GitLab

Removed dynamic dependencies from Makefile.

  ARM67Support      TRUE  (for 610,710,7500,7500FE)
  ARMSASupport      TRUE  (for StrongARM)
  ARMSASupport_RevS FALSE (for StrongARMs before rev S)
  IOMD1Support      TRUE  (for old machines)
  IOMD2Support      FALSE (They killed Phoebe!)
Version set to 4.00 (RISC OS 4)
This is the same as my last commit to the Ursula branch

Made the kernel control the layout of the free space map on the RAM disc, which avoids the rmreinit ramfs on every size change of the disc.

Also increase RAM disc max size to 128M bytes.

Change is enabled by NewRAMFS switch.

   If clamp was less than requested initial size, clamp was still being
   applied, which breaks API of OS_DynamicArea 0 by giving less than
   requested size without returning error.
   Fix is to allow requested initial size to defeat clamp, which is also
   much more useful (allows max size to float up to what client obviously
   needs).
 - Kernel now enumerates dynamic areas in alphabetical order - gives much
   neater switcher display with only small extra cost on area creation
   (getting switcher to do sort would be expensive/messy). Performance
   here still wipes the floor with old kernels.

- reestablish ownership of RAMFS dynamic area by kernel; this may be switched out again later, if new RAMFS takes over ownership

 - make ChangeDynamicArea allow re-entrancy slightly earlier on a shrink;
   now allows re-enter after page moves but before calling any post-shrink
   handler; this should fix problem with RAMFS and new FileCore (that now
   itself uses dynamic areas); needs testing once we pull a ROM together
 - Fix following bugs:
    - (new) kernel was keeping modules on active service chains during
      call to their finalise, now temporarily delinks them, as the API
      says it should; symptom eg. Zap 1.39 fails to quit (module Zap)
    - (longstanding) *help in a taskwindow was very dangerous (eg quit
      window, run other *help, start new window can all take out OS,
      by fundamentally cracking MessageTrans); fix: *help now does not
      keep MessageTrans descriptor on SVC stack (swapped out on pre-empt),
      now has re-entrancy guard, gives 'busy' error if reentered; minor
      infelicity is that busy message may be given on new *help after
      abandoning old one in taskwindow, but the next *help will work
    - (longstanding) *time could occasionally give a scrambled message on
      first call in a taskwindow (caused by multiple use of a general
      buffer)

added support for Sparse dynamic areas fixed performance disaster caused by naff API for Shrinkable areas implemented clamps for dynamic areas max size configured kernel to not own or create RAMFS area (needs new RAMFS)

AMBControl now uses system heap for space, not RMA
AMBControl enables Lazy task swapping if running on rev T or better SA
kernel now assumes there could be code above 64M
SWIS for limited 32 bit user code support implemented
Long command lines implemented (1k instead of 256)
Fast service call distribution implemented (uses Ursula module format)
*fx,*key etc now allow missing space before first parameter
*configure is reinstated (bug fix)

Memory map changes:
remove shadow ROM
move UNDEF stack, SoftCAM and MMU tables above 64M
expand RMA limit to 15M from 11M
expand SysHeap limit to 3M-32k from 2M-8k
expand SVC stack to 32k from 8k
partially protect kernel workspace from user access
protect SVC stack from user access

   - mapping of Cur/Sys/Sound area done more elegantly, and soft CAM info
     is now consistent with it
   - cached screen cleaning on VSync performed *after* VSync events
   - comments at top of ARM600 modernised
   - Pages_Unsafe/Safe code fixed to work properly on StrongARM with
     pages that are involved in interrupts (there is no fix for ARM8,
     since that is unlikely to be needed - an ASSERT checks use of ARM8
   - OS_DynamicArea code souped up, to be much more efficient for large
     numbers of dynamic areas (see comments near top of ChangeDyn)
   - cached screen is now suspended on h/w scroll (avoids possible cache
     incoherency)
2) API changes:
   - new OS_Memory reason code (10) allows Wimp to inform kernel of
     Wimp_ClaimFreeMemory, and can control VRAM rescue (see below)
   - new OS_ReadSysInfo reason code (6) allows reading of kernel values
     (reserved for Acorn use, eg. for SoftLoad, ROMPatch)
   - new OS_DynamicArea reason codes (6 and 7) allow for more efficient
     monitoring of dynamic areas by TaskManager (reserved for Acorn use)
3) Changes for Phoebe:
   - kernel runs a VRAM rescue process, which ensures that any VRAM not
     used for the screen is reclaimed if necessary and sinks to the bottom
     of the Free Pool. This is important for Phoebe, where VRAM is slower
     than SDRAM, but does no harm on other platforms.
   - logical copy of physical RAM is removed from memory map. This frees
     up 256M of address space that will later be used for PCI on Phoebe,
     but should do no harm on other platforms (this space is marked
     private in PRMs, so 3rd parties should not use it).

 2 Fix bug with Pages_Unsafe/Pages_Safe page moving for StrongARM
   (interrupt hole) - also better performance for StrongARM

 3 Improve perfromance of physical memory clear for StrongARM
   (make sure it uses burst write for STM)

 4 Suspend Chocolate task switching for StrongARM if SALDMIBbroken
   is TRUE

Module SWI chunks added to end of linked list on grounds that first loaded modules are probably more important, so should be checked first.

Some RISC OS 3.70 bits internationalised.
*ChangeDynamicArea moved into UtilityModule from TaskManager.

 - Chocolate screen mapping (section mapped and cached), StrongARM only
   Phoebe h/w (IOMD 2) will have register to assist this, but code currently
   relies on data abort mechanism to keep screen up to date wrt write-back
   data cache.

