Blackfin: punt dead cache locking code
[safe/jmp/linux-2.6] / arch / blackfin / kernel / setup.c
index 102a9db..ce02f98 100644 (file)
 #include <linux/bootmem.h>
 #include <linux/seq_file.h>
 #include <linux/cpu.h>
+#include <linux/mm.h>
 #include <linux/module.h>
 #include <linux/tty.h>
 #include <linux/pfn.h>
 
+#ifdef CONFIG_MTD_UCLINUX
+#include <linux/mtd/map.h>
 #include <linux/ext2_fs.h>
 #include <linux/cramfs_fs.h>
 #include <linux/romfs_fs.h>
+#endif
 
 #include <asm/cplb.h>
 #include <asm/cacheflush.h>
 #include <asm/blackfin.h>
 #include <asm/cplbinit.h>
 #include <asm/div64.h>
+#include <asm/cpu.h>
 #include <asm/fixed_code.h>
 #include <asm/early_printk.h>
 
-static DEFINE_PER_CPU(struct cpu, cpu_devices);
-
 u16 _bfin_swrst;
 EXPORT_SYMBOL(_bfin_swrst);
 
@@ -42,8 +45,10 @@ EXPORT_SYMBOL(memory_start);
 EXPORT_SYMBOL(memory_end);
 EXPORT_SYMBOL(physical_mem_end);
 EXPORT_SYMBOL(_ramend);
+EXPORT_SYMBOL(reserved_mem_dcache_on);
 
 #ifdef CONFIG_MTD_UCLINUX
+extern struct map_info uclinux_ram_map;
 unsigned long memory_mtd_end, memory_mtd_start, mtd_size;
 unsigned long _ebss;
 EXPORT_SYMBOL(memory_mtd_end);
@@ -52,12 +57,14 @@ EXPORT_SYMBOL(mtd_size);
 #endif
 
 char __initdata command_line[COMMAND_LINE_SIZE];
+void __initdata *init_retx, *init_saved_retx, *init_saved_seqstat,
+       *init_saved_icplb_fault_addr, *init_saved_dcplb_fault_addr;
 
 /* boot memmap, for parsing "memmap=" */
 #define BFIN_MEMMAP_MAX                128 /* number of entries in bfin_memmap */
 #define BFIN_MEMMAP_RAM                1
 #define BFIN_MEMMAP_RESERVED   2
-struct bfin_memmap {
+static struct bfin_memmap {
        int nr_map;
        struct bfin_memmap_entry {
                unsigned long long addr; /* start of memory segment */
@@ -76,69 +83,152 @@ static struct change_member *change_point[2*BFIN_MEMMAP_MAX] __initdata;
 static struct bfin_memmap_entry *overlap_list[BFIN_MEMMAP_MAX] __initdata;
 static struct bfin_memmap_entry new_map[BFIN_MEMMAP_MAX] __initdata;
 
-void __init bf53x_cache_init(void)
-{
+DEFINE_PER_CPU(struct blackfin_cpudata, cpu_data);
+
+static int early_init_clkin_hz(char *buf);
+
 #if defined(CONFIG_BFIN_DCACHE) || defined(CONFIG_BFIN_ICACHE)
-       generate_cpl_tables();
+void __init generate_cplb_tables(void)
+{
+       unsigned int cpu;
+
+       generate_cplb_tables_all();
+       /* Generate per-CPU I&D CPLB tables */
+       for (cpu = 0; cpu < num_possible_cpus(); ++cpu)
+               generate_cplb_tables_cpu(cpu);
+}
 #endif
 
+void __cpuinit bfin_setup_caches(unsigned int cpu)
+{
 #ifdef CONFIG_BFIN_ICACHE
-       bfin_icache_init();
-       printk(KERN_INFO "Instruction Cache Enabled\n");
+       bfin_icache_init(icplb_tbl[cpu]);
+#endif
+
+#ifdef CONFIG_BFIN_DCACHE
+       bfin_dcache_init(dcplb_tbl[cpu]);
+#endif
+
+       /*
+        * In cache coherence emulation mode, we need to have the
+        * D-cache enabled before running any atomic operation which
+        * might invove cache invalidation (i.e. spinlock, rwlock).
+        * So printk's are deferred until then.
+        */
+#ifdef CONFIG_BFIN_ICACHE
+       printk(KERN_INFO "Instruction Cache Enabled for CPU%u\n", cpu);
+       printk(KERN_INFO "  External memory:"
+# ifdef CONFIG_BFIN_EXTMEM_ICACHEABLE
+              " cacheable"
+# else
+              " uncacheable"
+# endif
+              " in instruction cache\n");
+       if (L2_LENGTH)
+               printk(KERN_INFO "  L2 SRAM        :"
+# ifdef CONFIG_BFIN_L2_ICACHEABLE
+                      " cacheable"
+# else
+                      " uncacheable"
+# endif
+                      " in instruction cache\n");
+
+#else
+       printk(KERN_INFO "Instruction Cache Disabled for CPU%u\n", cpu);
 #endif
 
 #ifdef CONFIG_BFIN_DCACHE
-       bfin_dcache_init();
-       printk(KERN_INFO "Data Cache Enabled"
-# if defined CONFIG_BFIN_WB
-               " (write-back)"
-# elif defined CONFIG_BFIN_WT
-               " (write-through)"
+       printk(KERN_INFO "Data Cache Enabled for CPU%u\n", cpu);
+       printk(KERN_INFO "  External memory:"
+# if defined CONFIG_BFIN_EXTMEM_WRITEBACK
+              " cacheable (write-back)"
+# elif defined CONFIG_BFIN_EXTMEM_WRITETHROUGH
+              " cacheable (write-through)"
+# else
+              " uncacheable"
+# endif
+              " in data cache\n");
+       if (L2_LENGTH)
+               printk(KERN_INFO "  L2 SRAM        :"
+# if defined CONFIG_BFIN_L2_WRITEBACK
+                      " cacheable (write-back)"
+# elif defined CONFIG_BFIN_L2_WRITETHROUGH
+                      " cacheable (write-through)"
+# else
+                      " uncacheable"
 # endif
-               "\n");
+                      " in data cache\n");
+#else
+       printk(KERN_INFO "Data Cache Disabled for CPU%u\n", cpu);
 #endif
 }
 
