1 files changed, 44 insertions, 80 deletions

diff --git a/sys/powerpc/booke/pmap.c b/sys/powerpc/booke/pmap.c
index 9eac669e0f7c..e6ce3e729cc6 100644
--- a/sys/powerpc/booke/pmap.c
+++ b/sys/powerpc/booke/pmap.c
@@ -237,11 +237,11 @@ static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
-static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
+static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t, int);
 
 static vm_size_t tsize2size(unsigned int);
 static unsigned int size2tsize(vm_size_t);
-static unsigned int ilog2(unsigned long);
+static unsigned long ilog2(unsigned long);
 
 static void set_mas4_defaults(void);
 
@@ -1619,10 +1619,16 @@ mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
 	debugf(" kernel pdir at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
 	    kernel_pdir, data_end);
 
+	/* Pre-round up to 1MB.  This wastes some space, but saves TLB entries */
+	data_end = roundup2(data_end, 1 << 20);
 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
+	debugf(" kernstart: %p\n", kernstart);
+	debugf(" kernsize: %lx\n", kernsize);
+
 	if (data_end - kernstart > kernsize) {
 		kernsize += tlb1_mapin_region(kernstart + kernsize,
-		    kernload + kernsize, (data_end - kernstart) - kernsize);
+		    kernload + kernsize, (data_end - kernstart) - kernsize,
+		    _TLB_ENTRY_MEM);
 	}
 	data_end = kernstart + kernsize;
 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
@@ -1819,7 +1825,7 @@ mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
 	 * Round so it fits into a single mapping.
 	 */
 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
-	    phys_avail[i + 1]);
+	    phys_avail[i + 1], _TLB_ENTRY_MEM);
 #endif
 
 	/*******************************************************/
@@ -3500,30 +3506,8 @@ mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
 #endif
 	res = (void *)va;
 
-	do {
-		sz = 1 << (ilog2(size) & ~1);
-		/* Align size to PA */
-		if (pa % sz != 0) {
-			do {
-				sz >>= 2;
-			} while (pa % sz != 0);
-		}
-		/* Now align from there to VA */
-		if (va % sz != 0) {
-			do {
-				sz >>= 2;
-			} while (va % sz != 0);
-		}
-		if (bootverbose)
-			printf("Wiring VA=%p to PA=%jx (size=%lx)\n",
-			    (void *)va, (uintmax_t)pa, (long)sz);
-		if (tlb1_set_entry(va, pa, sz,
-		    _TLB_ENTRY_SHARED | tlb_calc_wimg(pa, ma)) < 0)
-			return (NULL);
-		size -= sz;
-		pa += sz;
-		va += sz;
-	} while (size > 0);
+	if (tlb1_mapin_region(va, pa, size, tlb_calc_wimg(pa, ma)) != size)
+		return (NULL);
 
 	return (res);
 }
@@ -3864,7 +3848,7 @@ tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
 /*
  * Return the largest uint value log such that 2^log <= num.
  */
-static unsigned int
+static unsigned long
 ilog2(unsigned long num)
 {
 	long lz;
@@ -3952,69 +3936,49 @@ tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
 }
 
 /*
- * Map in contiguous RAM region into the TLB1 using maximum of
- * KERNEL_REGION_MAX_TLB_ENTRIES entries.
- *
- * If necessary round up last entry size and return total size
- * used by all allocated entries.
+ * Map in contiguous RAM region into the TLB1.
  */
-vm_size_t
-tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
+static vm_size_t
+tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size, int wimge)
 {
-	vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
-	vm_size_t mapped, pgsz, base, mask;
-	int idx, nents;
-
-	/* Round up to the next 1M */
-	size = roundup2(size, 1 << 20);
+	vm_offset_t base;
+	vm_size_t mapped, sz, ssize;
 
 	mapped = 0;
-	idx = 0;
 	base = va;
-	pgsz = 64*1024*1024;
-	while (mapped < size) {
-		while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
-			while (pgsz > (size - mapped))
-				pgsz >>= 2;
-			pgs[idx++] = pgsz;
-			mapped += pgsz;
-		}
+	ssize = size;
 
-		/* We under-map. Correct for this. */
-		if (mapped < size) {
-			while (pgs[idx - 1] == pgsz) {
-				idx--;
-				mapped -= pgsz;
-			}
-			/* XXX We may increase beyond out starting point. */
-			pgsz <<= 2;
-			pgs[idx++] = pgsz;
-			mapped += pgsz;
+	while (size > 0) {
+		sz = 1UL << (ilog2(size) & ~1);
+		/* Align size to PA */
+		if (pa % sz != 0) {
+			do {
+				sz >>= 2;
+			} while (pa % sz != 0);
 		}
-	}
-
-	nents = idx;
-	mask = pgs[0] - 1;
-	/* Align address to the boundary */
-	if (va & mask) {
-		va = (va + mask) & ~mask;
-		pa = (pa + mask) & ~mask;
-	}
-
-	for (idx = 0; idx < nents; idx++) {
-		pgsz = pgs[idx];
-		debugf("%u: %jx -> %jx, size=%jx\n", idx, (uintmax_t)pa,
-		    (uintmax_t)va, (uintmax_t)pgsz);
-		tlb1_set_entry(va, pa, pgsz,
-		    _TLB_ENTRY_SHARED | _TLB_ENTRY_MEM);
-		pa += pgsz;
-		va += pgsz;
+		/* Now align from there to VA */
+		if (va % sz != 0) {
+			do {
+				sz >>= 2;
+			} while (va % sz != 0);
+		}
+		/* Now align from there to VA */
+		if (bootverbose)
+			printf("Wiring VA=%p to PA=%jx (size=%lx)\n",
+			    (void *)va, (uintmax_t)pa, (long)sz);
+		if (tlb1_set_entry(va, pa, sz,
+		    _TLB_ENTRY_SHARED | wimge) < 0)
+			return (mapped);
+		size -= sz;
+		pa += sz;
+		va += sz;
 	}
 
 	mapped = (va - base);
 	if (bootverbose)
 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
-		    mapped, mapped - size);
+		    mapped, mapped - ssize);
+
 	return (mapped);
 }