core: finalize TLB management in timer callbacks (STABLE)

This change enables interrupts within timer callbacks, making it
possible to load pages to MMU while handling a timer underflow. The call
to TLB_LoadPTEH() has been moved directly into the VBR handler to avoid
jumping to ILRAM for a short call on SH4.

The TMU and ETMU handlers have been changed to callback through a new
function gint_inth_callback() that saves the user bank and a few
registers, then invokes the callback with interrupts enabled and in user
bank; until now, callbacks were invoked with interrupts disabled and in
kernel bank. Note that IMASK is still set so a callback can only be
interrupted by a high-priority interrupt.

A timer_wait() function has also been added to simplify tests that
involve timers. Finally, the priority level of the TMU0 underflow
interrupt has been set to 13 (as per the comments) instead of 7.

This version is the first stable version that handles TLB misses
transparently for large add-ins. It is suitable for every gint
application.

TODO

@@ -1,5 +1,4 @@
 For the 2.1.0 release:
-* core:      use gint_switch() to handle TLB misses
 * core:      the four basic memory functions (with automated tests)
 * bopti:     remove the deprecated image_t definition
 * project:   remove the compat branch
@@ -12,6 +11,7 @@ Issues:
 * #10        support fx-CG 20
 
 Extensions on existing code:
+* bopti:     try to display fullscreen images with TLB access + DMA on fxcg50
 * gray:      add gprint()
 * gray:      double-buffer gray settings and unify d* with g*
 * topti:     support unicode fonts

include/gint/timer.h

@@ -141,6 +141,12 @@ void timer_pause(int timer);
    @timer  Timer id, as returned by timer_setup() */
 void timer_stop(int timer);
 
+/* timer_wait() - wait for a timer to stop
+   Waits until the specified timer stops running. If the timer is not running,
+   returns immediately. The timer might not be free if it has just been paused
+   instead of stopped entirely. */
+void timer_wait(int timer);
+
 //---
 //	Predefined timer callbacks
 //---

src/core/inth.S

@@ -11,6 +11,8 @@
 .global _gint_inth_7705
 #endif
 
+.global _gint_inth_callback
+
 .section .gint.blocks, "ax"
 .align 4
 
@@ -152,3 +154,70 @@ _inth_remap:
 	.byte	0x2f, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff
 
 #endif
+
+/* CALLBACK HELPER
+   This function implements the callback with context saving. It is a general
+   function and does not need to reside in VBR space which is block-structured.
+   This function saves registers r0_bank...r7_bank, enables interrupts,
+   switches back to user bank and executes the callback. It does not save other
+   registers (pr/mach/macl/gbr) which are managed by the handler entry. */
+
+.section .gint.mapped
+
+/* gint_inth_callback()
+   Calls the specified function with the given argument after saving the user
+   context, enabling interrupts and going to user bank.
+
+   @r4  Callback function (volatile void * -> int)
+   @r5  Argument (volatile void *)
+   Returns the return value of the callback (int). */
+_gint_inth_callback:
+	stc.l	r0_bank, @-r15
+	stc.l	r1_bank, @-r15
+	stc.l	r2_bank, @-r15
+	stc.l	r3_bank, @-r15
+	stc.l	r4_bank, @-r15
+	stc.l	r5_bank, @-r15
+	stc.l	r6_bank, @-r15
+	stc.l	r7_bank, @-r15
+	stc.l	spc, @-r15
+	stc.l	ssr, @-r15
+	stc.l	sr, @-r15
+
+	/* Enable interrupts and go back to user bank. SR.IMASK is already set
+	   to the level of the current interrupt, which makes sure we can only
+	   be re-interrupted by something with a higher priority. */
+	mov.l	.SR_clear_RB_BL, r1
+	stc	sr, r0
+	and	r1, r0
+	ldc	r0, sr
+
+	/* We are now in the user bank, but we can still use r0..r7 as the
+	   important values have been saved on the stack. Call back. */
+	stc	r4_bank, r0
+	sts.l	pr, @-r15
+	jsr	@r0
+	stc	r5_bank, r4
+	lds.l	@r15+, pr
+
+	/* We want to forward the return value to the system bank */
+	ldc	r0, r0_bank
+
+	/* Restore the previous status register and the registers of the
+	   interrupted procedure. */
+	ldc.l	@r15+, sr
+	ldc.l	@r15+, ssr
+	ldc.l	@r15+, spc
+	ldc.l	@r15+, r7_bank
+	ldc.l	@r15+, r6_bank
+	ldc.l	@r15+, r5_bank
+	ldc.l	@r15+, r4_bank
+	ldc.l	@r15+, r3_bank
+	ldc.l	@r15+, r2_bank
+	ldc.l	@r15+, r1_bank
+	rts
+	ldc.l	@r15+, r0_bank
+
+.align 4
+.SR_clear_RB_BL:
+	.long	~((1 << 29) | (1 << 28))

src/core/syscalls.S

@@ -9,8 +9,7 @@
 **	* File system, because it's a mess and we might ruin the ROM.
 */
 
