65c816: fix bank setting bug

With this fixed we return from syscalls on the right DP and lo and behold
init now runs as far as fork().

diff --git a/Kernel/lowlevel-65c816.s b/Kernel/lowlevel-65c816.s
index 1f0df83..1c13428 100644 (file)
--- a/Kernel/lowlevel-65c816.s
+++ b/Kernel/lowlevel-65c816.s
@@ -86,6 +86,24 @@ syscall      =       $fe
 ;      Helper - given the udata bank in A set the DP value correctly. May
 ;      be worth optimizing one day. Must not corrupt X or Y
 ;
+;      FIXME: there are various pieces of code that break if bank is > 0x80
+;      and thus overflows. It might be cleaner if we re-arranged page here
+;      and in the bank code as
+;                      code bank.8     }  For split I/D
+;                      data bank.8     }  0xFFFF = swapped?
+;                      dp.16           }  Our direct page (and swap)
+;                      (if swapped -> swap bank number)
+;      that way doexec would do al the work along with bank65c816 and we
+;      would be able to do things like sparse bank numbering in the C code
+;
+;      Even then we can't really make much use of 16MB of RAM because we
+;      are worst case 128 processes with split I/D and we just don't have
+;      enough bank 0 memory for that many DP and S pages!
+;
+;      (actually we could lazy swap DP and S up and down memory but really
+;      who can use that much space except as a ramdisk ?)
+;
+;
 setdp:
        php
 
@@ -94,11 +112,11 @@ setdp:
        .i16
 
        pha
-       asl     a                       ; twice page
+       and     #$00FF                  ; clear top bits
+       asl     a                       ; twice page (clears c)
        adc     #STACK_BANKOFF          ; we now point at the stack
        inc     a                       ; plus 0x100
        xba                             ; Swap to get xx00 format we need
-       and     #$FF00
        tcd
        pla