- Use ISR-based read method for SD Gecko.

cube/patches/Makefile

@@ -30,17 +30,18 @@ clean:
 build-sd:
 	@echo Building SD Patch ...
 	@$(CC) -Os $(OPTS) -c base/base.S
-	@$(CC) -Os $(OPTS) -c base/emulator.c -DDTK
+	@$(CC) -Os $(OPTS) -c base/emulator.c -DASYNC_READ -DDTK -DISR
 	@$(CC) -Os $(OPTS) -c base/fifo.c
 	@$(CC) -Os $(OPTS) -c base/igr.c
 	@$(CC) -Os $(OPTS) -c base/mix.c
-	@$(CC) -Os $(OPTS) -c base/blockdevice.c
-	@$(CC) -Os $(OPTS) -c sdgecko/sd.c -DSINGLE_SECTOR=2
+	@$(CC) -Os $(OPTS) -c base/blockdevice.c -DASYNC_READ
+	@$(CC) -Os $(OPTS) -c sdgecko/sd.c -DISR_READ=1 -DSINGLE_SECTOR=1
+	@$(CC) -Os $(OPTS) -c sdgecko/sd_isr.S
 	@$(CC) -Os $(OPTS) -c base/adp.c
 	@$(CC) -Os $(OPTS) -c base/frag.c
 	@$(CC) -Os $(OPTS) -c base/usbgecko.c
 	@$(CC) -Os $(OPTS) -c base/dolphin/os.c
-	@$(CC) -Os $(OPTS) -o sd.elf -T base/base.ld -T base/common.ld -T base/dolphin/os.ld base.o emulator.o fifo.o igr.o mix.o blockdevice.o sd.o adp.o frag.o usbgecko.o os.o
+	@$(CC) -Os $(OPTS) -o sd.elf -T base/base.ld -T base/common.ld -T base/dolphin/os.ld base.o emulator.o fifo.o igr.o mix.o blockdevice.o sd.o sd_isr.o adp.o frag.o usbgecko.o os.o
 	@rm -rf *.o
 	@$(OBJDUMP) -D sd.elf > $(DISASM)/sd_disasm.txt
 	@$(OBJCOPY) -O binary sd.elf sd.bin

cube/patches/base/emulator.c

@@ -132,7 +132,9 @@ bool dtk_fill_buffer(void)
 		dtk_fill_buffer();
 	}
 
+	#ifndef ISR
 	DCInvalidateRange(__builtin_assume_aligned(dtk.buffer, 32), sizeof(*dtk.buffer));
+	#endif
 	read_disc_frag(__builtin_assume_aligned(dtk.buffer, 32), sizeof(*dtk.buffer), dtk.current.position, read_callback);
 	#else
 	OSCancelAlarm(&read_alarm);
@@ -693,6 +695,14 @@ void exception_handler(OSException exception, OSContext *context, ...);
 extern void load_context(void) __attribute((noreturn));
 extern uint32_t load_context_end[];
 
+void external_interrupt_vector(void);
+
+static void write_branch(void *a, void *b)
+{
+	*(uint32_t *)a = (uint32_t)(b - (OS_BASE_CACHED - 0x48000002));
+	asm volatile("dcbst 0,%0; sync; icbi 0,%0" :: "r" (a));
+}
+
 void service_exception(OSException exception, OSContext *context, uint32_t dsisr, uint32_t dar)
 {
 	OSExceptionHandler handler;
@@ -707,10 +717,7 @@ void service_exception(OSException exception, OSContext *context, uint32_t dsisr
 				break;
 			}
 			if (handler) {
-				ptrdiff_t offset = (ptrdiff_t)handler - (ptrdiff_t)load_context_end;
-
-				*load_context_end = 0x48000000 | (offset & 0x3FFFFFC);
-				asm volatile("dcbst 0,%0; sync; icbi 0,%0" :: "r" (load_context_end));
+				write_branch(load_context_end, handler);
 				load_context();
 				return;
 			}
@@ -773,6 +780,9 @@ void init(void)
 	OSCreateAlarm(&command_alarm);
 
