Mapped kernel support for Broadcom XLP

index a842154..74c5f6d 100644
@@ -610,6 +610,7 @@ static void __init request_crashkernel(struct resource *res)
 static void __init arch_mem_init(char **cmdline_p)
 {
 	extern void plat_mem_setup(void);
+	extern void plat_mem_fixup(void);
 
 	/* call board setup routine */
 	plat_mem_setup();

@@ -653,6 +654,9 @@ static void __init arch_mem_init(char **cmdline_p)
 		pr_info("User-defined physical RAM map:\n");
 		print_memory_map();
 	}
+#if defined(CONFIG_CPU_XLP) || defined(CONFIG_CPU_XLR)
+	plat_mem_fixup();
+#endif
 
 	bootmem_init();
 #ifdef CONFIG_PROC_VMCORE

53 lines not on this page · raw message

index 135a38f..bf4f6d3 100644
@@ -65,18 +65,6 @@ static void nlm_linux_exit(void)
 		cpu_wait();
 }
 
-static void nlm_fixup_mem(void)
-{
-	const int pref_backup = 512;
-	int i;
-
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		boot_mem_map.map[i].size -= pref_backup;
-	}
-}
-
 static void __init xlp_init_mem_from_bars(void)
 {
 	uint64_t map[16];

@@ -115,8 +103,6 @@ void __init plat_mem_setup(void)
 		pr_info("Using DRAM BARs for memory map.\n");
 		xlp_init_mem_from_bars();
 	}
-	/* Calculate and setup wired entries for mapped kernel */
-	nlm_fixup_mem();
 }
 
 const char *get_system_type(void)

index d118b9a..714f6a3 100644
@@ -144,7 +144,6 @@ static void prom_add_memory(void)
 {
 	struct nlm_boot_mem_map *bootm;
 	u64 start, size;
-	u64 pref_backup = 512;	/* avoid pref walking beyond end */
 	int i;
 
 	bootm = (void *)(long)nlm_prom_info.psb_mem_map;

@@ -158,7 +157,7 @@ static void prom_add_memory(void)
 		if (i == 0 && start == 0 && size == 0x0c000000)
 			size = 0x0ff00000;
 
-		add_memory_region(start, size - pref_backup, BOOT_MEM_RAM);
+		add_memory_region(start, size, BOOT_MEM_RAM);
 	}
 }
 

50 lines not on this page · raw message

new file mode 100644
index 0000000..6f32f07
@@ -0,0 +1,25 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1994 - 1999, 2000, 03, 04 Ralf Baechle
+ * Copyright (C) 2000, 2002  Maciej W. Rozycki
+ * Copyright (C) 1990, 1999, 2000 Silicon Graphics, Inc.
+ */
+#ifndef _ASM_NETLOGIC_SPACES_H
+#define _ASM_NETLOGIC_SPACES_H
+
+#if defined(CONFIG_MAPPED_KERNEL)
+#ifdef CONFIG_32BIT
+#define PAGE_OFFSET		CKSEG2
+#define MAP_BASE		CKSEG3
+#else
+#define PAGE_OFFSET		_AC(0xc000000000000000, UL)
+#define MAP_BASE		_AC(0xd000000000000000, UL)
+#endif
+#endif /* CONFIG_MAPPED_KERNEL */
+
+#include <asm/mach-generic/spaces.h>
+
+#endif /* __ASM_NETLOGIC_SPACES_H */

index c281f03..dde3a8d 100644
@@ -47,6 +47,17 @@
 #define BOOT_NMI_LOCK		4
 #define BOOT_NMI_HANDLER	8
 
+/* TLB entry save area for mapped kernel */
+#define BOOT_NTLBS		64
+#define BOOT_TLBS_START		72
+#define BOOT_TLB_SIZE		32
+
+/* four u64 entries per TLB */
+#define BOOT_TLB_ENTRYHI	0
+#define BOOT_TLB_ENTRYLO0	1
+#define BOOT_TLB_ENTRYLO1	2
+#define BOOT_TLB_PAGEMASK	3
+
 /* CPU ready flags for each CPU */
 #define BOOT_CPU_READY		2048
 

@@ -90,6 +101,9 @@ extern char nlm_reset_entry[], nlm_reset_entry_end[];
 /* SWIOTLB */
 extern struct dma_map_ops nlm_swiotlb_dma_ops;
 
+/* mapped kernel */
+void nlm_setup_wired_tlbs(void);
+
 extern unsigned int nlm_threads_per_core;
 extern cpumask_t nlm_cpumask;
 

index e1c49a9..21a204f 100644
@@ -135,7 +135,11 @@
 #if defined(CONFIG_MODULES) && defined(KBUILD_64BIT_SYM32) && \
 	VMALLOC_START != CKSSEG
 /* Load modules into 32bit-compatible segment. */
+#ifdef CONFIG_MAPPED_KERNEL
+#define MODULE_START	CKSEG3	/* don't overlap with mapped kernel */
+#else
 #define MODULE_START	CKSSEG
+#endif
 #define MODULE_END	(FIXADDR_START-2*PAGE_SIZE)
 #endif
 

index eeaf50f..81c4859 100644
@@ -431,7 +431,9 @@ void tlb_init(void)
 	 *     be set to fixed-size pages.
 	 */
 	write_c0_pagemask(PM_DEFAULT_MASK);
+#ifndef CONFIG_MAPPED_KERNEL
 	write_c0_wired(0);
+#endif
 	if (current_cpu_type() == CPU_R10000 ||
 	    current_cpu_type() == CPU_R12000 ||
 	    current_cpu_type() == CPU_R14000)