-void __init bf53x_relocate_l1_mem(void)
+void __cpuinit bfin_setup_cpudata(unsigned int cpu)
+{
+       struct blackfin_cpudata *cpudata = &per_cpu(cpu_data, cpu);
+
+       cpudata->idle = current;
+       cpudata->imemctl = bfin_read_IMEM_CONTROL();
+       cpudata->dmemctl = bfin_read_DMEM_CONTROL();
+}
+
+void __init bfin_cache_init(void)
+{
+#if defined(CONFIG_BFIN_DCACHE) || defined(CONFIG_BFIN_ICACHE)
+       generate_cplb_tables();
+#endif
+       bfin_setup_caches(0);
+}
+
+void __init bfin_relocate_l1_mem(void)
 {
        unsigned long l1_code_length;
        unsigned long l1_data_a_length;
        unsigned long l1_data_b_length;
        unsigned long l2_length;
 
-       l1_code_length = _etext_l1 - _stext_l1;
-       if (l1_code_length > L1_CODE_LENGTH)
-               panic("L1 Instruction SRAM Overflow\n");
-       /* cannot complain as printk is not available as yet.
-        * But we can continue booting and complain later!
-        */
-
-       /* Copy _stext_l1 to _etext_l1 to L1 instruction SRAM */
-       dma_memcpy(_stext_l1, _l1_lma_start, l1_code_length);
+       early_shadow_stamp();
 
-       l1_data_a_length = _ebss_l1 - _sdata_l1;
-       if (l1_data_a_length > L1_DATA_A_LENGTH)
-               panic("L1 Data SRAM Bank A Overflow\n");
-
-       /* Copy _sdata_l1 to _ebss_l1 to L1 data bank A SRAM */
-       dma_memcpy(_sdata_l1, _l1_lma_start + l1_code_length, l1_data_a_length);
+       /*
+        * due to the ALIGN(4) in the arch/blackfin/kernel/vmlinux.lds.S
+        * we know that everything about l1 text/data is nice and aligned,
+        * so copy by 4 byte chunks, and don't worry about overlapping
+        * src/dest.
+        *
+        * We can't use the dma_memcpy functions, since they can call
+        * scheduler functions which might be in L1 :( and core writes
+        * into L1 instruction cause bad access errors, so we are stuck,
+        * we are required to use DMA, but can't use the common dma
+        * functions. We can't use memcpy either - since that might be
+        * going to be in the relocated L1
+        */
 
-       l1_data_b_length = _ebss_b_l1 - _sdata_b_l1;
-       if (l1_data_b_length > L1_DATA_B_LENGTH)
-               panic("L1 Data SRAM Bank B Overflow\n");
+       blackfin_dma_early_init();
 
-       /* Copy _sdata_b_l1 to _ebss_b_l1 to L1 data bank B SRAM */
-       dma_memcpy(_sdata_b_l1, _l1_lma_start + l1_code_length +
+       /* if necessary, copy _stext_l1 to _etext_l1 to L1 instruction SRAM */
+       l1_code_length = _etext_l1 - _stext_l1;
+       if (l1_code_length)
+               early_dma_memcpy(_stext_l1, _l1_lma_start, l1_code_length);
+
+       /* if necessary, copy _sdata_l1 to _sbss_l1 to L1 data bank A SRAM */
+       l1_data_a_length = _sbss_l1 - _sdata_l1;
+       if (l1_data_a_length)
+               early_dma_memcpy(_sdata_l1, _l1_lma_start + l1_code_length, l1_data_a_length);
+
+       /* if necessary, copy _sdata_b_l1 to _sbss_b_l1 to L1 data bank B SRAM */
+       l1_data_b_length = _sbss_b_l1 - _sdata_b_l1;
+       if (l1_data_b_length)
+               early_dma_memcpy(_sdata_b_l1, _l1_lma_start + l1_code_length +
                        l1_data_a_length, l1_data_b_length);
 
-#ifdef L2_LENGTH
-       l2_length = _ebss_l2 - _stext_l2;
-       if (l2_length > L2_LENGTH)
-               panic("L2 SRAM Overflow\n");
+       early_dma_memcpy_done();
 
-       /* Copy _stext_l2 to _edata_l2 to L2 SRAM */
-       dma_memcpy(_stext_l2, _l2_lma_start, l2_length);
-#endif
+       /* if necessary, copy _stext_l2 to _edata_l2 to L2 SRAM */
+       if (L2_LENGTH != 0) {
+               l2_length = _sbss_l2 - _stext_l2;
+               if (l2_length)
+                       memcpy(_stext_l2, _l2_lma_start, l2_length);
+       }
 }
 
 /* add_memory_region to memmap */
@@ -227,7 +317,7 @@ static int __init sanitize_memmap(struct bfin_memmap_entry *map, int *pnr_map)
        /* record all known change-points (starting and ending addresses),
           omitting those that are for empty memory regions */
        chgidx = 0;
-       for (i = 0; i < old_nr; i++)    {
+       for (i = 0; i < old_nr; i++) {
                if (map[i].size != 0) {
                        change_point[chgidx]->addr = map[i].addr;
                        change_point[chgidx++]->pentry = &map[i];
@@ -235,13 +325,13 @@ static int __init sanitize_memmap(struct bfin_memmap_entry *map, int *pnr_map)
                        change_point[chgidx++]->pentry = &map[i];
                }
        }
-       chg_nr = chgidx;        /* true number of change-points */
+       chg_nr = chgidx;        /* true number of change-points */
 
        /* sort change-point list by memory addresses (low -> high) */
        still_changing = 1;
-       while (still_changing)  {
+       while (still_changing) {
                still_changing = 0;
-               for (i = 1; i < chg_nr; i++)  {
+               for (i = 1; i < chg_nr; i++) {
                        /* if <current_addr> > <last_addr>, swap */
                        /* or, if current=<start_addr> & last=<end_addr>, swap */
                        if ((change_point[i]->addr < change_point[i-1]->addr) ||
@@ -258,10 +348,10 @@ static int __init sanitize_memmap(struct bfin_memmap_entry *map, int *pnr_map)
        }
 
