlen, memset loops etc
+                 o)  multiple loads/stores in a row, e.g. relative to
+                     the stack pointer
+                 o)  lui + or, lui + add, and 64-bit variants
+                 o)  jr ra + addiu to the v0 register?
+                 o)  compare + branch
+         o)  DROTR32 and similar MIPS64 rev 2 instructions, which have
+             a rotation bit which differs from previous ISAs.
+         o)  EI and DI instructions for MIPS64/32 rev 2. NOTE: These are
+             _NOT_ the same as for R5900!
+         o)  (Re)implement 128-bit loads/stores for R5900.
+         o)  R4000 and others:
+                 x)  watchhi/watchlo exceptions, and other exception
+                     handling details
+         o)  R10000 and others:  (R12000, R14000 ?)
+                 x)  memory space, exceptions, ...
+                 x)  use cop0 framemask for tlb lookups
+                     (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)
  Dyntrans:
-         TOP Priority:
+         x)  Redesign/rethink the delay slot mechanism used for e.g. MIPS,
-                 x)  64-bit stuff: good generic virtual memory translation, and
+                 so that it caches a translation (that is, an instruction
-                     other structures. not all 64 bits need to be used
+                 word and the instr_call it was translated to the last
-                     (e.g. Alpha)
+                 time), so that it doesn't need to do slow
-                 x)  Delay slots!
+                 to_be_translated for each end of page?
-                 x)  Old MIPS -> dyntrans!
+         x)  Program Counter statistics:
-         Other stuff:
+                 Per machine? What about SMP? All data to the same file?
+                 A debugger command should be possible to use to enable/
+                 disable statistics gathering.
+                 Configuration file option!
+         x)  Common fatal_abort() function, which drops into the debugger
+                 without continuing.
          x)  INVALIDATION should cause translations in _all_ cpus to be
              invalidated, e.g. on a write to a write-protected page
              (containing code)
-         x)  Think about sharing code between CPUs of the same
-             arch/bitlength/endianness (the translation caches).
-             (Eg. a little-endian R4000 and a little-endian R5000 could
-             share code, but not an R3000 and an R4000.)
-         x)  better (meaningful) instr call statistics
          x)  Call/return hints?
          x)  16-bit encodings? (MIPS16, ARM Thumb, SH3, ...)
-         x)  PPC optimizations; instr combs
+         x)  H8?
-         x)  Alpha
-         x)  SPARC
-         x)  PPC (64-bit stuff)
          x)  Lots of other stuff: see src/cpus/README_DYNTRANS
          x)  true recompilation backend? think carefully about this,
              experiment in a separate project (not in GXemul)
+                 o) First test would be to just implement a simple
+                    instruction such as MIPS' addiu or lui, on AMD64
+                    hosts...
+         x)  Idle loop detection? (Depends on target.) Could be turned
+             into usleep(1) or similar on the host... except when doing
+             e.g. SMP emulation. Then it becomes trickier.
+ Alpha:
+         o)  Virtual memory (tlbs etc)
+         o)  Get {NetBSD,OpenBSD,Linux}/alpha booting. :)
+ SPARC:
+         o)  Add all registers (floating point, control regs etc)
+         o)  Save/restore register windows etc!
+         o)  Load/stores!
+         o)  Disassemly of some more instructions?
+         o)  Are sll etc 32-bit sign-extending or zero-extending?
+         o)  Finish the cmp (subcc) flag computation code.
+         o)  Finish the GDB register stuff.
+         o)  SPARC v8, v7 etc?
+ Debugger:
+         o)  How does SMP debugging work? Does it simply use "threads"?
+                 What if the guest OS (running on an emulated SMP machine)
+                 has a usertask running, with userland threads?
+         o)  Try to make the debugger more modular and, if possible, reentrant!
+         o)  Remove the emul command? (But show network info if showing
+                 machines?)
+         o)  Generalize the expression evaluator. (debugger_expr.c?)
+                 settable variables      ("show nr of instructions on average")
+                 emul[x]                 defaults to current emul
+                 machine[x]              defaults to current machine
+                 cpu[x]                  defaults to currently focused cpu
+                 registers               cpu arch dependent (#-prefix)
+                 symbols                 @-prefix
+                 numeric constants       decimal, hex, and octal ($-prefix)
+                 boolean                 yes,no, true,false
+                 operators (+ - * / % & | ^ !)
+                 parentheses for grouping subexpressions
+                 NOTE: the change from % to # for register prefix!
+                 examples:
+                                 emul[0].machine[2].cpu[0].pc
+                                 machine[test2].cpu[1].ra = main
+                                 settings.show_trace_tree = yes
+                 Settings:
+                         o)  Remove a setting.
+                         o)  Read/write a setting given a name. (Read as
+                             string and/or int64_t simultaneously?)