127 lines not on this page · raw message

index d61cba1..4ed5204 100644
@@ -53,6 +53,7 @@ int xlr_wakeup_secondary_cpus(void)
 	struct nlm_soc_info *nodep;
 	unsigned int i, j, boot_cpu;
 	volatile u32 *cpu_ready = nlm_get_boot_data(BOOT_CPU_READY);
+	void *handler;
 
 	/*
 	 *  In case of RMI boot, hit with NMI to get the cores

@@ -60,7 +61,11 @@ int xlr_wakeup_secondary_cpus(void)
 	 */
 	nodep = nlm_get_node(0);
 	boot_cpu = hard_smp_processor_id();
-	nlm_set_nmi_handler(nlm_rmiboot_preboot);
+	handler = nlm_rmiboot_preboot;
+#ifdef CONFIG_MAPPED_KERNEL
+	handler = (void *)CKSEG0ADDR((long)handler);
+#endif
+	nlm_set_nmi_handler(handler);
 	for (i = 0; i < NR_CPUS; i++) {
 		if (i == boot_cpu || !cpumask_test_cpu(i, &nlm_cpumask))
 			continue;

index 37fe58c..9791e39 100644
@@ -51,6 +51,15 @@ int main(int argc, char *argv[])
 
 	vmlinuz_load_addr += (16 - vmlinux_size % 16);
 
+	/* handle 32/64 bit mapped addresses */
+	if (vmlinuz_load_addr >= 0xffffffffc0000000ULL)
+		vmlinuz_load_addr = 0xffffffff80000000ull |
+					(vmlinuz_load_addr & 0x1fffffff);
+	if (vmlinuz_load_addr >= 0xc0000000ULL &&
+				vmlinuz_load_addr < 0xffffffffULL)
+		vmlinuz_load_addr = 0x80000000ull |
+					(vmlinuz_load_addr & 0x1fffffff);
+
 	printf("0x%llx\n", vmlinuz_load_addr);
 
 	return EXIT_SUCCESS;

index c00c4dd..8321c86 100644
@@ -70,6 +70,7 @@ void error(char *x)
 void decompress_kernel(unsigned long boot_heap_start)
 {
 	unsigned long zimage_start, zimage_size;
+	unsigned long long loadaddr = VMLINUX_LOAD_ADDRESS_ULL;
 
 	zimage_start = (unsigned long)(&__image_begin);
 	zimage_size = (unsigned long)(&__image_end) -

@@ -90,9 +91,12 @@ void decompress_kernel(unsigned long boot_heap_start)
 	puthex(VMLINUX_LOAD_ADDRESS_ULL);
 	puts("\n");
 
+#ifdef CONFIG_MAPPED_KERNEL
+	loadaddr = CKSEG0ADDR(loadaddr);
+#endif
 	/* Decompress the kernel with according algorithm */
 	decompress((char *)zimage_start, zimage_size, 0, 0,
-		   (void *)VMLINUX_LOAD_ADDRESS_ULL, 0, error);
+		   (void *)(unsigned long)loadaddr, 0, error);
 
 	/* FIXME: should we flush cache here? */
 	puts("Now, booting the kernel...\n");

index 6d967ce..83aeb7c 100644
@@ -34,6 +34,7 @@
 
 #include <linux/kernel.h>
 #include <linux/types.h>
+#include <linux/sizes.h>
 
 #include <asm/bootinfo.h>
 #include <asm/pgtable.h>

@@ -41,100 +42,187 @@
 #include <asm/tlb.h>
 
 #include <asm/netlogic/common.h>
+#include <asm/netlogic/xlp-hal/xlp.h>
 
-#define TLBSZ		(256 * 1024 * 1024)
-#define PM_TLBSZ	PM_256M
-#define PTE_MAPKERN(pa)	(((pa >> 12) << 6) | 0x2f)
+#define SZ_4G		(4ull * 1024 * 1024 * 1024)
+#define PM_4G		0x1ffffe000
+#define MINPGSZ		SZ_256M
+
+#define PTE_MAPKERN(pa)	((((pa) >> 12) << 6) | 0x2f)
 #define TLB_MAXWIRED	28
 
 static const int prefetch_backup = 512;
 