        /* create a new memmap, removing overlaps */
-       overlap_entries = 0;     /* number of entries in the overlap table */
-       new_entry = 0;   /* index for creating new memmap entries */
-       last_type = 0;           /* start with undefined memory type */
-       last_addr = 0;           /* start with 0 as last starting address */
+       overlap_entries = 0;    /* number of entries in the overlap table */
+       new_entry = 0;          /* index for creating new memmap entries */
+       last_type = 0;          /* start with undefined memory type */
+       last_addr = 0;          /* start with 0 as last starting address */
        /* loop through change-points, determining affect on the new memmap */
        for (chgidx = 0; chgidx < chg_nr; chgidx++) {
                /* keep track of all overlapping memmap entries */
@@ -283,14 +373,14 @@ static int __init sanitize_memmap(struct bfin_memmap_entry *map, int *pnr_map)
                        if (overlap_list[i]->type > current_type)
                                current_type = overlap_list[i]->type;
                /* continue building up new memmap based on this information */
-               if (current_type != last_type)  {
+               if (current_type != last_type) {
                        if (last_type != 0) {
                                new_map[new_entry].size =
                                        change_point[chgidx]->addr - last_addr;
                                /* move forward only if the new size was non-zero */
                                if (new_map[new_entry].size != 0)
                                        if (++new_entry >= BFIN_MEMMAP_MAX)
-                                               break;  /* no more space left for new entries */
+                                               break;  /* no more space left for new entries */
                        }
                        if (current_type != 0) {
                                new_map[new_entry].addr = change_point[chgidx]->addr;
@@ -300,9 +390,9 @@ static int __init sanitize_memmap(struct bfin_memmap_entry *map, int *pnr_map)
                        last_type = current_type;
                }
        }
-       new_nr = new_entry;   /* retain count for new entries */
+       new_nr = new_entry;     /* retain count for new entries */
 
-       /* copy new  mapping into original location */
+       /* copy new mapping into original location */
        memcpy(map, new_map, new_nr*sizeof(struct bfin_memmap_entry));
        *pnr_map = new_nr;
 
@@ -319,13 +409,14 @@ static void __init print_memory_map(char *who)
                        bfin_memmap.map[i].addr + bfin_memmap.map[i].size);
                switch (bfin_memmap.map[i].type) {
                case BFIN_MEMMAP_RAM:
-                               printk("(usable)\n");
-                               break;
+                       printk(KERN_CONT "(usable)\n");
+                       break;
                case BFIN_MEMMAP_RESERVED:
-                               printk("(reserved)\n");
-                               break;
-               default:        printk("type %lu\n", bfin_memmap.map[i].type);
-                               break;
+                       printk(KERN_CONT "(reserved)\n");
+                       break;
+               default:
+                       printk(KERN_CONT "type %lu\n", bfin_memmap.map[i].type);
+                       break;
                }
        }
 }
@@ -358,7 +449,6 @@ static __init int parse_memmap(char *arg)
  *  - "memmap=XXX[KkmM][@][$]XXX[KkmM]" defines a memory region
  *       @ from <start> to <start>+<mem>, type RAM
  *       $ from <start> to <start>+<mem>, type RESERVED
- *
  */
 static __init void parse_cmdline_early(char *cmdline_p)
 {
@@ -380,17 +470,20 @@ static __init void parse_cmdline_early(char *cmdline_p)
                                        if (*to != ' ') {
                                                if (*to == '$'
                                                    || *(to + 1) == '$')
-                                                       reserved_mem_dcache_on =
-                                                           1;
+                                                       reserved_mem_dcache_on = 1;
                                                if (*to == '#'
                                                    || *(to + 1) == '#')
-                                                       reserved_mem_icache_on =
-                                                           1;
+                                                       reserved_mem_icache_on = 1;
                                        }
                                }
+                       } else if (!memcmp(to, "clkin_hz=", 9)) {
+                               to += 9;
+                               early_init_clkin_hz(to);
+#ifdef CONFIG_EARLY_PRINTK
                        } else if (!memcmp(to, "earlyprintk=", 12)) {
                                to += 12;
                                setup_early_printk(to);
+#endif
                        } else if (!memcmp(to, "memmap=", 7)) {
                                to += 7;
                                parse_memmap(to);
@@ -409,31 +502,46 @@ static __init void parse_cmdline_early(char *cmdline_p)
  *  [_rambase, _ramstart]:             kernel image
  *  [memory_start, memory_end]:                dynamic memory managed by kernel
  *  [memory_end, _ramend]:             reserved memory
- *     [meory_mtd_start(memory_end),
+ *     [memory_mtd_start(memory_end),
  *             memory_mtd_start + mtd_size]:   rootfs (if any)
  *     [_ramend - DMA_UNCACHED_REGION,
  *             _ramend]:                       uncached DMA region
  *  [_ramend, physical_mem_end]:       memory not managed by kernel
- *
  */
-static __init void  memory_setup(void)
+static __init void memory_setup(void)
 {
 #ifdef CONFIG_MTD_UCLINUX
        unsigned long mtd_phys = 0;
 #endif
+       unsigned long max_mem;
 
        _rambase = (unsigned long)_stext;
        _ramstart = (unsigned long)_end;
 
        if (DMA_UNCACHED_REGION > (_ramend - _ramstart)) {
                console_init();
-               panic("DMA region exceeds memory limit: %lu.\n",
+               panic("DMA region exceeds memory limit: %lu.",
                        _ramend - _ramstart);
        }
-       memory_end = _ramend - DMA_UNCACHED_REGION;
+       max_mem = memory_end = _ramend - DMA_UNCACHED_REGION;
+
+#if (defined(CONFIG_BFIN_EXTMEM_ICACHEABLE) && ANOMALY_05000263)
+       /* Due to a Hardware Anomaly we need to limit the size of usable
+        * instruction memory to max 60MB, 56 if HUNT_FOR_ZERO is on
+        * 05000263 - Hardware loop corrupted when taking an ICPLB exception
+        */
+# if (defined(CONFIG_DEBUG_HUNT_FOR_ZERO))
+       if (max_mem >= 56 * 1024 * 1024)
+               max_mem = 56 * 1024 * 1024;
+# else
+       if (max_mem >= 60 * 1024 * 1024)
+               max_mem = 60 * 1024 * 1024;
+# endif                                /* CONFIG_DEBUG_HUNT_FOR_ZERO */
+#endif                         /* ANOMALY_05000263 */
+
 