-/* TLB management */
+.section .pretext
-.global ___TLB_LoadPTEH
 
 /* Dynamic allocation */
 .global _malloc
@@ -51,26 +50,6 @@
 
 #define syscall(id) syscall_(id, syscall_table)
 
-/*** Special syscalls that must remain mapped ***/
-
-.section .gint.mapped
-
-#ifdef FX9860G
-___TLB_LoadPTEH:
-	syscall_(0x0003, 2f)
-2:	.long 0x80010070
-#endif
-
-#ifdef FXCG50
-___TLB_LoadPTEH:
-	syscall_(0x000c, 2f)
-2:	.long 0x80020070
-#endif
-
-/*** Normal syscalls in ROM ***/
-
-.section .pretext
-
 #ifdef FX9860G
 
 /* Dynamic allocation */

src/core/tlbh.S

@@ -28,8 +28,15 @@ test_tea:
 
 map:
 	/* If TEA is mappable, map a page and return */
-	mov.l	.TLB_LoadPTEH, r0
+	#ifdef FX9860G
-	jsr	@r0
+	mov	#3, r0
+	#endif
+	#ifdef FXCG50
+	mov	#12, r0
+	#endif
+
+	mov.l	.syscall, r2
+	jsr	@r2
 	nop
 
 	lds.l	@r15+, macl
@@ -66,5 +73,12 @@ panic:
 	.long	0x00300000
 .max_mapped_rom:
 	.long	0x00300000 + _srom
-.TLB_LoadPTEH:
+
-	.long	___TLB_LoadPTEH
+#ifdef FX9860G
+.syscall:
+	.long	0x80010070
+#endif
+#ifdef FXCG50
+.syscall:
+	.long	0x80020070
+#endif

src/core/vbr.s

@@ -11,8 +11,8 @@
    priority registers or the interrupt masks, and make sure that all the
    interrupts that it leaves enabled are handled by the new VBR handlers.
 
-   @arg  vbr		New VBR address (uint32_t)
+   @r4  vbr        New VBR address (uint32_t)
-   @arg  configure	Configuration function (void -> void)
+   @r5  configure  Configuration function (void -> void)
    Returns the previous VBR address. */
 _gint_setvbr:
 	mov.l	r9, @-r15

src/tmu/inth.s

@@ -40,9 +40,8 @@ _inth_tmu_0:
 	mova	.storage0, r0
 	mov	#0, r1
 
-	/*** This is the first shared section ***/
+/*** This is the first shared section ***/
-
+.shared1:
-.clearflag:
 	sts.l	pr, @-r15
 	mov.l	r1, @-r15
 
@@ -54,32 +53,31 @@ _inth_tmu_0:
 	mov.w	r2, @r1
 
 	/* Invoke the callback function and pass the argument */
-	mov.l	@r0, r1
+	mov.l	.gint_inth_callback_1, r1
+	mov.l	@r0, r4
 	jsr	@r1
-	mov.l	@(4, r0), r4
+	mov.l	@(4, r0), r5
 
-	/* Prepare stopping the timer and jump to second section */
+	/* Jump to second section */
+	mov.l	.timer_stop_1, r1
+	bra	.shared2
 	mov.l	@r15+, r4
-	mov.l	.timer_stop, r1
-	bra	.stoptimer
-	nop
 
 /* SECOND GATE - TMU1 entry and stop timer */
 _inth_tmu_1:
 	mova	.storage1, r0
-	bra	.clearflag
+	bra	.shared1
 	mov	#1, r1
 
-	/*** This is the second shared section ***/
+/*** This is the second shared section ***/
-
+.shared2:
-.stoptimer:
 	/* Stop the timer if the return value is not zero */
 	tst	r0, r0
-	bt	.end
+	bt	.shared3
 	jsr	@r1
 	nop
 
-.end:
+.shared3:
 	lds.l	@r15+, pr
 	rts
 	nop
@@ -89,10 +87,13 @@ _inth_tmu_1:
 /* THIRD GATE - TMU2 entry and storage for TMU0 */
 _inth_tmu_2:
 	mova	.storage2, r0
-	bra	.clearflag
+	bra	.shared1
 	mov	#2, r1
 
-	.zero	14
+	.zero	10
+
+.gint_inth_callback_1:
+	.long	_gint_inth_callback
 
 .storage0:
 	.long	0		 /* Callback: Configured dynamically */
@@ -104,8 +105,8 @@ _inth_tmu_storage:
 
 .mask:
 	.long	0x0000feff
-.timer_stop:
+.timer_stop_1:
-	.long	_timer_stop	/* gint's function from <gint/timer.h> */
+	.long	_timer_stop
 