 #if defined(CONFIG_MAPPED_KERNEL) && defined(CONFIG_64BIT)
-static void nlm_tlb_align(struct boot_mem_map_entry *map)
+
+/* To track the allocated area of a boot_mem_map segment */
+struct alloc_entry {
+	/* Start and end of the va mapped */
+	unsigned long long start;
+	unsigned long long end;
+
+	/* When just one of lo0/lo1 is used, the valid area is half of above */
+	unsigned long long astart;
+	unsigned long long aend;
+} alloc_map[32];
+
+static inline int addtlb(phys_addr_t pa, u64 pgsz, u64 pmask,
+	unsigned int validmask, struct alloc_entry *ae)
 {
-	phys_t astart, aend, start, end;
+	phys_addr_t endpa;
+	u64 *t;
+	u32 *tlbcount;
+	int ntlb;
+
+	tlbcount = (u32 *)nlm_get_boot_data(BOOT_NTLBS);
+	ntlb = *tlbcount;
+	endpa = pa + 2 * pgsz;
 
-	start = map->addr;
-	end = start + map->size;
+	pr_debug("%2d - pa0 %llx pa1 %llx pgsz %llx valid %x\n",
+				ntlb, pa, endpa, pgsz, validmask);
+	if (ntlb == TLB_MAXWIRED) {
+		pr_err("Ran out of TLB entries pa %llx pgsz %llx\n", pa, pgsz);
+		return -1;
+	}
 
-	/* fudge first entry for now  */
-	if (start < 0x10000000) {
-		start = 0;
-		end = 0x10000000;
+	t = nlm_get_boot_data(BOOT_TLBS_START);
+	t += ntlb * (BOOT_TLB_SIZE / sizeof(t[0]));
+	t[BOOT_TLB_ENTRYHI] = pa + PAGE_OFFSET;
+	t[BOOT_TLB_ENTRYLO0] = (validmask & 0x1) ? PTE_MAPKERN(pa) : 1;
+	t[BOOT_TLB_ENTRYLO1] = (validmask & 0x2) ? PTE_MAPKERN(pa + pgsz) : 1;
+	t[BOOT_TLB_PAGEMASK] = pmask;
+
+	if (pa < ae->start) {
+		ae->astart = ae->start = pa;
+		if ((validmask & 0x1) == 0)
+			ae->astart += pgsz;
 	}
-	astart = round_up(start, TLBSZ);
-	aend = round_down(end, TLBSZ);
-	if (aend <= astart) {
-		pr_info("Boot mem map: discard seg %lx-%lx\n",
-				(unsigned long)start, (unsigned long)end);
-		map->size = 0;
-		return;
+	if (endpa > ae->end) {
+		ae->aend = ae->end = endpa;
+		if ((validmask & 0x2) == 0)
+			ae->aend -= pgsz;
 	}
-	if (astart != start || aend != end) {
-		if (start != 0) {
-			map->addr = astart;
-			map->size = aend - astart;
-		}
-		pr_info("Boot mem map: %lx - %lx -> %lx-%lx\n",
-			(unsigned long)start, (unsigned long)end,
-			(unsigned long)astart, (unsigned long)aend);
-	} else
-		pr_info("Boot mem map: added %lx - %lx\n",
-			(unsigned long)astart, (unsigned long)aend);
+	*tlbcount = ntlb + 1;
+	return 0;
 }
 
+/*
+ * Calculate the TLB entries needed to wire dowm the memory map
+ *
+ * Tries to use the largest pagesizes possible, discards memory which
+ * cannot be mapped
+ */
 static void nlm_calc_wired_tlbs(void)
 {
-	u64 *tlbarr;
-	u32 *tlbcount;
-	u64 lo0, lo1, vaddr;
-	phys_addr_t astart, aend, p;
-	unsigned long bootdata = CKSEG1ADDR(RESET_DATA_PHYS);
-	int i, pos;
+	u64 pgsz, pgmask, p;
+	phys_addr_t astart, aend, pend, nstart;
+	phys_addr_t tstart, tend, mstart, mend;
+	struct boot_mem_map_entry *bmap;
+	int i, nr_map;
+
+	nr_map = boot_mem_map.nr_map;
+	bmap = boot_mem_map.map;
+
+	for (i = 0; i < nr_map; i++) {
+		alloc_map[i].start = alloc_map[i].astart = ~0ull;
+		alloc_map[i].end = alloc_map[i].aend = 0;
+	}
 
-	tlbarr = (u64 *)(bootdata + BOOT_TLBS_START);
-	tlbcount = (u32 *)(bootdata + BOOT_NTLBS);
+	/* force the first entry with one 256M lo0 page */
+	addtlb(0, 0x10000000, PM_256M, 0x1, &alloc_map[0]);
 
-	pos = 0;
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		astart = boot_mem_map.map[i].addr;
-		aend =	astart + boot_mem_map.map[i].size;
+	/* starting page size and page mask */
+	if (cpu_is_xlpii()) {
+		pgsz = SZ_4G;
+		pgmask = PM_4G;
+	} else {
+		pgsz = SZ_256M;
+		pgmask = PM_256M;
+	}
 