 #ifdef CONFIG_MPU
-       /* Round up to multiple of 4MB */
+       /* Round up to multiple of 4MB */
        memory_start = (_ramstart + 0x3fffff) & ~0x3fffff;
 #else
        memory_start = PAGE_ALIGN(_ramstart);
@@ -459,52 +567,45 @@ static __init void  memory_setup(void)
 
 # if defined(CONFIG_ROMFS_FS)
        if (((unsigned long *)mtd_phys)[0] == ROMSB_WORD0
-           && ((unsigned long *)mtd_phys)[1] == ROMSB_WORD1)
+           && ((unsigned long *)mtd_phys)[1] == ROMSB_WORD1) {
                mtd_size =
                    PAGE_ALIGN(be32_to_cpu(((unsigned long *)mtd_phys)[2]));
-#  if (defined(CONFIG_BFIN_ICACHE) && ANOMALY_05000263)
-       /* Due to a Hardware Anomaly we need to limit the size of usable
-        * instruction memory to max 60MB, 56 if HUNT_FOR_ZERO is on
-        * 05000263 - Hardware loop corrupted when taking an ICPLB exception
-        */
-#   if (defined(CONFIG_DEBUG_HUNT_FOR_ZERO))
-       if (memory_end >= 56 * 1024 * 1024)
-               memory_end = 56 * 1024 * 1024;
-#   else
-       if (memory_end >= 60 * 1024 * 1024)
-               memory_end = 60 * 1024 * 1024;
-#   endif                              /* CONFIG_DEBUG_HUNT_FOR_ZERO */
-#  endif                               /* ANOMALY_05000263 */
-# endif                                /* CONFIG_ROMFS_FS */
-
-       memory_end -= mtd_size;
 
-       if (mtd_size == 0) {
-               console_init();
-               panic("Don't boot kernel without rootfs attached.\n");
+               /* ROM_FS is XIP, so if we found it, we need to limit memory */
+               if (memory_end > max_mem) {
+                       pr_info("Limiting kernel memory to %liMB due to anomaly 05000263\n", max_mem >> 20);
+                       memory_end = max_mem;
+               }
        }
+# endif                                /* CONFIG_ROMFS_FS */
 
-       /* Relocate MTD image to the top of memory after the uncached memory area */
-       dma_memcpy((char *)memory_end, _end, mtd_size);
-
-       memory_mtd_start = memory_end;
-       _ebss = memory_mtd_start;       /* define _ebss for compatible */
+       /* Since the default MTD_UCLINUX has no magic number, we just blindly
+        * read 8 past the end of the kernel's image, and look at it.
+        * When no image is attached, mtd_size is set to a random number
+        * Do some basic sanity checks before operating on things
+        */
+       if (mtd_size == 0 || memory_end <= mtd_size) {
+               pr_emerg("Could not find valid ram mtd attached.\n");
+       } else {
+               memory_end -= mtd_size;
+
+               /* Relocate MTD image to the top of memory after the uncached memory area */
+               uclinux_ram_map.phys = memory_mtd_start = memory_end;
+               uclinux_ram_map.size = mtd_size;
+               pr_info("Found mtd parition at 0x%p, (len=0x%lx), moving to 0x%p\n",
+                       _end, mtd_size, (void *)memory_mtd_start);
+               dma_memcpy((void *)uclinux_ram_map.phys, _end, uclinux_ram_map.size);
+       }
 #endif                         /* CONFIG_MTD_UCLINUX */
 
-#if (defined(CONFIG_BFIN_ICACHE) && ANOMALY_05000263)
-       /* Due to a Hardware Anomaly we need to limit the size of usable
-        * instruction memory to max 60MB, 56 if HUNT_FOR_ZERO is on
-        * 05000263 - Hardware loop corrupted when taking an ICPLB exception
+       /* We need lo limit memory, since everything could have a text section
+        * of userspace in it, and expose anomaly 05000263. If the anomaly
+        * doesn't exist, or we don't need to - then dont.
         */
-#if (defined(CONFIG_DEBUG_HUNT_FOR_ZERO))
-       if (memory_end >= 56 * 1024 * 1024)
-               memory_end = 56 * 1024 * 1024;
-#else
-       if (memory_end >= 60 * 1024 * 1024)
-               memory_end = 60 * 1024 * 1024;
-#endif                         /* CONFIG_DEBUG_HUNT_FOR_ZERO */
-       printk(KERN_NOTICE "Warning: limiting memory to %liMB due to hardware anomaly 05000263\n", memory_end >> 20);
-#endif                         /* ANOMALY_05000263 */
+       if (memory_end > max_mem) {
+               pr_info("Limiting kernel memory to %liMB due to anomaly 05000263\n", max_mem >> 20);
+               memory_end = max_mem;
+       }
 
 #ifdef CONFIG_MPU
        page_mask_nelts = ((_ramend >> PAGE_SHIFT) + 31) / 32;
@@ -525,19 +626,19 @@ static __init void  memory_setup(void)
        printk(KERN_INFO "Kernel Managed Memory: %ldMB\n", _ramend >> 20);
 
        printk(KERN_INFO "Memory map:\n"
-               KERN_INFO "  fixedcode = 0x%p-0x%p\n"
-               KERN_INFO "  text      = 0x%p-0x%p\n"
-               KERN_INFO "  rodata    = 0x%p-0x%p\n"
-               KERN_INFO "  bss       = 0x%p-0x%p\n"
-               KERN_INFO "  data      = 0x%p-0x%p\n"
-               KERN_INFO "    stack   = 0x%p-0x%p\n"
-               KERN_INFO "  init      = 0x%p-0x%p\n"
-               KERN_INFO "  available = 0x%p-0x%p\n"
+              "  fixedcode = 0x%p-0x%p\n"
+              "  text      = 0x%p-0x%p\n"
+              "  rodata    = 0x%p-0x%p\n"
+              "  bss       = 0x%p-0x%p\n"
+              "  data      = 0x%p-0x%p\n"
+              "    stack   = 0x%p-0x%p\n"
+              "  init      = 0x%p-0x%p\n"
+              "  available = 0x%p-0x%p\n"
 #ifdef CONFIG_MTD_UCLINUX
-               KERN_INFO "  rootfs    = 0x%p-0x%p\n"
+              "  rootfs    = 0x%p-0x%p\n"
 #endif
 #if DMA_UNCACHED_REGION > 0
-               KERN_INFO "  DMA Zone  = 0x%p-0x%p\n"
+              "  DMA Zone  = 0x%p-0x%p\n"
 #endif
                , (void *)FIXED_CODE_START, (void *)FIXED_CODE_END,
                _stext, _etext,
@@ -613,7 +714,7 @@ static __init void setup_bootmem_allocator(void)
        end_pfn = memory_end >> PAGE_SHIFT;
 