 .storage1:
 	.long	0 		/* Callback: Configured dynamically */
@@ -142,27 +143,24 @@ _inth_tmu_storage:
 
 /* FIRST GATE - ETMU2 entry, invoke callback and prepare clear flag */
 _inth_etmu2:
-	/* Warning: the size of the following section (4 bytes) is hardcoded in
-	   the jump in _inth_etmux */
 	mova	.storage_etmu2, r0
 	mov	#5, r1
 
-.extra_callback:
+.shared4:
 	sts.l	pr, @-r15
-	mov.l	r1, @-r15
+	mov.l	r0, @-r15
 
-	/* Invoke the callback function */
+	/* Clear interrupt flag */
-	mov.l	@r0, r1
+	mov.l	.timer_clear, r2
-	jsr	@r1
+	jsr	@r2
-	mov.l	@(4, r0), r4
+	mov	r1, r4
 
-	/* Load timer ID and forward the callback's return value */
+	/* Prepare invoking the callback function */
-	mov.l	.timer_clear, r1
+	mov.l	@r15+, r0
-	mov.l	@r15+, r4
+	mov.l	.gint_inth_callback_2, r1
+	mov.l	@r0, r4
 	bra	_inth_etmu_help
-	mov	r0, r5
+	mov.l	@(4, r0), r5
-
-	nop
 
 .storage_etmu2:
 	.long	0		/* Callback: Configured dynamically */
@@ -171,18 +169,28 @@ _inth_etmu2:
 /* SECOND GATE - Helper entry, invoke callback and stop timer if requested */
 _inth_etmu_help:
 
-	/* Clear the flag and possibly stop the timer */
+	/* Invoke callback; if return value is non-zero, stop timer */
+	jsr	@r1
+	nop
+	tst	r0, r0
+	bt	.shared5
+	mov.l	.timer_stop_2, r1
 	jsr	@r1
 	nop
 
+	/* Clear the flag and possibly stop the timer */
+
+.shared5:
 	lds.l	@r15+, pr
 	rts
 	nop
 
-	.zero	18
+.gint_inth_callback_2:
-
+	.long	_gint_inth_callback
 .timer_clear:
-	.long	_timer_clear	/* gint's function from src/tmu/tmu.c */
+	.long	_timer_clear
+.timer_stop_2:
+	.long	_timer_stop
 
 /* THIRD GATE - All other ETMU entries, deferred to the previous ones */
 _inth_etmux:
@@ -198,11 +206,11 @@ _inth_etmux:
 	nop
 
 	/* Offset from VBR where extra timer 2 is located:
-	   - 0x600 to reach the interrupt handlers
+	   * 0x600 to reach the interrupt handlers
-	   - 0x040 to jump over the entry gate
+	   * 0x040 to jump over the entry gate
-	   - 0x840 to reach the handler of ETMU2
+	   * 0x840 to reach the handler of ETMU2
-	   - 0x004 to skip its first instructions (the size is hardcoded) */
+	   * Skip over the first instructions */
-1:	.long	0xe84
+1:	.long	0xe80 + (.shared4 - _inth_etmu2)
 
 .id_etmux:
 	.long	0		/* Timer ID */

src/tmu/tmu.c

@@ -162,6 +162,22 @@ GMAPPED void timer_stop(int id)
 	}
 }
 
+/* timer_wait() - wait until a timer is stopped */
+void timer_wait(int id)
+{
+	if(id < 3)
+	{
+		tmu_t *T = &TMU[id];
+		/* Sleep if an interruption will wake us up */
+		while(*TSTR & (1 << id)) if(T->TCR.UNIE) sleep();
+	}
+	else
+	{
+		etmu_t *T = &ETMU[id];
+		while(T->TSTR) if(T->TCR.UNIE) sleep();
+	}
+}
+
 //---
 //	Predefined timer callbacks
 //---
@@ -178,15 +194,12 @@ GMAPPED int timer_timeout(volatile void *arg)
 //	Low-level functions
 //---
 
-/* timer_clear() - clear an ETMU flag and possibly stop the timer
+/* timer_clear() - clear an ETMU flag
-   @timer  Timer ID, must be an ETMU
+   @timer  Timer ID, must be an ETMU */
-   @stop   Non-zero to stop the timer */
+GMAPPED void timer_clear(int id)
-GMAPPED void timer_clear(int id, int stop)
 {
 	do ETMU[id-3].TCR.UNF = 0;
 	while(ETMU[id-3].TCR.UNF);
-
-	if(stop) timer_stop(id);
 }
 
 //---
@@ -270,7 +283,7 @@ static void init(void)
 	gint_inthandler(0xc60, inth_etmu_help, 32);
 
 	/* Enable TMU0 at level 13, TMU1 at level 11, TMU2 at level 9 */
-	gint_intlevel(0, 7);
+	gint_intlevel(0, 13);
 	gint_intlevel(1, 11);
 	gint_intlevel(2, 9);