-		/* fudge first entry for now  */
-		if (astart < 0x10000000) {
-			astart = 0;
-			aend = 0x10000000;
-		}
-		for (p = round_down(astart, 2 * TLBSZ);
-			p < round_up(aend, 2 * TLBSZ);) {
-				vaddr = PAGE_OFFSET + p;
-				lo0 = (p >= astart) ? PTE_MAPKERN(p) : 1;
-				p += TLBSZ;
-				lo1 = (p < aend) ? PTE_MAPKERN(p) : 1;
-				p += TLBSZ;
-
-				tlbarr[BOOT_TLB_ENTRYHI] = vaddr;
-				tlbarr[BOOT_TLB_ENTRYLO0] = lo0;
-				tlbarr[BOOT_TLB_ENTRYLO1] = lo1;
-				tlbarr[BOOT_TLB_PAGEMASK] = PM_TLBSZ;
-				tlbarr += (BOOT_TLB_SIZE / sizeof(tlbarr[0]));
-
-				if (++pos >= TLB_MAXWIRED) {
-					pr_err("Ran out of TLBs at %llx, ",
-							(unsigned long long)p);
-					pr_err("Discarding rest of memory!\n");
-					boot_mem_map.nr_map = i + 1;
-					boot_mem_map.map[i].size = p -
-						boot_mem_map.map[i].addr;
+	/* do multiple passes with successively smaller page sizes */
+	for (; pgsz >= MINPGSZ; pgsz /= 4, pgmask = (pgmask >> 2) ^ 0x1800) {
+		for (i = 0; i < nr_map; i++) {
+			if (bmap[i].type != BOOT_MEM_RAM)
+				continue;
+
+			/* previous mapping end and next mapping start */
+			pend = alloc_map[i - 1].end;
+			nstart = (i == nr_map - 1) ? ~0ull : bmap[i + 1].addr;
+
+			/* mem block start and end */
+			mstart = round_up(bmap[i].addr, MINPGSZ);
+			mend = round_down(bmap[i].addr + bmap[i].size, MINPGSZ);
+
+			/* allocated area in the memory block, start and end */
+			astart = alloc_map[i].start;
+			aend = alloc_map[i].end;
+
+			/* skip fully mapped blocks */
+			if (mstart >= astart && mend <= aend)
+				continue;
+
+			/* boundaries aligned to the current page size */
+			tstart = round_up(mstart, 2 * pgsz);
+			tend = round_down(mend, 2 * pgsz);
+			if (tstart > tend)
+				continue;
+
+			/* use LO1 of a TLB entry */
+			if (mstart + pgsz == tstart && pend <= mstart - pgsz)
+				if (addtlb(mstart - pgsz, pgsz,
+						pgmask, 0x2, &alloc_map[i]))
 					goto out;
+
+			for (p = tstart; p < tend;) {
+				if (astart < aend && p == astart) {
+					p = aend;
+					continue;
 				}
+				if (addtlb(p, pgsz, pgmask, 0x3, &alloc_map[i]))
+					goto out;
+				p += 2 * pgsz;
+			}
+
+			/* use LO0 of a TLB entry */
+			if (tend + pgsz == mend && nstart >= mend + pgsz)
+				if (addtlb(tend, pgsz,
+						pgmask, 0x1, &alloc_map[i]))
+					goto out;
 		}
 	}
 out:
-	*tlbcount = pos;
-	pr_info("%d TLB entires used for mapped kernel.\n", pos);
+	for (i = 0; i < nr_map; i++) {
+		mstart = bmap[i].addr;
+		mend = bmap[i].addr + bmap[i].size;
+		astart = alloc_map[i].astart;
+		aend = alloc_map[i].aend;
+
+		pr_info("%2d alloc: %10llx %10llx mem %10llx %10llx\n", i,
+						astart, aend, bmap[i].addr,
+						bmap[i].addr + bmap[i].size);
+		if (astart >= aend) {
+			bmap[i].size = 0;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)mstart,
+				(unsigned long long)mend);
+			continue;
+		}
+		if (bmap[i].addr < astart) {
+			bmap[i].addr = astart;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)bmap[i].addr,
+				(unsigned long long)astart);
+		}
+		if (mend > aend) {
+			bmap[i].size = aend - bmap[i].addr;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)aend,
+				(unsigned long long)mend);
+		}
+	}
+	pr_info("%d TLB entires used for mapped kernel.\n",
+				*(u32 *)nlm_get_boot_data(BOOT_NTLBS));
 }
 #endif
 

@@ -143,13 +231,6 @@ void __init plat_mem_fixup(void)
 	int i;
 
 #if defined(CONFIG_MAPPED_KERNEL) && defined(CONFIG_64BIT)
-	/* trim memory regions to PM_TLBSZ boundaries */
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		nlm_tlb_align(&boot_mem_map.map[i]);
-	}
-
 	/* calculate and save wired TLB entries */
 	nlm_calc_wired_tlbs();
 

index a842154..74c5f6d 100644
@@ -610,6 +610,7 @@ static void __init request_crashkernel(struct resource *res)
 static void __init arch_mem_init(char **cmdline_p)
 {
 	extern void plat_mem_setup(void);
+	extern void plat_mem_fixup(void);
 
