Improve kernel stack handling on e500.

- Allocate thread0.td_kstack in pmap_bootstrap(), provide guard page
- Switch to thread0.td_kstack as soon as possible i.e. right after return
  from e500_init() and before mi_startup() happens
- Clean up temp stack area
- Other minor cosmetics in machdep.c

Obtained from:	Semihalf

3 files changed, 56 insertions, 24 deletions

diff --git a/sys/powerpc/booke/locore.S b/sys/powerpc/booke/locore.S
index 5cf3f6f..e4d8714 100644
--- a/sys/powerpc/booke/locore.S
+++ b/sys/powerpc/booke/locore.S
@@ -39,6 +39,8 @@
 #include <machine/tlb.h>
 #include <machine/bootinfo.h>
 
+#define TMPSTACKSZ	16384
+
 /*
  * This symbol is here for the benefit of kvm_mkdb, and is supposed to
  * mark the start of kernel text.
@@ -258,9 +260,9 @@ __start:
 /*
  * Setup a temporary stack
  */
-	lis	%r1, kstack0_space@ha
-	addi	%r1, %r1, kstack0_space@l
-	addi	%r1, %r1, (16384 - 512)
+	lis	%r1, tmpstack@ha
+	addi	%r1, %r1, tmpstack@l
+	addi	%r1, %r1, (TMPSTACKSZ - 8)
 
 /*
  * Intialise exception vector offsets
@@ -277,7 +279,13 @@ __start:
 	lis	%r4, _end@ha
 	addi	%r4, %r4, _end@l
 
-	bl	e500_init   /* Prepare e500 core */
+	bl	e500_init
+
+	/* Switch to thread0.td_kstack */
+	mr	%r1, %r3
+	li	%r3, 0
+	stw	%r3, 0(%r1)
+
 	bl	mi_startup  /* Machine independet part, does not return */
 
 /************************************************************************/
@@ -469,8 +477,8 @@ setfault:
 /************************************************************************/
 	.data
 	.align	4
-GLOBAL(kstack0_space)
-	.space	16384
+tmpstack:
+	.space	TMPSTACKSZ
 
 /*
  * Compiled KERNBASE locations
@@ -481,7 +489,6 @@ GLOBAL(kstack0_space)
 /*
  * Globals
  */
-#define	INTSTK		16384		/* 16K interrupt stack */
 #define	INTRCNT_COUNT	256		/* max(HROWPIC_IRQMAX,OPENPIC_IRQMAX) */
 
 GLOBAL(kernload)
diff --git a/sys/powerpc/booke/machdep.c b/sys/powerpc/booke/machdep.c
index 6c1be7f..4e9141e 100644
--- a/sys/powerpc/booke/machdep.c
+++ b/sys/powerpc/booke/machdep.c
@@ -154,10 +154,6 @@ extern unsigned char _end[];
 extern struct mem_region availmem_regions[];
 extern int availmem_regions_sz;
 
-extern void *trapcode, *trapsize;
-
-extern unsigned char kstack0_space[];
-
 extern void dcache_enable(void);
 extern void dcache_inval(void);
 extern void icache_enable(void);
@@ -185,11 +181,12 @@ SYSINIT(cpu, SI_SUB_CPU, SI_ORDER_FIRST, cpu_e500_startup, NULL);
 void print_kernel_section_addr(void);
 void dump_bootinfo(void);
 void dump_kenv(void);
-void e500_init(u_int32_t, u_int32_t, void *);
+u_int e500_init(u_int32_t, u_int32_t, void *);
 
 static void
 cpu_e500_startup(void *dummy)
 {
+	int indx, size;
 
 	/* Initialise the decrementer-based clock. */
 	decr_init();
@@ -198,20 +195,17 @@ cpu_e500_startup(void *dummy)
 	cpu_setup(PCPU_GET(cpuid));
 
 	printf("real memory  = %ld (%ld MB)\n", ptoa(physmem),
-			ptoa(physmem) / 1048576);
+	    ptoa(physmem) / 1048576);
 	realmem = physmem;
 
 	/* Display any holes after the first chunk of extended memory. */
 	if (bootverbose) {
-		int indx;
-
 		printf("Physical memory chunk(s):\n");
 		for (indx = 0; phys_avail[indx + 1] != 0; indx += 2) {
-			int size1 = phys_avail[indx + 1] - phys_avail[indx];
-
+			size = phys_avail[indx + 1] - phys_avail[indx];
 			printf("0x%08x - 0x%08x, %d bytes (%d pages)\n",
-					phys_avail[indx], phys_avail[indx + 1] - 1, size1,
-					size1 / PAGE_SIZE);
+			    phys_avail[indx], phys_avail[indx + 1] - 1, size,
+			    size / PAGE_SIZE);
 		}
 	}
 