        /*
-        * give all the memory to the bootmap allocator,  tell it to put the
+        * give all the memory to the bootmap allocator, tell it to put the
         * boot mem_map at the start of memory.
         */
        bootmap_size = init_bootmem_node(NODE_DATA(0),
@@ -711,6 +812,13 @@ void __init setup_arch(char **cmdline_p)
 {
        unsigned long sclk, cclk;
 
+       enable_shadow_console();
+
+       /* Check to make sure we are running on the right processor */
+       if (unlikely(CPUID != bfin_cpuid()))
+               printk(KERN_ERR "ERROR: Not running on ADSP-%s: unknown CPUID 0x%04x Rev 0.%d\n",
+                       CPU, bfin_cpuid(), bfin_revid());
+
 #ifdef CONFIG_DUMMY_CONSOLE
        conswitchp = &dummy_con;
 #endif
@@ -725,26 +833,37 @@ void __init setup_arch(char **cmdline_p)
        memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
        boot_command_line[COMMAND_LINE_SIZE - 1] = '\0';
 
-       /* setup memory defaults from the user config */
-       physical_mem_end = 0;
-       _ramend = get_mem_size() * 1024 * 1024;
-
        memset(&bfin_memmap, 0, sizeof(bfin_memmap));
 
+       /* If the user does not specify things on the command line, use
+        * what the bootloader set things up as
+        */
+       physical_mem_end = 0;
        parse_cmdline_early(&command_line[0]);
 
+       if (_ramend == 0)
+               _ramend = get_mem_size() * 1024 * 1024;
+
        if (physical_mem_end == 0)
                physical_mem_end = _ramend;
 
        memory_setup();
 
+       /* Initialize Async memory banks */
+       bfin_write_EBIU_AMBCTL0(AMBCTL0VAL);
+       bfin_write_EBIU_AMBCTL1(AMBCTL1VAL);
+       bfin_write_EBIU_AMGCTL(AMGCTLVAL);
+#ifdef CONFIG_EBIU_MBSCTLVAL
+       bfin_write_EBIU_MBSCTL(CONFIG_EBIU_MBSCTLVAL);
+       bfin_write_EBIU_MODE(CONFIG_EBIU_MODEVAL);
+       bfin_write_EBIU_FCTL(CONFIG_EBIU_FCTLVAL);
+#endif
+
        cclk = get_cclk();
        sclk = get_sclk();
 
-#if !defined(CONFIG_BFIN_KERNEL_CLOCK)
-       if (ANOMALY_05000273 && cclk == sclk)
-               panic("ANOMALY 05000273, SCLK can not be same as CCLK");
-#endif
+       if ((ANOMALY_05000273 || ANOMALY_05000274) && (cclk >> 1) < sclk)
+               panic("ANOMALY 05000273 or 05000274: CCLK must be >= 2*SCLK");
 
 #ifdef BF561_FAMILY
        if (ANOMALY_05000266) {
@@ -754,55 +873,81 @@ void __init setup_arch(char **cmdline_p)
 #endif
        printk(KERN_INFO "Hardware Trace ");
        if (bfin_read_TBUFCTL() & 0x1)
-               printk("Active ");
+               printk(KERN_CONT "Active ");
        else
-               printk("Off ");
+               printk(KERN_CONT "Off ");
        if (bfin_read_TBUFCTL() & 0x2)
-               printk("and Enabled\n");
+               printk(KERN_CONT "and Enabled\n");
        else
-       printk("and Disabled\n");
+               printk(KERN_CONT "and Disabled\n");
 
-#if defined(CONFIG_CHR_DEV_FLASH) || defined(CONFIG_BLK_DEV_FLASH)
-       /* we need to initialize the Flashrom device here since we might
-        * do things with flash early on in the boot
-        */
-       flash_probe();
-#endif
+       printk(KERN_INFO "Boot Mode: %i\n", bfin_read_SYSCR() & 0xF);
 
+       /* Newer parts mirror SWRST bits in SYSCR */
+#if defined(CONFIG_BF53x) || defined(CONFIG_BF561) || \
+    defined(CONFIG_BF538) || defined(CONFIG_BF539)
        _bfin_swrst = bfin_read_SWRST();
+#else
+       /* Clear boot mode field */
+       _bfin_swrst = bfin_read_SYSCR() & ~0xf;
+#endif
 
-       if (_bfin_swrst & RESET_DOUBLE)
-               printk(KERN_INFO "Recovering from Double Fault event\n");
-       else if (_bfin_swrst & RESET_WDOG)
+#ifdef CONFIG_DEBUG_DOUBLEFAULT_PRINT
+       bfin_write_SWRST(_bfin_swrst & ~DOUBLE_FAULT);
+#endif
+#ifdef CONFIG_DEBUG_DOUBLEFAULT_RESET
+       bfin_write_SWRST(_bfin_swrst | DOUBLE_FAULT);
+#endif
+
+#ifdef CONFIG_SMP
+       if (_bfin_swrst & SWRST_DBL_FAULT_A) {
+#else
+       if (_bfin_swrst & RESET_DOUBLE) {
+#endif
+               printk(KERN_EMERG "Recovering from DOUBLE FAULT event\n");
+#ifdef CONFIG_DEBUG_DOUBLEFAULT
+               /* We assume the crashing kernel, and the current symbol table match */
+               printk(KERN_EMERG " While handling exception (EXCAUSE = 0x%x) at %pF\n",
+                       (int)init_saved_seqstat & SEQSTAT_EXCAUSE, init_saved_retx);
+               printk(KERN_NOTICE "   DCPLB_FAULT_ADDR: %pF\n", init_saved_dcplb_fault_addr);
+               printk(KERN_NOTICE "   ICPLB_FAULT_ADDR: %pF\n", init_saved_icplb_fault_addr);
+#endif
+               printk(KERN_NOTICE " The instruction at %pF caused a double exception\n",
+                       init_retx);
+       } else if (_bfin_swrst & RESET_WDOG)
                printk(KERN_INFO "Recovering from Watchdog event\n");
        else if (_bfin_swrst & RESET_SOFTWARE)
                printk(KERN_NOTICE "Reset caused by Software reset\n");
 