 	/* call board setup routine */
 	plat_mem_setup();

@@ -653,6 +654,9 @@ static void __init arch_mem_init(char **cmdline_p)
 		pr_info("User-defined physical RAM map:\n");
 		print_memory_map();
 	}
+#if defined(CONFIG_CPU_XLP) || defined(CONFIG_CPU_XLR)
+	plat_mem_fixup();
+#endif
 
 	bootmem_init();
 #ifdef CONFIG_PROC_VMCORE

53 lines not on this page · raw message

index 135a38f..bf4f6d3 100644
@@ -65,18 +65,6 @@ static void nlm_linux_exit(void)
 		cpu_wait();
 }
 
-static void nlm_fixup_mem(void)
-{
-	const int pref_backup = 512;
-	int i;
-
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		boot_mem_map.map[i].size -= pref_backup;
-	}
-}
-
 static void __init xlp_init_mem_from_bars(void)
 {
 	uint64_t map[16];

@@ -115,8 +103,6 @@ void __init plat_mem_setup(void)
 		pr_info("Using DRAM BARs for memory map.\n");
 		xlp_init_mem_from_bars();
 	}
-	/* Calculate and setup wired entries for mapped kernel */
-	nlm_fixup_mem();
 }
 
 const char *get_system_type(void)

index d118b9a..714f6a3 100644
@@ -144,7 +144,6 @@ static void prom_add_memory(void)
 {
 	struct nlm_boot_mem_map *bootm;
 	u64 start, size;
-	u64 pref_backup = 512;	/* avoid pref walking beyond end */
 	int i;
 
 	bootm = (void *)(long)nlm_prom_info.psb_mem_map;

@@ -158,7 +157,7 @@ static void prom_add_memory(void)
 		if (i == 0 && start == 0 && size == 0x0c000000)
 			size = 0x0ff00000;
 
-		add_memory_region(start, size - pref_backup, BOOT_MEM_RAM);
+		add_memory_region(start, size, BOOT_MEM_RAM);
 	}
 }
 

50 lines not on this page · raw message

new file mode 100644
index 0000000..6f32f07
@@ -0,0 +1,25 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 1994 - 1999, 2000, 03, 04 Ralf Baechle
+ * Copyright (C) 2000, 2002  Maciej W. Rozycki
+ * Copyright (C) 1990, 1999, 2000 Silicon Graphics, Inc.
+ */
+#ifndef _ASM_NETLOGIC_SPACES_H
+#define _ASM_NETLOGIC_SPACES_H
+
+#if defined(CONFIG_MAPPED_KERNEL)
+#ifdef CONFIG_32BIT
+#define PAGE_OFFSET		CKSEG2
+#define MAP_BASE		CKSEG3
+#else
+#define PAGE_OFFSET		_AC(0xc000000000000000, UL)
+#define MAP_BASE		_AC(0xd000000000000000, UL)
+#endif
+#endif /* CONFIG_MAPPED_KERNEL */
+
+#include <asm/mach-generic/spaces.h>
+
+#endif /* __ASM_NETLOGIC_SPACES_H */

index c281f03..dde3a8d 100644
@@ -47,6 +47,17 @@
 #define BOOT_NMI_LOCK		4
 #define BOOT_NMI_HANDLER	8
 
+/* TLB entry save area for mapped kernel */
+#define BOOT_NTLBS		64
+#define BOOT_TLBS_START		72
+#define BOOT_TLB_SIZE		32
+
+/* four u64 entries per TLB */
+#define BOOT_TLB_ENTRYHI	0
+#define BOOT_TLB_ENTRYLO0	1
+#define BOOT_TLB_ENTRYLO1	2
+#define BOOT_TLB_PAGEMASK	3
+
 /* CPU ready flags for each CPU */
 #define BOOT_CPU_READY		2048
 

@@ -90,6 +101,9 @@ extern char nlm_reset_entry[], nlm_reset_entry_end[];
 /* SWIOTLB */
 extern struct dma_map_ops nlm_swiotlb_dma_ops;
 
+/* mapped kernel */
+void nlm_setup_wired_tlbs(void);
+
 extern unsigned int nlm_threads_per_core;
 extern cpumask_t nlm_cpumask;
 

index e1c49a9..21a204f 100644
@@ -135,7 +135,11 @@
 #if defined(CONFIG_MODULES) && defined(KBUILD_64BIT_SYM32) && \
 	VMALLOC_START != CKSSEG
 /* Load modules into 32bit-compatible segment. */
+#ifdef CONFIG_MAPPED_KERNEL
+#define MODULE_START	CKSEG3	/* don't overlap with mapped kernel */
+#else
 #define MODULE_START	CKSSEG
+#endif
 #define MODULE_END	(FIXADDR_START-2*PAGE_SIZE)
 #endif
 

index eeaf50f..81c4859 100644
@@ -431,7 +431,9 @@ void tlb_init(void)
 	 *     be set to fixed-size pages.
 	 */
 	write_c0_pagemask(PM_DEFAULT_MASK);
+#ifndef CONFIG_MAPPED_KERNEL
 	write_c0_wired(0);
+#endif
 	if (current_cpu_type() == CPU_R10000 ||
 	    current_cpu_type() == CPU_R12000 ||
 	    current_cpu_type() == CPU_R14000)