+                         o)  Warnings when exiting the emulator, if the
+                             settings have not been removed exactly in
+                             the same way as they were added? This would
+                             improve code cleanliness in the long term.
+                             (I.e. require a corresponding _destroy()
+                             function for all _new functions... machine_
+                             cpu_ etc.)
+                 Help command should have subsections! One for "expressions",
+                 mirrored in the documentation, but the internal help should
+                 be the one that should be considered correct.
+         o)  see src/debugger.c for more
+ POWER/PowerPC:
+         x)  PPC optimizations; instr combs
+         x)  64-bit stuff: either Linux on G5, or perhaps some hobbyist
+                 version of AIX? (if there exists such a thing)
+         x)  find and fix the bug which causes NetBSD/macppc to fail after
+             an install!
+         x)  macppc: adb controller; keyboard (for framebuffer mode)
+         x)  make OpenBSD/macppc work (PCI controller stuff)
+ Algor:
+         PCI and ISA and LOCAL interrupts! --> wdc could start working
+         Add interrupt controller in dev_algor.c.
+ ARM:
+         o)  try to get netbsd/evbarm 3.x running (iq80321)
+         o)  make the xscale counter registers (ccnt) work
+         o)  make the ata controller usable for FreeBSD!
+         o)  zaurus for openbsd...
+         o)  debian/cats crashes because of unimplemented coproc stuff.
+             fix this?
+ Cache simulation:
+         o)  Command line flags for:
+                 o)  CPU endianness?
+                 o)  Cache sizes? (multiple levels)
+         o)  Separate from the CPU concept, so that multi-core CPUs sharing
+             e.g. a L2 cache can be simulated (?)
+         o)  Instruction cache emulation is easiest (if separate from the
+             data cache); similar hack as the S;I; hack in cpu_dyntrans.c.
+             NOTE: if the architecture has a delay slot, then an instruction
+             slot can actually be executed as 2 instructions.
+         o)  Data cache emulation = harder; each arch's load/store routines
+             must include support? running one instruction at a time and
+             having a cpu-dependant lookup function for each instruction
+             is another option (easier to implement, but very very slow).
+ Documentation:
+         o)  machines, cpus, devices.
+         o)  Automagic documentation generation:
+                 x)  REMEMBER that several machines/devices can be in
+                         the same source file!
+         o)  Try to rewrite the install instructions for those machines
+             that use 3MAX into using CATS? (To remove the need to a raw
+             ffs partition using up all of the disk image.)
  More generic out_of_memory error reporting, and check everywhere!
          Causes: OpenBSD has low default limits for normal users.
-Line 44 
 Breakpoints: 32-bit vs 64-bit sign exten
+Line 256 
 Breakpoints: 32-bit vs 64-bit sign exten
          extended soon to support stuff like  "2*x + symbol + y" etc. cool
          stuff)
- Sprite (guest OS for DECstation emulation)
-         x)  Timing problems during bootup?
  The Device subsystem:
          x)  allow devices to be moved and/or changed in size (down to a
              minimum size, etc, or up to a max size)
          x)  keep track of interrupts and busses? actually, allowing any device
-             to be a bus might be a nice idea
+             to be a bus might be a nice idea.
+         x)  turn interrupt controllers into devices? :-)
          x)  refactor various clocks/nvram/cmos into one device?
- ARM:
+ Clocks:
-         o)  add ID for "i80321 600MHz rev 2 (XScale core)"
+         x)  General framework for automagic clock adjustment for _all_
-         o)  make the ata controller usable for FreeBSD!
+             kinds of clocks and timers. (Which should be possible to turn
-         x)  zaurus for openbsd...
+             off, of course, like the way DECstation emulation works now.)
  PCI:
+         x)  last write was ffffffff ==> fix this, it should be used
+             together with a mask to get the correct bits. also, not ALL
+             bits are size bits! (lowest 4 vs lowest 2?)
          x)  add support for address fixups
          x)  generalize the interrupt routing stuff (lines etc). this should
              be per machine? or per bus, that's better
+         x)  add a "pcn" NIC (AMD PCnet32 Lance 79c970 (PCI 1022:2000)),
- MacPPC:
+             could be useful for several machine modes (Malta, Algor, evbarm,
-         x)  dev_zs / interrupts?
+             hp700?, macppc, etc.)
-         x)  adb controller; keyboard
-         x)  make OpenBSD/macppc work (PCI controller stuff)
  Network layer:
          o)  DHCP (for Debian and BSD installers :-)
-Line 87 
 Busses:
+Line 298 
 Busses:
          o)  dev->bus: TurboChannel, PCMCIA, ADB?
  Config file parser:
-         o)  Refresh/rewrite it :)
+         o)  Rewrite it from scratch!
          o)  Usage of any expression available through the debugger
-         o)  Expressions such as "add device" would be nice to be able
-             to do on the command line manually.
-         o)  Allow machine() entries even if there is no emul() entry (but
-             then don't allow any emul() entries at all).
