Refactor PowerPC (especially AIM) init sequence to be less baroque.

MFC after:	2 months

11 files changed, 92 insertions, 178 deletions

diff --git a/sys/conf/ldscript.powerpc b/sys/conf/ldscript.powerpc
index efc0dfd..b190dc1 100644
--- a/sys/conf/ldscript.powerpc
+++ b/sys/conf/ldscript.powerpc
@@ -11,6 +11,7 @@ SECTIONS
   /* Read-only sections, merged into text segment: */
 
   . = kernbase + SIZEOF_HEADERS;
+  PROVIDE (begin = . - SIZEOF_HEADERS);
 
   .text      :
   {
diff --git a/sys/conf/ldscript.powerpc64 b/sys/conf/ldscript.powerpc64
index ce23551..dba7455 100644
--- a/sys/conf/ldscript.powerpc64
+++ b/sys/conf/ldscript.powerpc64
@@ -11,6 +11,7 @@ SECTIONS
   /* Read-only sections, merged into text segment: */
 
   . = kernbase + SIZEOF_HEADERS;
+  PROVIDE (begin = . - SIZEOF_HEADERS);
 
   .text      :
   {
@@ -68,7 +69,7 @@ SECTIONS
   .toc1      : ALIGN(8) { *(.toc1) }
   .opd       : ALIGN(8) { KEEP (*(.opd)) }
   .branch_lt : ALIGN(8) { *(.branch_lt) }
-  .got       : ALIGN(8) { *(.got .toc) }
+  .got       : ALIGN(8) { __tocbase = .; *(.got .toc) }
 
   .dynamic        : { *(.dynamic) }
   /* Put .ctors and .dtors next to the .got2 section, so that the pointers
diff --git a/sys/powerpc/aim/locore32.S b/sys/powerpc/aim/locore32.S
index 8cee665..6e462a0 100644
--- a/sys/powerpc/aim/locore32.S
+++ b/sys/powerpc/aim/locore32.S
@@ -76,29 +76,19 @@
         .globl  kernbase
         .set    kernbase, KERNBASE
 
-#define	TMPSTKSZ	8192		/* 8K temporary stack */
-
 /*
  * Globals
  */
 	.data
+	.align 3
+GLOBAL(__startkernel)
+	.long	begin
+GLOBAL(__endkernel)
+	.long	end
 	.align	4
+#define	TMPSTKSZ	8192		/* 8K temporary stack */
 GLOBAL(tmpstk)
 	.space	TMPSTKSZ
-GLOBAL(esym)
-	.long	0			/* end of symbol table */
-
-#define	INTRCNT_COUNT	256		/* max(HROWPIC_IRQMAX,OPENPIC_IRQMAX) */
-GLOBAL(intrnames)
-	.space	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-GLOBAL(sintrnames)
-	.long	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-
-	.align 4
-GLOBAL(intrcnt)
-	.space	INTRCNT_COUNT * 4 * 2
-GLOBAL(sintrcnt)
-	.long	INTRCNT_COUNT * 4 * 2
 
 	.text
 	.globl	btext
@@ -142,43 +132,9 @@ __start:
 	cmplw	8,9
 	blt	2b
 	
-	/* Save the argument pointer and length */
-	mr	20,6
-	mr	21,7
-
-	lis	8,openfirmware_entry@ha
-	stw	5,openfirmware_entry@l(8) /* save client interface handler */
-
 	lis	1,(tmpstk+TMPSTKSZ-16)@ha
 	addi	1,1,(tmpstk+TMPSTKSZ-16)@l
 