127 lines not on this page · raw message

index d61cba1..4ed5204 100644
@@ -53,6 +53,7 @@ int xlr_wakeup_secondary_cpus(void)
 	struct nlm_soc_info *nodep;
 	unsigned int i, j, boot_cpu;
 	volatile u32 *cpu_ready = nlm_get_boot_data(BOOT_CPU_READY);
+	void *handler;
 
 	/*
 	 *  In case of RMI boot, hit with NMI to get the cores

@@ -60,7 +61,11 @@ int xlr_wakeup_secondary_cpus(void)
 	 */
 	nodep = nlm_get_node(0);
 	boot_cpu = hard_smp_processor_id();
-	nlm_set_nmi_handler(nlm_rmiboot_preboot);
+	handler = nlm_rmiboot_preboot;
+#ifdef CONFIG_MAPPED_KERNEL
+	handler = (void *)CKSEG0ADDR((long)handler);
+#endif
+	nlm_set_nmi_handler(handler);
 	for (i = 0; i < NR_CPUS; i++) {
 		if (i == boot_cpu || !cpumask_test_cpu(i, &nlm_cpumask))
 			continue;

index 37fe58c..9791e39 100644
@@ -51,6 +51,15 @@ int main(int argc, char *argv[])
 
 	vmlinuz_load_addr += (16 - vmlinux_size % 16);
 
+	/* handle 32/64 bit mapped addresses */
+	if (vmlinuz_load_addr >= 0xffffffffc0000000ULL)
+		vmlinuz_load_addr = 0xffffffff80000000ull |
+					(vmlinuz_load_addr & 0x1fffffff);
+	if (vmlinuz_load_addr >= 0xc0000000ULL &&
+				vmlinuz_load_addr < 0xffffffffULL)
+		vmlinuz_load_addr = 0x80000000ull |
+					(vmlinuz_load_addr & 0x1fffffff);
+
 	printf("0x%llx\n", vmlinuz_load_addr);
 
 	return EXIT_SUCCESS;

index c00c4dd..8321c86 100644
@@ -70,6 +70,7 @@ void error(char *x)
 void decompress_kernel(unsigned long boot_heap_start)
 {
 	unsigned long zimage_start, zimage_size;
+	unsigned long long loadaddr = VMLINUX_LOAD_ADDRESS_ULL;
 
 	zimage_start = (unsigned long)(&__image_begin);
 	zimage_size = (unsigned long)(&__image_end) -

@@ -90,9 +91,12 @@ void decompress_kernel(unsigned long boot_heap_start)
 	puthex(VMLINUX_LOAD_ADDRESS_ULL);
 	puts("\n");
 
+#ifdef CONFIG_MAPPED_KERNEL
+	loadaddr = CKSEG0ADDR(loadaddr);
+#endif
 	/* Decompress the kernel with according algorithm */
 	decompress((char *)zimage_start, zimage_size, 0, 0,
-		   (void *)VMLINUX_LOAD_ADDRESS_ULL, 0, error);
+		   (void *)(unsigned long)loadaddr, 0, error);
 
 	/* FIXME: should we flush cache here? */
 	puts("Now, booting the kernel...\n");

index 6d967ce..83aeb7c 100644
@@ -34,6 +34,7 @@
 
 #include <linux/kernel.h>
 #include <linux/types.h>
+#include <linux/sizes.h>
 
 #include <asm/bootinfo.h>
 #include <asm/pgtable.h>

@@ -41,100 +42,187 @@
 #include <asm/tlb.h>
 
 #include <asm/netlogic/common.h>
+#include <asm/netlogic/xlp-hal/xlp.h>
 
-#define TLBSZ		(256 * 1024 * 1024)
-#define PM_TLBSZ	PM_256M
-#define PTE_MAPKERN(pa)	(((pa >> 12) << 6) | 0x2f)
+#define SZ_4G		(4ull * 1024 * 1024 * 1024)
+#define PM_4G		0x1ffffe000
+#define MINPGSZ		SZ_256M
+
+#define PTE_MAPKERN(pa)	((((pa) >> 12) << 6) | 0x2f)
 #define TLB_MAXWIRED	28
 
 static const int prefetch_backup = 512;
 