@@ -328,7 +322,7 @@ bootinfo_eth(void)
 	return (eth);
 }
 
-void
+u_int
 e500_init(u_int32_t startkernel, u_int32_t endkernel, void *mdp)
 {
 	struct pcpu *pc;
@@ -445,9 +439,9 @@ e500_init(u_int32_t startkernel, u_int32_t endkernel, void *mdp)
 	init_param2(physmem);
 
 	/* Finish setting up thread0. */
-	thread0.td_kstack = (vm_offset_t)kstack0_space;
 	thread0.td_pcb = (struct pcb *)
-	    (thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
+	    ((thread0.td_kstack + thread0.td_kstack_pages * PAGE_SIZE -
+	    sizeof(struct pcb)) & ~15);
 	bzero((void *)thread0.td_pcb, sizeof(struct pcb));
 	pc->pc_curpcb = thread0.td_pcb;
 
@@ -480,7 +474,10 @@ e500_init(u_int32_t startkernel, u_int32_t endkernel, void *mdp)
 		printf("L1 I-cache %sabled\n",
 		    (csr & L1CSR1_ICE) ? "en" : "dis");
 
+	debugf("e500_init: SP = 0x%08x\n", ((uintptr_t)thread0.td_pcb - 16) & ~15);
 	debugf("e500_init: e\n");
+
+	return (((uintptr_t)thread0.td_pcb - 16) & ~15);
 }
 
 /* Initialise a struct pcpu. */
diff --git a/sys/powerpc/booke/pmap.c b/sys/powerpc/booke/pmap.c
index fde1593..a6fff41 100644
--- a/sys/powerpc/booke/pmap.c
+++ b/sys/powerpc/booke/pmap.c
@@ -897,8 +897,9 @@ mmu_booke_bootstrap(mmu_t mmu, vm_offset_t kernelstart, vm_offset_t kernelend)
 	int cnt, i, j;
 	u_int s, e, sz;
 	u_int phys_avail_count;
-	vm_size_t physsz, hwphyssz;
-	vm_offset_t kernel_pdir;
+	vm_size_t physsz, hwphyssz, kstack0_sz;
+	vm_offset_t kernel_pdir, kstack0;
+	vm_paddr_t kstack0_phys;
 
 	debugf("mmu_booke_bootstrap: entered\n");
 
@@ -1056,6 +1057,16 @@ mmu_booke_bootstrap(mmu_t mmu, vm_offset_t kernelstart, vm_offset_t kernelend)
 	availmem_regions_sz = cnt;
 
 	/*******************************************************/
+	/* Steal physical memory for kernel stack from the end */
+	/* of the first avail region                           */
+	/*******************************************************/
+	kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
+	kstack0_phys = availmem_regions[0].mr_start +
+	    availmem_regions[0].mr_size;
+	kstack0_phys -= kstack0_sz;
+	availmem_regions[0].mr_size -= kstack0_sz;
+
+	/*******************************************************/
 	/* Fill in phys_avail table, based on availmem_regions */
 	/*******************************************************/
 	phys_avail_count = 0;
@@ -1125,6 +1136,23 @@ mmu_booke_bootstrap(mmu_t mmu, vm_offset_t kernelstart, vm_offset_t kernelend)
 	/*******************************************************/
 	/* Final setup */
 	/*******************************************************/
+	/* Enter kstack0 into kernel map, provide guard page */
+	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
+	thread0.td_kstack = kstack0;
+	thread0.td_kstack_pages = KSTACK_PAGES;
+
+	debugf("kstack_sz = 0x%08x\n", kstack0_sz);
+	debugf("kstack0_phys at 0x%08x - 0x%08x\n",
+	    kstack0_phys, kstack0_phys + kstack0_sz);
+	debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
+	
+	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
+	for (i = 0; i < KSTACK_PAGES; i++) {
+		mmu_booke_kenter(mmu, kstack0, kstack0_phys);
+		kstack0 += PAGE_SIZE;
+		kstack0_phys += PAGE_SIZE;
+	}
+
 	/* Initialize TLB0 handling. */
 	tlb0_init();