-       printk(KERN_INFO "Blackfin support (C) 2004-2008 Analog Devices, Inc.\n");
+       printk(KERN_INFO "Blackfin support (C) 2004-2009 Analog Devices, Inc.\n");
        if (bfin_compiled_revid() == 0xffff)
                printk(KERN_INFO "Compiled for ADSP-%s Rev any\n", CPU);
        else if (bfin_compiled_revid() == -1)
                printk(KERN_INFO "Compiled for ADSP-%s Rev none\n", CPU);
        else
                printk(KERN_INFO "Compiled for ADSP-%s Rev 0.%d\n", CPU, bfin_compiled_revid());
-       if (bfin_revid() != bfin_compiled_revid()) {
-               if (bfin_compiled_revid() == -1)
-                       printk(KERN_ERR "Warning: Compiled for Rev none, but running on Rev %d\n",
-                              bfin_revid());
-               else if (bfin_compiled_revid() != 0xffff)
-                       printk(KERN_ERR "Warning: Compiled for Rev %d, but running on Rev %d\n",
-                              bfin_compiled_revid(), bfin_revid());
+
+       if (likely(CPUID == bfin_cpuid())) {
+               if (bfin_revid() != bfin_compiled_revid()) {
+                       if (bfin_compiled_revid() == -1)
+                               printk(KERN_ERR "Warning: Compiled for Rev none, but running on Rev %d\n",
+                                      bfin_revid());
+                       else if (bfin_compiled_revid() != 0xffff) {
+                               printk(KERN_ERR "Warning: Compiled for Rev %d, but running on Rev %d\n",
+                                      bfin_compiled_revid(), bfin_revid());
+                               if (bfin_compiled_revid() > bfin_revid())
+                                       panic("Error: you are missing anomaly workarounds for this rev");
+                       }
+               }
+               if (bfin_revid() < CONFIG_BF_REV_MIN || bfin_revid() > CONFIG_BF_REV_MAX)
+                       printk(KERN_ERR "Warning: Unsupported Chip Revision ADSP-%s Rev 0.%d detected\n",
+                              CPU, bfin_revid());
        }
-       if (bfin_revid() < SUPPORTED_REVID)
-               printk(KERN_ERR "Warning: Unsupported Chip Revision ADSP-%s Rev 0.%d detected\n",
-                      CPU, bfin_revid());
+
        printk(KERN_INFO "Blackfin Linux support by http://blackfin.uclinux.org/\n");
 
        printk(KERN_INFO "Processor Speed: %lu MHz core clock and %lu MHz System Clock\n",
-              cclk / 1000000,  sclk / 1000000);
-
-       if (ANOMALY_05000273 && (cclk >> 1) <= sclk)
-               printk("\n\n\nANOMALY_05000273: CCLK must be >= 2*SCLK !!!\n\n\n");
+              cclk / 1000000, sclk / 1000000);
 
        setup_bootmem_allocator();
 
@@ -831,18 +976,21 @@ void __init setup_arch(char **cmdline_p)
        BUG_ON((char *)&safe_user_instruction - (char *)&fixed_code_start
                != SAFE_USER_INSTRUCTION - FIXED_CODE_START);
 
+#ifdef CONFIG_SMP
+       platform_init_cpus();
+#endif
        init_exception_vectors();
-       bf53x_cache_init();
+       bfin_cache_init();      /* Initialize caches for the boot CPU */
 }
 
 static int __init topology_init(void)
 {
-       int cpu;
+       unsigned int cpu;
+       /* Record CPU-private information for the boot processor. */
+       bfin_setup_cpudata(0);
 
        for_each_possible_cpu(cpu) {
-               struct cpu *c = &per_cpu(cpu_devices, cpu);
-
-               register_cpu(c, cpu);
+               register_cpu(&per_cpu(cpu_data, cpu).cpu, cpu);
        }
 
        return 0;
@@ -850,52 +998,94 @@ static int __init topology_init(void)
 
 subsys_initcall(topology_init);
 
+/* Get the input clock frequency */
+static u_long cached_clkin_hz = CONFIG_CLKIN_HZ;
+static u_long get_clkin_hz(void)
+{
+       return cached_clkin_hz;
+}
+static int __init early_init_clkin_hz(char *buf)
+{
+       cached_clkin_hz = simple_strtoul(buf, NULL, 0);
+#ifdef BFIN_KERNEL_CLOCK
+       if (cached_clkin_hz != CONFIG_CLKIN_HZ)
+               panic("cannot change clkin_hz when reprogramming clocks");
+#endif
+       return 1;
+}
+early_param("clkin_hz=", early_init_clkin_hz);
+
+/* Get the voltage input multiplier */
 static u_long get_vco(void)
 {
-       u_long msel;
-       u_long vco;
+       static u_long cached_vco;
+       u_long msel, pll_ctl;
+
+       /* The assumption here is that VCO never changes at runtime.
+        * If, someday, we support that, then we'll have to change this.
+        */
+       if (cached_vco)
+               return cached_vco;
 
-       msel = (bfin_read_PLL_CTL() >> 9) & 0x3F;
+       pll_ctl = bfin_read_PLL_CTL();
+       msel = (pll_ctl >> 9) & 0x3F;
        if (0 == msel)
                msel = 64;
 
-       vco = CONFIG_CLKIN_HZ;
-       vco >>= (1 & bfin_read_PLL_CTL());      /* DF bit */
-       vco = msel * vco;
-       return vco;
+       cached_vco = get_clkin_hz();
+       cached_vco >>= (1 & pll_ctl);   /* DF bit */
+       cached_vco *= msel;
+       return cached_vco;
 }
 