 #if defined(CONFIG_MAPPED_KERNEL) && defined(CONFIG_64BIT)
-static void nlm_tlb_align(struct boot_mem_map_entry *map)
+
+/* To track the allocated area of a boot_mem_map segment */
+struct alloc_entry {
+	/* Start and end of the va mapped */
+	unsigned long long start;
+	unsigned long long end;
+
+	/* When just one of lo0/lo1 is used, the valid area is half of above */
+	unsigned long long astart;
+	unsigned long long aend;
+} alloc_map[32];
+
+static inline int addtlb(phys_addr_t pa, u64 pgsz, u64 pmask,
+	unsigned int validmask, struct alloc_entry *ae)
 {
-	phys_t astart, aend, start, end;
+	phys_addr_t endpa;
+	u64 *t;
+	u32 *tlbcount;
+	int ntlb;
+
+	tlbcount = (u32 *)nlm_get_boot_data(BOOT_NTLBS);
+	ntlb = *tlbcount;
+	endpa = pa + 2 * pgsz;
 
-	start = map->addr;
-	end = start + map->size;
+	pr_debug("%2d - pa0 %llx pa1 %llx pgsz %llx valid %x\n",
+				ntlb, pa, endpa, pgsz, validmask);
+	if (ntlb == TLB_MAXWIRED) {
+		pr_err("Ran out of TLB entries pa %llx pgsz %llx\n", pa, pgsz);
+		return -1;
+	}
 
-	/* fudge first entry for now  */
-	if (start < 0x10000000) {
-		start = 0;
-		end = 0x10000000;
+	t = nlm_get_boot_data(BOOT_TLBS_START);
+	t += ntlb * (BOOT_TLB_SIZE / sizeof(t[0]));
+	t[BOOT_TLB_ENTRYHI] = pa + PAGE_OFFSET;
+	t[BOOT_TLB_ENTRYLO0] = (validmask & 0x1) ? PTE_MAPKERN(pa) : 1;
+	t[BOOT_TLB_ENTRYLO1] = (validmask & 0x2) ? PTE_MAPKERN(pa + pgsz) : 1;
+	t[BOOT_TLB_PAGEMASK] = pmask;
+
+	if (pa < ae->start) {
+		ae->astart = ae->start = pa;
+		if ((validmask & 0x1) == 0)
+			ae->astart += pgsz;
 	}
-	astart = round_up(start, TLBSZ);
-	aend = round_down(end, TLBSZ);
-	if (aend <= astart) {
-		pr_info("Boot mem map: discard seg %lx-%lx\n",
-				(unsigned long)start, (unsigned long)end);
-		map->size = 0;
-		return;
+	if (endpa > ae->end) {
+		ae->aend = ae->end = endpa;
+		if ((validmask & 0x2) == 0)
+			ae->aend -= pgsz;
 	}
-	if (astart != start || aend != end) {
-		if (start != 0) {
-			map->addr = astart;
-			map->size = aend - astart;
-		}
-		pr_info("Boot mem map: %lx - %lx -> %lx-%lx\n",
-			(unsigned long)start, (unsigned long)end,
-			(unsigned long)astart, (unsigned long)aend);
-	} else
-		pr_info("Boot mem map: added %lx - %lx\n",
-			(unsigned long)astart, (unsigned long)aend);
+	*tlbcount = ntlb + 1;
+	return 0;
 }
 
+/*
+ * Calculate the TLB entries needed to wire dowm the memory map
+ *
+ * Tries to use the largest pagesizes possible, discards memory which
+ * cannot be mapped
+ */
 static void nlm_calc_wired_tlbs(void)
 {
-	u64 *tlbarr;
-	u32 *tlbcount;
-	u64 lo0, lo1, vaddr;
-	phys_addr_t astart, aend, p;
-	unsigned long bootdata = CKSEG1ADDR(RESET_DATA_PHYS);
-	int i, pos;
+	u64 pgsz, pgmask, p;
+	phys_addr_t astart, aend, pend, nstart;
+	phys_addr_t tstart, tend, mstart, mend;
+	struct boot_mem_map_entry *bmap;
+	int i, nr_map;
+
+	nr_map = boot_mem_map.nr_map;
+	bmap = boot_mem_map.map;
+
+	for (i = 0; i < nr_map; i++) {
+		alloc_map[i].start = alloc_map[i].astart = ~0ull;
+		alloc_map[i].end = alloc_map[i].aend = 0;
+	}
 
-	tlbarr = (u64 *)(bootdata + BOOT_TLBS_START);
-	tlbcount = (u32 *)(bootdata + BOOT_NTLBS);
+	/* force the first entry with one 256M lo0 page */
+	addtlb(0, 0x10000000, PM_256M, 0x1, &alloc_map[0]);
 
-	pos = 0;
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		astart = boot_mem_map.map[i].addr;
-		aend =	astart + boot_mem_map.map[i].size;
+	/* starting page size and page mask */
+	if (cpu_is_xlpii()) {
+		pgsz = SZ_4G;
+		pgmask = PM_4G;
+	} else {
+		pgsz = SZ_256M;
+		pgmask = PM_256M;
+	}
 