 	OSSetExceptionHandler(OS_EXCEPTION_DSI, exception_handler);
+	#ifdef ISR
+	write_branch((void *)0x80000500, external_interrupt_vector);
+	#endif
 }
 
 bool exi_probe(int32_t chan)

cube/patches/base/frag.c

@@ -129,5 +129,4 @@ void device_frag_read(void *dst, u32 len, u32 offset)
 		dst+=amountRead;
 		offset+=amountRead;
 	}
-	end_read();
 }

cube/patches/base/igr.c

@@ -51,6 +51,7 @@ static void load_dol(uint32_t offset, uint32_t size)
 		dcache_flush_icache_inv(image.data[i], image.dataLen[i]);
 	}
 
+	end_read();
 	image.entry();
 }
 

cube/patches/sdgecko/sd.c

@@ -4,9 +4,11 @@
 #**************************************************************************/
 
 #include <stdint.h>
+#include <stdbool.h>
 #include <string.h>
 #include "../base/common.h"
 #include "../base/dolphin/exi.h"
+#include "../base/dolphin/os.h"
 
 //CMD12 - Stop multiple block read command
 #define CMD12				0x4C
@@ -25,11 +27,28 @@
 #define exi_channel			(*(u8*)VAR_EXI_SLOT)
 #define exi_regs			(*(vu32**)VAR_EXI_REGS)
 
+static OSInterruptHandler TCIntrruptHandler = NULL;
+
 static struct {
 	uint32_t next_sector;
 	uint32_t last_sector;
+	#if ISR_READ
+	int items;
+	struct {
+		void *address;
+		uint32_t length;
+		uint32_t offset;
+		uint32_t sector;
+		read_frag_cb callback;
+	} queue[2];
+	#endif
 } mmc = {0};
 
+extern intptr_t isr_registers;
+extern int isr_transferred;
+
+void tc_interrupt_handler(OSInterrupt interrupt, OSContext *context);
+
 // EXI Functions
 static void exi_select()
 {
@@ -94,22 +113,34 @@ static void exi_read_to_buffer(void *dest, u32 len) {
 	}
 }
 
-static void rcvr_datablock(void *dest, u32 start_byte, u32 bytes_to_read) {
+static void rcvr_datablock(void *dest, u32 start_byte, u32 bytes_to_read, int sync) {
 	exi_select();
 
 	while(rcvr_spi() != 0xFE);
 
-	// Skip the start if it's a misaligned read
-	exi_read_to_buffer(0, start_byte);
-
-	// Read however much we need to in this block
-	exi_read_to_buffer(dest, bytes_to_read);
+	if(sync) {
+		// Skip the start if it's a misaligned read
+		exi_read_to_buffer(0, start_byte);
 
-	// Read out the rest from the SD as we've requested it anyway and can't have it hanging off the bus (2 for CRC discard)
-	u32 remainder = 2 + (SECTOR_SIZE - (start_byte+bytes_to_read));
-	exi_read_to_buffer(0, remainder);
+		// Read however much we need to in this block
+		exi_read_to_buffer(dest, bytes_to_read);
 
-	exi_deselect();
+		// Read out the rest from the SD as we've requested it anyway and can't have it hanging off the bus (2 for CRC discard)
+		u32 remainder = 2 + (SECTOR_SIZE - (start_byte+bytes_to_read));
+		exi_read_to_buffer(0, remainder);
+
+		exi_deselect();
+	}
+	else {
+		#if ISR_READ
+		isr_registers = OSUncachedToPhysical(exi_regs);
+		isr_transferred = -4;
+
+		OSInterrupt interrupt = OS_INTERRUPT_EXI_0_TC + (3 * exi_channel);
+		TCIntrruptHandler = OSSetInterruptHandler(interrupt, tc_interrupt_handler);
+		OSUnmaskInterrupts(OS_INTERRUPTMASK(interrupt));
+		#endif
+	}
 }
 