 /* Get the Core clock */
 u_long get_cclk(void)
 {
+       static u_long cached_cclk_pll_div, cached_cclk;
        u_long csel, ssel;
+
        if (bfin_read_PLL_STAT() & 0x1)
-               return CONFIG_CLKIN_HZ;
+               return get_clkin_hz();
 
        ssel = bfin_read_PLL_DIV();
+       if (ssel == cached_cclk_pll_div)
+               return cached_cclk;
+       else
+               cached_cclk_pll_div = ssel;
+
        csel = ((ssel >> 4) & 0x03);
        ssel &= 0xf;
        if (ssel && ssel < (1 << csel)) /* SCLK > CCLK */
-               return get_vco() / ssel;
-       return get_vco() >> csel;
+               cached_cclk = get_vco() / ssel;
+       else
+               cached_cclk = get_vco() >> csel;
+       return cached_cclk;
 }
 EXPORT_SYMBOL(get_cclk);
 
 /* Get the System clock */
 u_long get_sclk(void)
 {
+       static u_long cached_sclk;
        u_long ssel;
 
+       /* The assumption here is that SCLK never changes at runtime.
+        * If, someday, we support that, then we'll have to change this.
+        */
+       if (cached_sclk)
+               return cached_sclk;
+
        if (bfin_read_PLL_STAT() & 0x1)
-               return CONFIG_CLKIN_HZ;
+               return get_clkin_hz();
 
-       ssel = (bfin_read_PLL_DIV() & 0xf);
+       ssel = bfin_read_PLL_DIV() & 0xf;
        if (0 == ssel) {
                printk(KERN_WARNING "Invalid System Clock\n");
                ssel = 1;
        }
 
-       return get_vco() / ssel;
+       cached_sclk = get_vco() / ssel;
+       return cached_sclk;
 }
 EXPORT_SYMBOL(get_sclk);
 
@@ -922,17 +1112,18 @@ static int show_cpuinfo(struct seq_file *m, void *v)
 {
        char *cpu, *mmu, *fpu, *vendor, *cache;
        uint32_t revid;
-
-       u_long cclk = 0, sclk = 0;
-       u_int dcache_size = 0, dsup_banks = 0;
+       int cpu_num = *(unsigned int *)v;
+       u_long sclk, cclk;
+       u_int icache_size = BFIN_ICACHESIZE / 1024, dcache_size = 0, dsup_banks = 0;
+       struct blackfin_cpudata *cpudata = &per_cpu(cpu_data, cpu_num);
 
        cpu = CPU;
        mmu = "none";
        fpu = "none";
        revid = bfin_revid();
 
-       cclk = get_cclk();
        sclk = get_sclk();
+       cclk = get_cclk();
 
        switch (bfin_read_CHIPID() & CHIPID_MANUFACTURE) {
        case 0xca:
@@ -943,14 +1134,16 @@ static int show_cpuinfo(struct seq_file *m, void *v)
                break;
        }
 
-       seq_printf(m, "processor\t: %d\n"
-               "vendor_id\t: %s\n"
-               "cpu family\t: 0x%x\n"
-               "model name\t: ADSP-%s %lu(MHz CCLK) %lu(MHz SCLK) (%s)\n"
-               "stepping\t: %d\n",
-               0,
-               vendor,
-               (bfin_read_CHIPID() & CHIPID_FAMILY),
+       seq_printf(m, "processor\t: %d\n" "vendor_id\t: %s\n", cpu_num, vendor);
+
+       if (CPUID == bfin_cpuid())
+               seq_printf(m, "cpu family\t: 0x%04x\n", CPUID);
+       else
+               seq_printf(m, "cpu family\t: Compiled for:0x%04x, running on:0x%04x\n",
+                       CPUID, bfin_cpuid());
+
+       seq_printf(m, "model name\t: ADSP-%s %lu(MHz CCLK) %lu(MHz SCLK) (%s)\n"
+               "stepping\t: %d ",
                cpu, cclk/1000000, sclk/1000000,
 #ifdef CONFIG_MPU
                "mpu on",
@@ -959,7 +1152,16 @@ static int show_cpuinfo(struct seq_file *m, void *v)
 #endif
                revid);
 
-       seq_printf(m, "cpu MHz\t\t: %lu.%03lu/%lu.%03lu\n",
+       if (bfin_revid() != bfin_compiled_revid()) {
+               if (bfin_compiled_revid() == -1)
+                       seq_printf(m, "(Compiled for Rev none)");
+               else if (bfin_compiled_revid() == 0xffff)
+                       seq_printf(m, "(Compiled for Rev any)");
+               else
+                       seq_printf(m, "(Compiled for Rev %d)", bfin_compiled_revid());
+       }
+
+       seq_printf(m, "\ncpu MHz\t\t: %lu.%03lu/%lu.%03lu\n",
                cclk/1000000, cclk%1000000,
                sclk/1000000, sclk%1000000);
        seq_printf(m, "bogomips\t: %lu.%02lu\n"
@@ -969,7 +1171,7 @@ static int show_cpuinfo(struct seq_file *m, void *v)
                (loops_per_jiffy * HZ));
 
        /* Check Cache configutation */
-       switch (bfin_read_DMEM_CONTROL() & (1 << DMC0_P | 1 << DMC1_P)) {
+       switch (cpudata->dmemctl & (1 << DMC0_P | 1 << DMC1_P)) {
        case ACACHE_BSRAM:
                cache = "dbank-A/B\t: cache/sram";
                dcache_size = 16;
@@ -993,78 +1195,72 @@ static int show_cpuinfo(struct seq_file *m, void *v)
        }
 
        /* Is it turned on? */
-       if (!((bfin_read_DMEM_CONTROL()) & (ENDCPLB | DMC_ENABLE)))
+       if ((cpudata->dmemctl & (ENDCPLB | DMC_ENABLE)) != (ENDCPLB | DMC_ENABLE))
                dcache_size = 0;
 
