Rework interfaces

We now use a portion of the ucode itself for version information
and to configure the base of SRAM, this will allow the driver
to use a pool allocator with less constraints.

We put the status register as part of the command register, this
speeds things up a bit and remove the need to clear it from
the host.

We also turn some SRAM stores from byte to long stores, they seem
to be faster (still room for improvement)

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

1 files changed, 48 insertions, 28 deletions

diff --git a/cf-code/cf-fsi-fw.S b/cf-code/cf-fsi-fw.S
index aa6a388..b771b79 100644
--- a/cf-code/cf-fsi-fw.S
+++ b/cf-code/cf-fsi-fw.S
@@ -161,13 +161,37 @@ _vecs:
 	.endr
 
 	/*
-	 * Main entry point
+	 * Header info
 	 */
 	.org	0x400
+_header_info:
+	.short	SYS_SIG			/* 0x00 */
+	.short	FW_VERSION		/* 0x02 */
+	.byte	API_VERSION_MAJ		/* 0x04 */
+	.byte	API_VERSION_MIN		/* 0x05 */
+	.byte	0,0			/* 0x06 pad */
+#ifdef ENABLE_TRACE
+	.long	FW_OPTION_TRACE_EN	/* 0x08 */
+#else
+	.long	0
+#endif
+
+	/*
+	 * Config area
+	*/
+	.org	0x400 + HDR_CMD_STAT_AREA
+_cmd_stat_base:
+	.long	SRAM_BASE_BE
+
+	/*
+	 * Main entry point
+	 */
+	.org	0x500
 	.global	_start
 _start:
 	/* Get base addresses */
-	movea.l	#SRAM_BASE_BE,%a1
+	lea	%pc@(_cmd_stat_base),%a0
+	movea.l	%a0@(0),%a1
 	movea.l	#GPIO_BASE,%a4
 	movea.l	%a4,%a5
 	add.l	#CLOCK_GPIO_VREG,%a5
@@ -240,18 +264,8 @@ _start:
 	/* Configure GPIOs to output */
 	bsr	config_gpio_out
 
-	/* Populate version & signature */
-	move.w	#SYS_SIG,%a1@(SYS_SIG_REG)
-	move.b	#FW_VERSION,%a1@(FW_VERS_REG)
-
-	/* This must happen last as it's the signal to the host
-	 * that we are ready
-	 */
-#ifdef ENABLE_TRACE
-	move.b	#(API_VERSION + API_VERSION_TRACE_EN),%a1@(API_VERS_REG)
-#else
-	move.b	#API_VERSION,%a1@(API_VERS_REG)
-#endif
+	/* Mark command complete, indicates we are ready */
+	move.b	#STAT_COMPLETE,%a1@(CMD_STAT_REG)
 
 	/*
 	 * Main command loop
@@ -266,7 +280,7 @@ main_loop:
 	 */
 1:	move.b	%a1@(ARB_REG),%d2
 	bne	arbitration_request
-	move.l	%a1@(CMD_REG),%d2
+	move.l	%a1@(CMD_STAT_REG),%d2
 	tst.b	%d2
 	bne	command_request
 	stop	#0x2000
@@ -301,11 +315,9 @@ command_request:
 	/* Mask interrupts */
 	move.w	#0x2007,%sr
 
-	/* Mark ourselves as sending a command */
-	move.b	#STAT_SENDING,%a1@(STAT_REG)
-
-	/* Clear command register */
-	move.b	#CMD_NONE,%a1@(CMD_REG + 3)
+	/* Clear the command/status register */
+	moveq.l	#0,%d0
+	move.l	%d0,%a1@(CMD_STAT_REG)
 
 	/* Start command ? */
 	cmpi.b	#CMD_COMMAND,%d2
@@ -320,7 +332,7 @@ command_request:
 	beq	start_idle_clocks
 
 	/* Error */
-	move.b	#STAT_ERR_INVAL_CMD,%a1@(STAT_REG)
+	move.b	#STAT_ERR_INVAL_CMD,%a1@(CMD_STAT_REG)
 	bra	main_loop
 
 	/*
@@ -389,7 +401,7 @@ start_command:
 	beq	1f
 	subq.l	#1,%d3
 	bne	0b
-	move.b	#STAT_ERR_MTOE,%a1@(STAT_REG)
+	move.b	#STAT_ERR_MTOE,%a1@(CMD_STAT_REG)
 	bra	send_delay
 
 1:	/* Got start bit, clock in slave ID and response tag */
@@ -454,7 +466,9 @@ start_command:
 	move.b	%d4,%a1@(STAT_RCRC)
 
 	/* Mark command complete */
-	move.b	#STAT_COMPLETE,%a1@(STAT_REG)
+	moveq.l	#STAT_COMPLETE,%d0
+	byterev	%d0
+	move.l	%d0,%a1@(CMD_STAT_REG)
 
 send_delay:
 	/* Send delay after every command */
@@ -485,8 +499,11 @@ start_break:
 	move.l	#FSI_POST_BREAK_CLOCKS,%d3
 	clock_out_zeros %d3
 
-	/* That's it */
-	move.b	#STAT_COMPLETE,%a1@(STAT_REG)
+	/* Mark command complete */
+	moveq.l	#STAT_COMPLETE,%d0
+	byterev	%d0
+	move.l	%d0,%a1@(CMD_STAT_REG)
+
 	bra	main_loop
 
 start_idle_clocks:
@@ -498,8 +515,11 @@ start_idle_clocks:
 	move.b	%d2,%d3
 	clock_out_zeros %d3
 
-	/* That's it */
-	move.b	#STAT_COMPLETE,%a1@(STAT_REG)
+	/* Mark command complete */
+	moveq.l	#STAT_COMPLETE,%d0
+	byterev	%d0
+	move.l	%d0,%a1@(CMD_STAT_REG)
+
 	bra	main_loop
 
 config_gpio_out:
@@ -549,7 +569,7 @@ bad_exception:
 	sub.l	%d1,%d0
 	lsr.l	#2,%d0
 	move.b	%d0,%a1@(BAD_INT_VEC)
-	move.b	#STAT_ERR_INVAL_IRQ,%a1@(STAT_REG)
+	move.b	#STAT_ERR_INVAL_IRQ,%a1@(CMD_STAT_REG)
 	halt
 
 	/* Bad exception stubs */