 #ifndef SINGLE_SECTOR
@@ -128,7 +159,7 @@ u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 	if(startByte) {
 		// amount to read in first block may vary if our read is small enough to fit in it
 		u32 amountInBlock = (len + startByte > SECTOR_SIZE) ? SECTOR_SIZE-startByte : len;
-		rcvr_datablock(dst,startByte, amountInBlock);
+		rcvr_datablock(dst,startByte, amountInBlock, 1);
 		numBytes-=amountInBlock;
 		dst+=amountInBlock;
 	}
@@ -137,21 +168,93 @@ u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 	u32 numFullBlocks = numBytes>>9; 
 	numBytes -= numFullBlocks << 9;
 	while(numFullBlocks) {
-		rcvr_datablock(dst, 0,SECTOR_SIZE);
+		rcvr_datablock(dst, 0,SECTOR_SIZE, 1);
 		dst+=SECTOR_SIZE; 
 		numFullBlocks--;
 	}
 
 	// Read any trailing half block
 	if(numBytes) {
-		rcvr_datablock(dst,0, numBytes);
+		rcvr_datablock(dst,0, numBytes, 1);
 		dst+=numBytes;
 	}
 	// End the read by sending CMD12
 	send_cmd(CMD12, 0);
 	return len;
 }
 #else
+#if ISR_READ
+static void mmc_read_queued(void)
+{
+	if (!EXILock(exi_channel, EXI_DEVICE_0, (EXICallback)mmc_read_queued))
+		return;
+
+	void *address = mmc.queue[0].address;
+	uint32_t length = mmc.queue[0].length;
+	uint32_t offset = mmc.queue[0].offset;
+	uint32_t sector = mmc.queue[0].sector;
+
+	if (sector != mmc.next_sector) {
+		end_read();
+		send_cmd(CMD18, sector);
+	}
+
+	rcvr_datablock(sectorBuf, 0, SECTOR_SIZE, 0);
+}
+
+void do_read_disc(void *address, uint32_t length, uint32_t offset, uint32_t sector, read_frag_cb callback)
+{
+	int i;
+
+	for (i = 0; i < mmc.items; i++)
+		if (mmc.queue[i].callback == callback)
+			return;
+
+	sector = offset / SECTOR_SIZE + sector;
+	offset = offset % SECTOR_SIZE;
+	length = MIN(length, SECTOR_SIZE - offset);
+
+	mmc.queue[i].address = address;
+	mmc.queue[i].length = length;
+	mmc.queue[i].offset = offset;
+	mmc.queue[i].sector = sector;
+	mmc.queue[i].callback = callback;
+	if (mmc.items++) return;
+
+	mmc_read_queued();
+}
+
+void tc_interrupt_handler(OSInterrupt interrupt, OSContext *context)
+{
+	if (isr_transferred < SECTOR_SIZE)
+		return;
+
+	OSMaskInterrupts(OS_INTERRUPTMASK(interrupt));
+	OSSetInterruptHandler(interrupt, TCIntrruptHandler);
+	exi_imm_read(2);
+	exi_deselect();
+	EXIUnlock(exi_channel);
+
+	void *address = mmc.queue[0].address;
+	uint32_t length = mmc.queue[0].length;
+	uint32_t offset = mmc.queue[0].offset;
+	uint32_t sector = mmc.queue[0].sector;
+	read_frag_cb callback = mmc.queue[0].callback;
+	mmc.last_sector = sector;
+	mmc.next_sector = sector + (1 << *VAR_SD_SHIFT);
+
+	if (address != VAR_SECTOR_BUF + offset)
+		memcpy(address, sectorBuf + offset, length);
+
+	if (--mmc.items) {
+		memcpy(mmc.queue, mmc.queue + 1, mmc.items * sizeof(*mmc.queue));
+		mmc_read_queued();
+	}
+
+	callback(address, length);
+}
+#endif
+
 u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 	// Try locking EXI bus
 	if(EXILock && !EXILock(exi_channel, EXI_DEVICE_0, 0)) {
@@ -169,7 +272,7 @@ u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 	// Send single block read command and the LBA we want to read at
 	send_cmd(CMD17, lba);
 	// Read block
-	rcvr_datablock(dst, startByte, numBytes);
+	rcvr_datablock(dst, startByte, numBytes, 1);
 	#else
 	// If we saved this sector
 	if(lba == mmc.last_sector) {
@@ -184,14 +287,14 @@ u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 	}
 	if(numBytes < SECTOR_SIZE) {
 		// Read half block
-		rcvr_datablock(sectorBuf, 0, SECTOR_SIZE);
+		rcvr_datablock(sectorBuf, 0, SECTOR_SIZE, 1);
 		memcpy(dst, sectorBuf + startByte, numBytes);
 		// Save current LBA
 		mmc.last_sector = lba;
 	}
 	else {
 		// Read full block
-		rcvr_datablock(dst, 0, SECTOR_SIZE);
+		rcvr_datablock(dst, 0, SECTOR_SIZE, 1);
 		// If we're reusing the sector buffer
 		if(dst == VAR_SECTOR_BUF) {
 			// Save current LBA
@@ -209,7 +312,7 @@ u32 do_read(void *dst, u32 len, u32 offset, u32 sectorLba) {
 #endif
 