+       if ((cpudata->imemctl & (IMC | ENICPLB)) != (IMC | ENICPLB))
+               icache_size = 0;
+
        seq_printf(m, "cache size\t: %d KB(L1 icache) "
-               "%d KB(L1 dcache-%s) %d KB(L2 cache)\n",
-               BFIN_ICACHESIZE / 1024, dcache_size,
-#if defined CONFIG_BFIN_WB
-               "wb"
-#elif defined CONFIG_BFIN_WT
-               "wt"
+               "%d KB(L1 dcache) %d KB(L2 cache)\n",
+               icache_size, dcache_size, 0);
+       seq_printf(m, "%s\n", cache);
+       seq_printf(m, "external memory\t: "
+#if defined(CONFIG_BFIN_EXTMEM_ICACHEABLE)
+                  "cacheable"
+#else
+                  "uncacheable"
 #endif
-               "", 0);
+                  " in instruction cache\n");
+       seq_printf(m, "external memory\t: "
+#if defined(CONFIG_BFIN_EXTMEM_WRITEBACK)
+                     "cacheable (write-back)"
+#elif defined(CONFIG_BFIN_EXTMEM_WRITETHROUGH)
+                     "cacheable (write-through)"
+#else
+                     "uncacheable"
+#endif
+                     " in data cache\n");
 
-       seq_printf(m, "%s\n", cache);
+       if (icache_size)
+               seq_printf(m, "icache setup\t: %d Sub-banks/%d Ways, %d Lines/Way\n",
+                          BFIN_ISUBBANKS, BFIN_IWAYS, BFIN_ILINES);
+       else
+               seq_printf(m, "icache setup\t: off\n");
 
-       seq_printf(m, "icache setup\t: %d Sub-banks/%d Ways, %d Lines/Way\n",
-                  BFIN_ISUBBANKS, BFIN_IWAYS, BFIN_ILINES);
        seq_printf(m,
                   "dcache setup\t: %d Super-banks/%d Sub-banks/%d Ways, %d Lines/Way\n",
                   dsup_banks, BFIN_DSUBBANKS, BFIN_DWAYS,
                   BFIN_DLINES);
-#ifdef CONFIG_BFIN_ICACHE_LOCK
-       switch (read_iloc()) {
-       case WAY0_L:
-               seq_printf(m, "Way0 Locked-Down\n");
-               break;
-       case WAY1_L:
-               seq_printf(m, "Way1 Locked-Down\n");
-               break;
-       case WAY01_L:
-               seq_printf(m, "Way0,Way1 Locked-Down\n");
-               break;
-       case WAY2_L:
-               seq_printf(m, "Way2 Locked-Down\n");
-               break;
-       case WAY02_L:
-               seq_printf(m, "Way0,Way2 Locked-Down\n");
-               break;
-       case WAY12_L:
-               seq_printf(m, "Way1,Way2 Locked-Down\n");
-               break;
-       case WAY012_L:
-               seq_printf(m, "Way0,Way1 & Way2 Locked-Down\n");
-               break;
-       case WAY3_L:
-               seq_printf(m, "Way3 Locked-Down\n");
-               break;
-       case WAY03_L:
-               seq_printf(m, "Way0,Way3 Locked-Down\n");
-               break;
-       case WAY13_L:
-               seq_printf(m, "Way1,Way3 Locked-Down\n");
-               break;
-       case WAY013_L:
-               seq_printf(m, "Way 0,Way1,Way3 Locked-Down\n");
-               break;
-       case WAY32_L:
-               seq_printf(m, "Way3,Way2 Locked-Down\n");
-               break;
-       case WAY320_L:
-               seq_printf(m, "Way3,Way2,Way0 Locked-Down\n");
-               break;
-       case WAY321_L:
-               seq_printf(m, "Way3,Way2,Way1 Locked-Down\n");
-               break;
-       case WAYALL_L:
-               seq_printf(m, "All Ways are locked\n");
-               break;
-       default:
-               seq_printf(m, "No Ways are locked\n");
-       }
+#ifdef __ARCH_SYNC_CORE_DCACHE
+       seq_printf(m, "SMP Dcache Flushes\t: %lu\n\n", cpudata->dcache_invld_count);
 #endif
+#ifdef __ARCH_SYNC_CORE_ICACHE
+       seq_printf(m, "SMP Icache Flushes\t: %lu\n\n", cpudata->icache_invld_count);
+#endif
+
+       if (cpu_num != num_possible_cpus() - 1)
+               return 0;
+
+       if (L2_LENGTH) {
+               seq_printf(m, "L2 SRAM\t\t: %dKB\n", L2_LENGTH/0x400);
+               seq_printf(m, "L2 SRAM\t\t: "
+#if defined(CONFIG_BFIN_L2_ICACHEABLE)
+                             "cacheable"
+#else
+                             "uncacheable"
+#endif
+                             " in instruction cache\n");
+               seq_printf(m, "L2 SRAM\t\t: "
+#if defined(CONFIG_BFIN_L2_WRITEBACK)
+                             "cacheable (write-back)"
+#elif defined(CONFIG_BFIN_L2_WRITETHROUGH)
+                             "cacheable (write-through)"
+#else
+                             "uncacheable"
+#endif
+                             " in data cache\n");
+       }
        seq_printf(m, "board name\t: %s\n", bfin_board_name);
        seq_printf(m, "board memory\t: %ld kB (0x%p -> 0x%p)\n",
                 physical_mem_end >> 10, (void *)0, (void *)physical_mem_end);
@@ -1072,18 +1268,25 @@ static int show_cpuinfo(struct seq_file *m, void *v)
                ((int)memory_end - (int)_stext) >> 10,
                _stext,
                (void *)memory_end);
+       seq_printf(m, "\n");
 
        return 0;
 }
 
 static void *c_start(struct seq_file *m, loff_t *pos)
 {
-       return *pos < NR_CPUS ? ((void *)0x12345678) : NULL;
+       if (*pos == 0)
+               *pos = first_cpu(cpu_online_map);
+       if (*pos >= num_online_cpus())
+               return NULL;
+
+       return pos;
 }
 
 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       ++*pos;
+       *pos = next_cpu(*pos, cpu_online_map);
+
        return c_start(m, pos);
 }
 
@@ -1100,6 +1303,7 @@ const struct seq_operations cpuinfo_op = {
 
 void __init cmdline_init(const char *r0)
 {
+       early_shadow_stamp();
        if (r0)
                strncpy(command_line, r0, COMMAND_LINE_SIZE);
 }