Max/SameBoy
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Added mechanism to handle MMIO read/write conflicts. Fixes #65

Browse Source
This commit is contained in:
Lior Halphon 2018-04-14 17:57:00 +03:00
parent 2c6f7906c5
commit be9df4d658
2 changed files with 71 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
Core/timing.c
View File

@ -156,13 +156,6 @@ static void GB_timers_run(GB_gameboy_t *gb, uint8_t cycles)
void GB_advance_cycles(GB_gameboy_t *gb, uint8_t cycles)
{
/* It appears that on the CGB, write timing is a bit different then on the DMG, effectively
making writes 1 T-cycle late when compared to the DMG. */
if (gb->is_cgb) {
cycles = (cycles + 1) & ~3;
if (cycles == 0) return;
}
// Affected by speed boost
gb->dma_cycles += cycles;

75
Core/z80_cpu.c
View File

@ -6,23 +6,42 @@
typedef void GB_opcode_t(GB_gameboy_t *gb, uint8_t opcode);
/*
About memory timings:
typedef enum {
/* Default behavior. If the CPU writes while another component reads, it reads the old value */
GB_CONFLICT_READ_OLD,
/* If the CPU writes while another component reads, it reads the new value */
GB_CONFLICT_READ_NEW,
/* If the CPU writes while another component reads, it reads a bitwise OR between the new and old values */
GB_CONFLICT_READ_OR,
/* If the CPU and another component write at the same time, the CPU's value "wins"*/
GB_CONFLICT_WRITE_CPU,
} GB_conflict_t;
Each M-cycle consists of 4 T-cycles. Every time the CPU accesses the memory it happens on the 1st T-cycle of an
M-cycle. During that cycle, other things may happen, such the PPU drawing to the screen. Since we can't really run
things in parallel, we run non-CPU "activities" serially using advnace_cycles(...). This is normally not a problem,
unless two entities (e.g. both the CPU and the PPU) read the same register at the same time (e.g. BGP). Since memory
accesses happen for an enitre T-cycle, if someone reads a value while someone else changes it during in the same
T-cycle, the read will return the new value. To correctly emulate this, a memory access T-cycle looks like this:
static const GB_conflict_t cgb_conflict_map[0x80] = {
[GB_IO_IF] = GB_CONFLICT_WRITE_CPU,
- Perform memory write (If needed)
- Run everything else
- Perform memory read (If needed)
/* Todo: most values not verified, and probably differ between revisions */
};
This is equivalent to running the memory write 1 T-cycle before the memory read.
static const GB_conflict_t dmg_conflict_map[0x80] = {
[GB_IO_IF] = GB_CONFLICT_WRITE_CPU,
*/
/* Todo: these are GB_CONFLICT_READ_NEW on MGB/SGB2 */
[GB_IO_BGP] = GB_CONFLICT_READ_OR,
[GB_IO_OBP0] = GB_CONFLICT_READ_OR,
[GB_IO_OBP1] = GB_CONFLICT_READ_OR,
/* Todo: These were verified on an SGB2 */
[GB_IO_LCDC] = GB_CONFLICT_READ_NEW,
[GB_IO_STAT] = GB_CONFLICT_READ_NEW,
[GB_IO_SCY] = GB_CONFLICT_READ_NEW,
[GB_IO_SCX] = GB_CONFLICT_READ_NEW,
/* Todo: these were not verified at all */
[GB_IO_WY] = GB_CONFLICT_READ_NEW,
[GB_IO_WX] = GB_CONFLICT_READ_NEW,
[GB_IO_LYC] = GB_CONFLICT_READ_NEW,
};
static uint8_t cycle_read(GB_gameboy_t *gb, uint16_t addr)
{
@ -48,10 +67,40 @@ static uint8_t cycle_read_inc_oam_bug(GB_gameboy_t *gb, uint16_t addr)
static void cycle_write(GB_gameboy_t *gb, uint16_t addr, uint8_t value)
{
assert(gb->pending_cycles);
GB_conflict_t conflict = GB_CONFLICT_READ_OLD;
if ((addr & 0xFF80) == 0xFF00) {
conflict = (gb->is_cgb? cgb_conflict_map : dmg_conflict_map)[addr & 0x7F];
}
switch (conflict) {
case GB_CONFLICT_READ_OLD:
GB_advance_cycles(gb, gb->pending_cycles);
GB_write_memory(gb, addr, value);
gb->pending_cycles = 4;
return;
case GB_CONFLICT_READ_NEW:
GB_advance_cycles(gb, gb->pending_cycles - 1);
GB_write_memory(gb, addr, value);
gb->pending_cycles = 5;
return;
case GB_CONFLICT_READ_OR: {
GB_advance_cycles(gb, gb->pending_cycles - 1);
uint8_t old_value = GB_read_memory(gb, addr);
GB_write_memory(gb, addr, value | old_value);
GB_advance_cycles(gb, 1);
GB_write_memory(gb, addr, value);
gb->pending_cycles = 4;
return;
case GB_CONFLICT_WRITE_CPU:
GB_advance_cycles(gb, gb->pending_cycles + 1);
GB_write_memory(gb, addr, value);
gb->pending_cycles = 3;
return;
}
}
}
static void cycle_no_access(GB_gameboy_t *gb)