-		/* fudge first entry for now  */
-		if (astart < 0x10000000) {
-			astart = 0;
-			aend = 0x10000000;
-		}
-		for (p = round_down(astart, 2 * TLBSZ);
-			p < round_up(aend, 2 * TLBSZ);) {
-				vaddr = PAGE_OFFSET + p;
-				lo0 = (p >= astart) ? PTE_MAPKERN(p) : 1;
-				p += TLBSZ;
-				lo1 = (p < aend) ? PTE_MAPKERN(p) : 1;
-				p += TLBSZ;
-
-				tlbarr[BOOT_TLB_ENTRYHI] = vaddr;
-				tlbarr[BOOT_TLB_ENTRYLO0] = lo0;
-				tlbarr[BOOT_TLB_ENTRYLO1] = lo1;
-				tlbarr[BOOT_TLB_PAGEMASK] = PM_TLBSZ;
-				tlbarr += (BOOT_TLB_SIZE / sizeof(tlbarr[0]));
-
-				if (++pos >= TLB_MAXWIRED) {
-					pr_err("Ran out of TLBs at %llx, ",
-							(unsigned long long)p);
-					pr_err("Discarding rest of memory!\n");
-					boot_mem_map.nr_map = i + 1;
-					boot_mem_map.map[i].size = p -
-						boot_mem_map.map[i].addr;
+	/* do multiple passes with successively smaller page sizes */
+	for (; pgsz >= MINPGSZ; pgsz /= 4, pgmask = (pgmask >> 2) ^ 0x1800) {
+		for (i = 0; i < nr_map; i++) {
+			if (bmap[i].type != BOOT_MEM_RAM)
+				continue;
+
+			/* previous mapping end and next mapping start */
+			pend = alloc_map[i - 1].end;
+			nstart = (i == nr_map - 1) ? ~0ull : bmap[i + 1].addr;
+
+			/* mem block start and end */
+			mstart = round_up(bmap[i].addr, MINPGSZ);
+			mend = round_down(bmap[i].addr + bmap[i].size, MINPGSZ);
+
+			/* allocated area in the memory block, start and end */
+			astart = alloc_map[i].start;
+			aend = alloc_map[i].end;
+
+			/* skip fully mapped blocks */
+			if (mstart >= astart && mend <= aend)
+				continue;
+
+			/* boundaries aligned to the current page size */
+			tstart = round_up(mstart, 2 * pgsz);
+			tend = round_down(mend, 2 * pgsz);
+			if (tstart > tend)
+				continue;
+
+			/* use LO1 of a TLB entry */
+			if (mstart + pgsz == tstart && pend <= mstart - pgsz)
+				if (addtlb(mstart - pgsz, pgsz,
+						pgmask, 0x2, &alloc_map[i]))
 					goto out;
+
+			for (p = tstart; p < tend;) {
+				if (astart < aend && p == astart) {
+					p = aend;
+					continue;
 				}
+				if (addtlb(p, pgsz, pgmask, 0x3, &alloc_map[i]))
+					goto out;
+				p += 2 * pgsz;
+			}
+
+			/* use LO0 of a TLB entry */
+			if (tend + pgsz == mend && nstart >= mend + pgsz)
+				if (addtlb(tend, pgsz,
+						pgmask, 0x1, &alloc_map[i]))
+					goto out;
 		}
 	}
 out:
-	*tlbcount = pos;
-	pr_info("%d TLB entires used for mapped kernel.\n", pos);
+	for (i = 0; i < nr_map; i++) {
+		mstart = bmap[i].addr;
+		mend = bmap[i].addr + bmap[i].size;
+		astart = alloc_map[i].astart;
+		aend = alloc_map[i].aend;
+
+		pr_info("%2d alloc: %10llx %10llx mem %10llx %10llx\n", i,
+						astart, aend, bmap[i].addr,
+						bmap[i].addr + bmap[i].size);
+		if (astart >= aend) {
+			bmap[i].size = 0;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)mstart,
+				(unsigned long long)mend);
+			continue;
+		}
+		if (bmap[i].addr < astart) {
+			bmap[i].addr = astart;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)bmap[i].addr,
+				(unsigned long long)astart);
+		}
+		if (mend > aend) {
+			bmap[i].size = aend - bmap[i].addr;
+			pr_info("%2d: Discarded %#10llx - %#10llx\n", i,
+				(unsigned long long)aend,
+				(unsigned long long)mend);
+		}
+	}
+	pr_info("%d TLB entires used for mapped kernel.\n",
+				*(u32 *)nlm_get_boot_data(BOOT_NTLBS));
 }
 #endif
 

@@ -143,13 +231,6 @@ void __init plat_mem_fixup(void)
 	int i;
 
 #if defined(CONFIG_MAPPED_KERNEL) && defined(CONFIG_64BIT)
-	/* trim memory regions to PM_TLBSZ boundaries */
-	for (i = 0; i < boot_mem_map.nr_map; i++) {
-		if (boot_mem_map.map[i].type != BOOT_MEM_RAM)
-			continue;
-		nlm_tlb_align(&boot_mem_map.map[i]);
-	}
-
 	/* calculate and save wired TLB entries */
 	nlm_calc_wired_tlbs();