          o)  Support for running debugger commands (like the -c
              command line option)
  Floating point layer:
          o)  make it common enough to be used by _all_ emulation modes
+         o)  implement more stuff
          o)  non-IEEE modes (i.e. x86)?
  Userland emulation:
-Line 108 
 Sound:
+Line 316 
 Sound:
          x)  generic sound framework
          x)  add one or more sound cards as devices
+ ASC SCSI controller:
+         x)  NetBSD/arc 2.0 uses the ASC controller in a way which GXemul
+             cannot yet handle. (NetBSD 1.6.2 works ok.) (Possibly a problem
+             in NetBSD itself, http://mail-index.netbsd.org/source-changes/
+/11/06/0024.html suggests that.)
  Caches / memory hierarchies: (this is mostly MIPS-specific)
-         o)  MIPS coproc.c: bits in config registers should reflect
-             correct cache sizes for _all_ CPU types. (currently only
-             implemented for R4000, R1x000, and a few others)
          o)  src/memory*.c: Implement correct cache emulation for
              all CPU types. (currently only R2000/R3000 is implemented)
-             (per CPU, multiple levels should be possible,
+             (per CPU, multiple levels should be possible, associativity etc!)
-             associativity etc!)
          o)  R2000/R3000 isn't _100%_ correct, just almost correct :)
          o)  Move the -S (fill mem with random) functionality into the
              memory.c subsystem, not machine.c or wherever it is now
-Line 128 
 Caches / memory hierarchies: (this is mo
+Line 338 
 Caches / memory hierarchies: (this is mo
              memory latencies should be emulated as correctly as
              possible.
- MIPS CPU emulation:  (note: this is for the OLD mips stuff)
-         o)  i386 bintrans backend: movn etc, slt[u] for
--bit mode, 64-bit shifts etc
-         o)  Instructions:
-                 o)  All ISAs:
-                         o)  Floating point exception handling, and
-                                 add more instructions.
-                         o)  Finish the MIPS16 translator, and test it!
-                         o)  MIPS ISA I, II, III, IV
-                         o)  MIPS V (SIMD vector stuff?)
-                         o)  MDMX  (MIPS Digital Media Extension)
-                         o)  MIPS 3D
-                         o)  MIPS MT (Multi-thread stuff) (What's this?)
-                 o)  Warn about mis-used bit fields (ie bits that
-                     should be all zeroes, warn about if they are not)!
-                     Both for coprocessor registers and for instruction
-                     opcodes.
-                 o)  the special2 stuff is a mess right now
-                 o)  warn and/or cause exceptions for unimplemented
-                     instructions (depending on CPU type)
-         o)  R2000/R3000:
-                 x)  R3000 "tri-byte stores". (What's this?)
-         o)  R4000 and others:
-                 x)  watchhi/watchlo exceptions, and other exception
-                     handling details
-         o)  R10000 and others:  (R12000, R14000 ?)
-                 x)  memory space, exceptions, ...
-                 x)  use cop0 framemask for tlb lookups
-                     (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)
-         o)  Implement load delays?  Warnings on interlocks.
-         o)  Implement all coprocessor 0 bits / functions.
-                 x)  coproc 0 selectors! (R4000 ?)
-         o)  R4300 (nintendo64, no mmu?), R5900 (playstation2, weird
-                 TLB/cache? 128-bit GPRs, new instructions),
-K (note: NOT R4000), 5K (note: NOT R5000),
-                 R6000 (ISA II), R8000
-         o)  Multi-cpu stuff:
-                 +)  Interrupt routing (ie devices vs mainbus, or
-                     connect each device to a fixed cpu)
-                 +)  SGI's NUMA architecture. Study
-                         x)  Linux sources
-                         x)  SGI's specs on NUMA address space
-                 +)  Ultrix?  NetBSD doesn't do SMP on MIPS yet :-(
-                 +)  Own experiments with ycx2.
  File/disk/symbol handling:
+         o)  Remove some of the complexity in file format guessing, for
+                 Ultrix kernels that are actually disk images?
          o)  Better handling of tape files
          o)  Read function argument count and types from binaries? (ELF?)
-         o)  Demangle C++ names.
+         o)  Better demangling of C++ names. Note: GNU's C++ differs from e.g.
+             Microsoft's C++, so multiple schemes must be possible. See
- Debugger:
+             URL at top of src/symbol_demangle.c for more info.
-         o)  see src/debugger.c for more
  Userland ABI emulation:
          o)  see src/useremul.c
-Line 206 
 Better framebuffer and X-windows functio
+Line 372 
 Better framebuffer and X-windows functio
                  on size (and center the image, with a black border).
          o)  Different scaledown on different windows?
          o)  Switch scaledown during runtime? (Ala CTRL-ALT-plus/minus)
+         o)  Bug reported by Elijah Rutschman on MacOS with weird
+             keys (F5 = cursor down?).
          o)  Keyboard and mouse events:
                  x)  Do this for more machines than just DECstation
                  x)  more X11 cursor keycodes

 Legend:



Removed from v.22
 


changed lines


 
Added in v.28
 Legend:



Removed from v.22
 


changed lines


 
Added in v.28
-Removed from v.22
+Added in v.28

	ViewVC Help
Powered by ViewVC 1.1.26