-	mfmsr	0
-	lis	9,ofmsr@ha
-	stwu	0,ofmsr@l(9)
-
-	mfsprg0	0			/* save SPRG0-3 */
-	stw	0,4(9)			/* ofmsr[1] = sprg0 */
-	mfsprg1 0
-	stw	0,8(9)			/* ofmsr[2] = sprg1 */
-	mfsprg2 0
-	stw	0,12(9)			/* ofmsr[3] = sprg2 */
-	mfsprg3 0
-	stw	0,16(9)			/* ofmsr[4] = sprg3 */
-	
-	bl	OF_initial_setup
-
-	lis	3,kernel_text@ha
-	addi	3,3,kernel_text@l
-
-	lis	4,end@ha
-	addi	4,4,end@l
-	add	4,4,3
-	mr	5,4
-
-	/* Restore the argument pointer and length */
-	mr	6,20
-	mr	7,21
-
 	bl	powerpc_init
 	mr	%r1, %r3
 	li	%r3, 0
diff --git a/sys/powerpc/aim/locore64.S b/sys/powerpc/aim/locore64.S
index 0c26e01..4865e75 100644
--- a/sys/powerpc/aim/locore64.S
+++ b/sys/powerpc/aim/locore64.S
@@ -68,36 +68,28 @@
 /* Locate the per-CPU data structure */
 #define GET_CPUINFO(r)  \
         mfsprg0  r
+#define GET_TOCBASE(r)  \
+	li	r,TRAP_TOCBASE;	/* Magic address for TOC */ \
+	ld	r,0(r)
 
-/*
- * Compiled KERNBASE location and the kernel load address
- */
-        .globl  kernbase
-        .set    kernbase, KERNBASE
-
-#define	TMPSTKSZ	16384		/* 16K temporary stack */
+/* Glue for linker script */
+.globl  kernbase
+.set    kernbase, KERNBASE
 
 /*
  * Globals
  */
 	.data
+	.align 3
+GLOBAL(__startkernel)
+	.llong	begin
+GLOBAL(__endkernel)
+	.llong	end
+
 	.align	4
+#define	TMPSTKSZ	16384		/* 16K temporary stack */
 GLOBAL(tmpstk)
 	.space	TMPSTKSZ
-GLOBAL(esym)
-	.llong	0			/* end of symbol table */
-
-#define	INTRCNT_COUNT	256		/* max(HROWPIC_IRQMAX,OPENPIC_IRQMAX) */
-GLOBAL(intrnames)
-	.space	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-GLOBAL(sintrnames)
-	.quad	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-
-	.align 4
-GLOBAL(intrcnt)
-	.space	INTRCNT_COUNT * 4 * 2
-GLOBAL(sintrcnt)
-	.quad	INTRCNT_COUNT * 4 * 2
 
 	.text
 	.globl	btext
@@ -113,90 +105,52 @@ kernel_text:
 /*
  * Startup entry.  Note, this must be the first thing in the text
  * segment!
+ *
+ * Calling convention:
+ * r3: Flattened Device Tree pointer (or zero)
+ * r4: ignored
+ * r5: OF client interface pointer (or zero)
+ * r6: Loader metadata pointer (or zero)
  */
 	.text
 ASENTRY_NOPROF(__start)
-	li	8,0
-	li	9,0x100
-	mtctr	9
-1:
-	dcbf	0,8
-	icbi	0,8
-	addi	8,8,0x20
-	bdnz	1b
-	sync
-	isync
-
-	/* Save the argument pointer and length */
-	mr	20,6
-	mr	21,7
-
-	lis	8,openfirmware_entry@ha
-	std	5,openfirmware_entry@l(8) /* save client interface handler */
+	/* Set up the TOC pointer */
+	b	0f
+	.align 3
+0:	nop
+	bl	1f
+	.llong	__tocbase + 0x8000
+1:	mflr	%r2
+	ld	%r2,0(%r2)
 
 	/* Set up the stack pointer */
-	lis	1,(tmpstk+TMPSTKSZ-48)@ha
-	addi	1,1,(tmpstk+TMPSTKSZ-48)@l
-
-	/* Set up the TOC pointer */
-	lis	2,tocbase@ha
-	ld	2,tocbase@l(2)
-
-	mfmsr	0
-	lis	9,ofmsr@ha
-	stdu	0,ofmsr@l(9)
-
-	mfsprg0	0			/* save SPRG0-3 */
-	std	0,8(9)			/* ofmsr[1] = sprg0 */
-	mfsprg1 0
-	std	0,16(9)			/* ofmsr[2] = sprg1 */
-	mfsprg2 0
-	std	0,24(9)			/* ofmsr[3] = sprg2 */
-	mfsprg3 0
-	std	0,32(9)			/* ofmsr[4] = sprg3 */
+	lis	%r1,(tmpstk+TMPSTKSZ-48)@ha
+	addi	%r1,%r1,(tmpstk+TMPSTKSZ-48)@l
 