 - Chocolate AMBControl task switching (lazy page mapping), StrongARM only
   Improves task swapping speed. There appears to be a StrongAEM silicon
   bug rev 2 and 3) which means that LDMIB rn, {regs includind rn} cannot
   be reliably restarted after a data abort. This stuffs Chocolate AMBControl
   (awaiting response from Digital).

Both enhancements need more work to complete for Phoebe. Chocolate AMBControl
may well have to be made dormant because of silicon bug.

Note that this kernel *will* cause problems with task switching on StrongARM,
unless Chocolate task switching is disabled via !Flavour application.

 - Chocolate screen mapping (section mapped and cached), StrongARM only
   Phoebe h/w (IOMD 2) will have register to assist this, but code currently
   relies on data abort mechanism to keep screen up to date wrt write-back
   data cache.

 - Chocolate AMBControl task switching (lazy page mapping), StrongARM only
   Improves task swapping speed. There appears to be a StrongAEM silicon
   bug rev 2 and 3) which means that LDMIB rn, {regs includind rn} cannot
   be reliably restarted after a data abort. This stuffs Chocolate AMBControl
   (awaiting response from Digital).

Both enhancements need more work to complete for Phoebe. Chocolate AMBControl
may well have to be made dormant because of silicon bug.

Note that this kernel *will* cause problems with task switching on StrongARM,
unless Chocolate task switching is disabled via !Flavour application.

Sprout from Black 1996-11-05 09:41:48 UTC Neil Turton <nturton@gitlab.riscosopen.org> 'Import from cleaned 360 CD'
Cherrypick from master 1997-05-14 12:40:00 UTC Kevin Bracey <kbracey@gitlab.riscosopen.org> 'Taught AMBControl about shrinkable dynamic areas':
    BlackLog
    Docs/5thColumn/+Access,ffd
    Docs/5thColumn/+SrcIndex
    Docs/5thColumn/+SrcIndexO
    Docs/MemMaps/+Access,ffd
    Docs/MemMaps/+SrcIndex
    Docs/MemMaps/+SrcIndexO
    Docs/PrivDoc/+Access,ffd
    Docs/PrivDoc/+SrcIndex
    Docs/PrivDoc/+SrcIndexO
    Docs/PrivDoc/5thColumn/+Access,ffd
    Docs/PrivDoc/5thColumn/+SrcIndex
    Docs/PrivDoc/5thColumn/+SrcIndexO
    Makefile
    MkRom,fd7
    MkRomInst,fd7
    TestSrc/Begin
    TestSrc/ExtIO
    TestSrc/Ioc
    TestSrc/Mem1IOMD
    TestSrc/Mem2
    TestSrc/Mem3
    Version
    hdr/ExportVals/s/GetVals
    hdr/ExportVals/values
    hdr/KernelWS
    hdr/KeyWS
    hdr/PublicWS
    hdr/RISCOS
    hdr/VduExt
    s/AMBControl/growp
    s/ARM600
    s/Arthur2
    s/ArthurSWIs
    s/ChangeDyn
    s/Convrsions
    s/GetAll
    s/HeapMan
    s/KbdResPC
    s/Kernel
    s/MemInfo
    s/Middle
    s/ModHand
    s/NewIRQs
    s/NewReset
    s/PMF/i2cutils
    s/PMF/mouse
    s/PMF/osbyte
    s/PMF/osinit
    s/SWINaming
    s/SysComms
    s/vdu/vdudecl
    s/vdu/vdudriver
    s/vdu/vdugrafl
    s/vdu/vdumodes
    s/vdu/vdupal20
    s/vdu/vdupointer
    s/vdu/vduswis
Cherrypick from StrongARM 1997-05-07 06:30:31 UTC Kevin Bracey <kbracey@gitlab.riscosopen.org> 'Correct RISC OS 3.71 import':
    Docs/!ReadMe
    Docs/0197276.02
    Docs/5thColumn/Manual
    Docs/A540Extend
    Docs/AMBControl
    Docs/Kernel
    Docs/KernlSplit
    Docs/MMUControl
    Docs/MemMaps/130
    Docs/MemMaps/258
    Docs/Mode22
    Docs/Modes
    Docs/MonLead
    Docs/PaletteV
    Docs/PrivDoc/5thColumn/Concept
    Docs/PrivDoc/MMPM
    Docs/PrivDoc/ScreenMode
    Docs/ReadSysInf
    Docs/TVmodesMed,dde
    s/AMBControl/AMB
    s/AMBControl/Memory
    s/AMBControl/Options
    s/AMBControl/Workspace
    s/AMBControl/allocate
    s/AMBControl/deallocate
    s/AMBControl/growshrink
    s/AMBControl/main
    s/AMBControl/mapslot
    s/AMBControl/mapsome
    s/AMBControl/memmap
    s/AMBControl/readinfo
    s/AMBControl/service
    s/AMBControl/shrinkp
    s/Copro15ops
Cherrypick from Spin_xx 1996-11-21 12:14:21 UTC Neil Turton <nturton@gitlab.riscosopen.org> 'Import from SrcFiler':
    TestSrc/InitModule
    TestSrc/LEDDelay
    TestSrc/ROMCard
    TestSrc/ShowIOMDRs
    TestSrc/ToggleLED
Delete:
    Doc/!ReadMe
    Doc/0197276.02
    Doc/A540Extend
    Doc/Kernel
    Doc/KernlSplit
    Doc/MMUControl
    Doc/Mode22
    Doc/Modes
    Doc/MonLead
    Doc/PaletteV
    Doc/ReadSysInf
    Doc/TVmodesMed,dde

Clean reimport of hdr.RISCOS (real commit date 2008-03-28 by bavison), without any of the 3rd party allocations.