 void end_read() {
-	#if SINGLE_SECTOR == 2
+	#if SINGLE_SECTOR == 2 || ISR_READ
 	if(mmc.next_sector) {
 		mmc.next_sector = 0;
 		// End the read by sending CMD12

cube/patches/sdgecko/sd_isr.S

@@ -0,0 +1,78 @@
+/* 
+ * Copyright (c) 2020, Extrems <extrems@extremscorner.org>
+ * 
+ * This file is part of Swiss.
+ * 
+ * Swiss is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ * 
+ * Swiss is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ * 
+ * You should have received a copy of the GNU General Public License
+ * with Swiss.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#define _LANGUAGE_ASSEMBLY
+#include "../../reservedarea.h"
+
+.section .text.external_interrupt_vector,"ax",@progbits
+	.balign 32
+	.globl isr_registers, isr_transferred
+isr_registers:
+	.long	0x0C006800
+isr_transferred:
+	.long	512
+
+	.globl external_interrupt_vector
+external_interrupt_vector:
+	mtsprg	0, r4
+	mtsprg	1, r5
+	mtsprg	2, r6
+	mfcr	r6
+	mtsprg	3, r6
+	lwz		r4, isr_registers - 0x80000000 (r0)
+	eciwx	r5, r0, r4
+	rlwinm	r6, r5, 1, 28, 28
+	and.	r6, r5, r6
+	beq		3f
+	andi.	r5, r5, (0x3FFF & ~0x80A) | (1 << 3)
+	lwz		r6, isr_transferred - 0x80000000 (r0)
+	cmpwi	r6, 512 - 4
+	bgt		3f
+	addi	r6, r6, 4
+	stw		r6, isr_transferred - 0x80000000 (r0)
+	beq		1f
+	cmpwi	r6, 0
+	ecowx	r5, r0, r4
+	ble		2f
+1:	li		r5, 4*4
+	eciwx	r5, r5, r4
+	stw		r5, VAR_SECTOR_BUF - 4 (r6)
+	beq		3f
+2:	li		r5, 4*4
+	li		r6, -1
+	ecowx	r6, r5, r4
+	li		r5, 3*4
+	li		r6, ((4 - 1) << 4) | 0b01
+	ecowx	r6, r5, r4
+	lis		r4, 0x0C00
+	li		r5, 0x3000
+	eciwx	r5, r5, r4
+	rlwinm.	r6, r5, 0, 16, 14
+	bne		3f
+	mfsprg	r6, 3
+	mtcr	r6
+	mfsprg	r6, 2
+	mfsprg	r5, 1
+	mfsprg	r4, 0
+	rfi
+3:	mfsprg	r6, 3
+	mtcr	r6
+	mfsprg	r6, 2
+	mfsprg	r5, 1
+	ba		0x00000504