 	/* Switch to 64-bit mode */
-	mfmsr	9
-	li	8,1
-	insrdi	9,8,1,0
-	mtmsrd	9
+	mfmsr	%r9
+	li	%r8,1
+	insrdi	%r9,%r8,1,0
+	mtmsrd	%r9
 	isync
 	
-	bl	OF_initial_setup
-	nop
-
-	lis	3,kernbase@ha
-	addi	3,3,kernbase@l
-
-	lis	4,end@ha
-	addi	4,4,end@l
-	add	4,4,3
-	mr	5,4
-
-	/* Restore the argument pointer and length */
-	mr	6,20
-	mr	7,21
-
+	/* Begin CPU init */
+	mr	%r4,%r2 /* Replace ignored r4 with tocbase for trap handlers */
 	bl	powerpc_init
 	nop
+
+	/* Set stack pointer to new value and branch to mi_startup */
 	mr	%r1, %r3
 	li	%r3, 0
 	std	%r3, 0(%r1)
 	bl	mi_startup
 	nop
+
+	/* If this returns (it won't), go back to firmware */
 	b	OF_exit
 	nop
 
 /*
- * PPC64 ABI TOC base
- */
-
-        .align  3
-	.globl	tocbase
-tocbase:
-        .llong  .TOC.@tocbase
-
-/*
  * int setfault()
  *
  * Similar to setjmp to setup for handling faults on accesses to user memory.
diff --git a/sys/powerpc/aim/machdep.c b/sys/powerpc/aim/machdep.c
index d1413a2..fdbe48b 100644
--- a/sys/powerpc/aim/machdep.c
+++ b/sys/powerpc/aim/machdep.c
@@ -223,7 +223,7 @@ cpu_startup(void *dummy)
 	vm_pager_bufferinit();
 }
 
-extern char	kernel_text[], _end[];
+extern vm_offset_t	__startkernel, __endkernel;
 
 #ifndef __powerpc64__
 /* Bits for running on 64-bit systems in 32-bit mode. */
@@ -244,13 +244,12 @@ extern void	*dblow, *dbsize;
 extern void	*imisstrap, *imisssize;
 extern void	*dlmisstrap, *dlmisssize;
 extern void	*dsmisstrap, *dsmisssize;
-char 		save_trap_init[0x2f00];		/* EXC_LAST */
 
 uintptr_t
-powerpc_init(vm_offset_t startkernel, vm_offset_t endkernel,
-    vm_offset_t basekernel, void *mdp)
+powerpc_init(vm_offset_t fdt, vm_offset_t toc, vm_offset_t ofentry, void *mdp)
 {
 	struct		pcpu *pc;
+	vm_offset_t	startkernel, endkernel;
 	void		*generictrap;
 	size_t		trap_offset;
 	void		*kmdp;
@@ -273,8 +272,12 @@ powerpc_init(vm_offset_t startkernel, vm_offset_t endkernel,
 	trap_offset = 0;
 	cacheline_warn = 0;
 
-	/* Save trap vectors. */
-	ofw_save_trap_vec(save_trap_init);
+	/* Store boot environment state */
+	OF_initial_setup((void *)fdt, NULL, (int (*)(void *))ofentry);
+
+	/* First guess at start/end kernel positions */
+	startkernel = __startkernel;
+	endkernel = __endkernel;
 
 #ifdef WII
 	/*
@@ -490,6 +493,9 @@ powerpc_init(vm_offset_t startkernel, vm_offset_t endkernel,
 	#else /* powerpc64 */
 	cpu_features |= PPC_FEATURE_64;
 	generictrap = &trapcode;
+
+	/* Set TOC base so that the interrupt code can get at it */
+	*((register_t *)TRAP_TOCBASE) = toc;
 	#endif
 
 	bcopy(&rstcode, (void *)(EXC_RST + trap_offset),  (size_t)&rstsize);
diff --git a/sys/powerpc/aim/trap_subr64.S b/sys/powerpc/aim/trap_subr64.S
index 0dc46b3..ec3c925 100644
--- a/sys/powerpc/aim/trap_subr64.S
+++ b/sys/powerpc/aim/trap_subr64.S
@@ -310,8 +310,7 @@ cpu_reset:
 	lis	%r1,(tmpstk+TMPSTKSZ-48)@ha	/* get new SP */
 	addi	%r1,%r1,(tmpstk+TMPSTKSZ-48)@l
 
-	lis	%r3,tocbase@ha
-	ld	%r2,tocbase@l(%r3)
+	GET_TOCBASE(%r2)
 	bl	CNAME(cpudep_ap_early_bootstrap) /* Set PCPU */
 	nop
 	lis	%r3,1@l
@@ -445,8 +444,7 @@ kern_slbtrap:
 	addi	%r1,%r1,PC_SLBSTACK-48+1024
 	li	%r2,~15
 	and	%r1,%r1,%r2
-	lis	%r3,tocbase@ha
-	ld	%r2,tocbase@l(%r3)
+	GET_TOCBASE(%r2)
 	mflr	%r3
 	andi.	%r3,%r3,0xff80
 	mfdar	%r4
@@ -683,8 +681,7 @@ k_trap:
 	FRAME_SETUP(PC_TEMPSAVE)
 /* Call C interrupt dispatcher: */
 trapagain:
-	lis	%r3,tocbase@ha
-	ld	%r2,tocbase@l(%r3)
+	GET_TOCBASE(%r2)
 	addi	%r3,%r1,48
 	bl	CNAME(powerpc_interrupt)
 	nop
@@ -711,8 +708,7 @@ CNAME(trapexit):
 	ori	%r3,%r3,PSL_EE@l
 	mtmsr	%r3
 	isync
-	lis	%r3,tocbase@ha
-	ld	%r2,tocbase@l(%r3)
+	GET_TOCBASE(%r2)
 	addi	%r3,%r1,48
 	bl	CNAME(ast)
 	nop
@@ -760,8 +756,7 @@ dbtrap:
 
 	FRAME_SETUP(PC_DBSAVE)
 /* Call C trap code: */
-	lis	%r3,tocbase@ha
-	ld	%r2,tocbase@l(%r3)
+	GET_TOCBASE(%r2)
 	addi	%r3,%r1,48
 	bl	CNAME(db_trap_glue)
 	nop
diff --git a/sys/powerpc/booke/locore.S b/sys/powerpc/booke/locore.S
index 514afa6..0fcca91 100644
--- a/sys/powerpc/booke/locore.S
+++ b/sys/powerpc/booke/locore.S
@@ -207,7 +207,7 @@ done_mapping:
  */
 	lis	%r1, tmpstack@ha
 	addi	%r1, %r1, tmpstack@l
-	addi	%r1, %r1, (TMPSTACKSZ - 8)
+	addi	%r1, %r1, (TMPSTACKSZ - 16)
 
 /*
  * Initialise exception vector offsets
@@ -367,7 +367,7 @@ bp_tlb1_end:
  */
 	lis	%r1, tmpstack@ha
 	addi	%r1, %r1, tmpstack@l
-	addi	%r1, %r1, (TMPSTACKSZ - 8)
+	addi	%r1, %r1, (TMPSTACKSZ - 16)
 
 /*
  * Initialise exception vector offsets
@@ -757,6 +757,8 @@ setfault:
 	.align	4
 tmpstack:
 	.space	TMPSTACKSZ
+tmpstackbound:
+	.space 10240	/* XXX: this really should not be necessary */
 
 /*
  * Compiled KERNBASE locations
@@ -764,20 +766,4 @@ tmpstack:
 	.globl	kernbase
 	.set	kernbase, KERNBASE
 
-/*
- * Globals
- */
-#define	INTRCNT_COUNT	256		/* max(HROWPIC_IRQMAX,OPENPIC_IRQMAX) */
-
-GLOBAL(intrnames)
-	.space	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-GLOBAL(sintrnames)
-	.long	INTRCNT_COUNT * (MAXCOMLEN + 1) * 2
-
-	.align 4
-GLOBAL(intrcnt)
-	.space	INTRCNT_COUNT * 4 * 2
-GLOBAL(sintrcnt)
-	.long	INTRCNT_COUNT * 4 * 2
-
 #include <powerpc/booke/trap_subr.S>
diff --git a/sys/powerpc/include/trap.h b/sys/powerpc/include/trap.h
index 211afcf..3b1ade5 100644
--- a/sys/powerpc/include/trap.h
+++ b/sys/powerpc/include/trap.h
@@ -123,6 +123,9 @@
 /* DTrace trap opcode. */
 #define EXC_DTRACE	0x7c810808
 
+/* Magic pointer to store TOC base for trap handlers on ppc64 */
+#define TRAP_TOCBASE	0x1f8
+
 #ifndef LOCORE
 struct	trapframe;
 struct	pcb;
diff --git a/sys/powerpc/ofw/ofw_machdep.c b/sys/powerpc/ofw/ofw_machdep.c
index cc44847..1c02fa3 100644
--- a/sys/powerpc/ofw/ofw_machdep.c
+++ b/sys/powerpc/ofw/ofw_machdep.c
@@ -66,8 +66,8 @@ extern register_t ofmsr[5];
 extern void	*openfirmware_entry;
 static void	*fdt;
 int		ofw_real_mode;
-extern char     save_trap_init[0x2f00];          /* EXC_LAST */
-char            save_trap_of[0x2f00];            /* EXC_LAST */
+char		save_trap_init[0x2f00];          /* EXC_LAST */
+char		save_trap_of[0x2f00];            /* EXC_LAST */
 
 int		ofwcall(void *);
 static int	openfirmware(void *args);
@@ -257,18 +257,30 @@ ofw_mem_regions(struct mem_region *memp, int *memsz,
 void
 OF_initial_setup(void *fdt_ptr, void *junk, int (*openfirm)(void *))
 {
+	ofmsr[0] = mfmsr();
+	#ifdef __powerpc64__
+	ofmsr[0] &= ~PSL_SF;
+	#endif
+	__asm __volatile("mfsprg0 %0" : "=&r"(ofmsr[1]));
+	__asm __volatile("mfsprg1 %0" : "=&r"(ofmsr[2]));
+	__asm __volatile("mfsprg2 %0" : "=&r"(ofmsr[3]));
+	__asm __volatile("mfsprg3 %0" : "=&r"(ofmsr[4]));
+
 	if (ofmsr[0] & PSL_DR)
 		ofw_real_mode = 0;
 	else
 		ofw_real_mode = 1;
 
 	fdt = fdt_ptr;
+	openfirmware_entry = openfirm;
 
 	#ifdef FDT_DTB_STATIC
 	/* Check for a statically included blob */
 	if (fdt == NULL)
 		fdt = &fdt_static_dtb;
 	#endif
+
+	ofw_save_trap_vec(save_trap_init);
 }
 
 boolean_t
diff --git a/sys/powerpc/powerpc/intr_machdep.c b/sys/powerpc/powerpc/intr_machdep.c
index cc7624f..b481279 100644
--- a/sys/powerpc/powerpc/intr_machdep.c
+++ b/sys/powerpc/powerpc/intr_machdep.c
@@ -127,6 +127,11 @@ static u_int nirqs = 0;		/* Allocated IRQs. */
 #endif
 static u_int stray_count;
 
+u_long intrcnt[INTR_VECTORS];
+char intrnames[INTR_VECTORS * MAXCOMLEN];
+size_t sintrcnt = sizeof(intrcnt);
+size_t sintrnames = sizeof(intrnames);
+
 device_t root_pic;
 
 #ifdef SMP
diff --git a/sys/powerpc/powerpc/vm_machdep.c b/sys/powerpc/powerpc/vm_machdep.c
index d684094..c816e32 100644
--- a/sys/powerpc/powerpc/vm_machdep.c
+++ b/sys/powerpc/powerpc/vm_machdep.c
@@ -99,11 +99,6 @@
 #include <vm/vm_map.h>
 #include <vm/vm_extern.h>
 
-#ifdef __powerpc64__
-extern uintptr_t tocbase;
-#endif
-
-
 /*
  * Finish a fork operation, with process p2 nearly set up.
  * Copy and update the pcb, set up the stack so that the child
@@ -149,7 +144,7 @@ cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
 	cf = (struct callframe *)tf - 1;
 	memset(cf, 0, sizeof(struct callframe));
 	#ifdef __powerpc64__
-	cf->cf_toc = tocbase;
+	cf->cf_toc = ((register_t *)fork_return)[1];
 	#endif
 	cf->cf_func = (register_t)fork_return;
 	cf->cf_arg0 = (register_t)td2;