sh: Populate initial secondary CPU info from boot_cpu_data.
[safe/jmp/linux-2.6] / arch / powerpc / kernel / prom.c
index eac97f4..d4405b9 100644 (file)
 #include <linux/kexec.h>
 #include <linux/debugfs.h>
 #include <linux/irq.h>
+#include <linux/lmb.h>
 
 #include <asm/prom.h>
 #include <asm/rtas.h>
-#include <asm/lmb.h>
 #include <asm/page.h>
 #include <asm/processor.h>
 #include <asm/irq.h>
@@ -51,7 +51,9 @@
 #include <asm/machdep.h>
 #include <asm/pSeries_reconfig.h>
 #include <asm/pci-bridge.h>
+#include <asm/phyp_dump.h>
 #include <asm/kexec.h>
+#include <mm/mmu_decl.h>
 
 #ifdef DEBUG
 #define DBG(fmt...) printk(KERN_ERR fmt)
@@ -424,7 +426,7 @@ static int __init early_parse_mem(char *p)
                return 1;
 
        memory_limit = PAGE_ALIGN(memparse(p, &p));
-       DBG("memory limit = 0x%lx\n", memory_limit);
+       DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
 
        return 0;
 }
@@ -436,7 +438,7 @@ early_param("mem", early_parse_mem);
  * The device tree may be allocated beyond our memory limit, or inside the
  * crash kernel region for kdump. If so, move it out of the way.
  */
-static void move_device_tree(void)
+static void __init move_device_tree(void)
 {
        unsigned long start, size;
        void *p;
@@ -583,11 +585,16 @@ static void __init check_cpu_pa_features(unsigned long node)
                      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
 }
 
-#ifdef CONFIG_PPC64
+#ifdef CONFIG_PPC_STD_MMU_64
 static void __init check_cpu_slb_size(unsigned long node)
 {
        u32 *slb_size_ptr;
 
+       slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
+       if (slb_size_ptr != NULL) {
+               mmu_slb_size = *slb_size_ptr;
+               return;
+       }
        slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
        if (slb_size_ptr != NULL) {
                mmu_slb_size = *slb_size_ptr;
@@ -607,6 +614,10 @@ static struct feature_property {
        {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
        {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
 #endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_VSX
+       /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
+       {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
+#endif /* CONFIG_VSX */
 #ifdef CONFIG_PPC64
        {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
        {"ibm,purr", 1, CPU_FTR_PURR, 0},
@@ -818,11 +829,11 @@ static int __init early_init_dt_scan_chosen(unsigned long node,
 #endif
 
 #ifdef CONFIG_KEXEC
-       lprop = (u64*)of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
+       lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
        if (lprop)
                crashk_res.start = *lprop;
 
-       lprop = (u64*)of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
+       lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
        if (lprop)
                crashk_res.end = crashk_res.start + *lprop - 1;
 #endif
@@ -882,16 +893,17 @@ static u64 __init dt_mem_next_cell(int s, cell_t **cellp)
  */
 static int __init early_init_dt_scan_drconf_memory(unsigned long node)
 {
-       cell_t *dm, *ls;
+       cell_t *dm, *ls, *usm;
        unsigned long l, n, flags;
        u64 base, size, lmb_size;
+       unsigned int is_kexec_kdump = 0, rngs;
 
-       ls = (cell_t *)of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
+       ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
        if (ls == NULL || l < dt_root_size_cells * sizeof(cell_t))
                return 0;
        lmb_size = dt_mem_next_cell(dt_root_size_cells, &ls);
 
-       dm = (cell_t *)of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
+       dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
        if (dm == NULL || l < sizeof(cell_t))
                return 0;
 
@@ -899,6 +911,12 @@ static int __init early_init_dt_scan_drconf_memory(unsigned long node)
        if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(cell_t))
                return 0;
 
+       /* check if this is a kexec/kdump kernel. */
+       usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
+                                                &l);
+       if (usm != NULL)
+               is_kexec_kdump = 1;
+
        for (; n != 0; --n) {
                base = dt_mem_next_cell(dt_root_addr_cells, &dm);
                flags = dm[3];
@@ -909,13 +927,34 @@ static int __init early_init_dt_scan_drconf_memory(unsigned long node)
                if ((flags & 0x80) || !(flags & 0x8))
                        continue;
                size = lmb_size;
-               if (iommu_is_off) {
-                       if (base >= 0x80000000ul)
+               rngs = 1;
+               if (is_kexec_kdump) {
+                       /*
+                        * For each lmb in ibm,dynamic-memory, a corresponding
+                        * entry in linux,drconf-usable-memory property contains
+                        * a counter 'p' followed by 'p' (base, size) duple.
+                        * Now read the counter from
+                        * linux,drconf-usable-memory property
+                        */
+                       rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
+                       if (!rngs) /* there are no (base, size) duple */
                                continue;
-                       if ((base + size) > 0x80000000ul)
-                               size = 0x80000000ul - base;
                }
-               lmb_add(base, size);
+               do {
+                       if (is_kexec_kdump) {
+                               base = dt_mem_next_cell(dt_root_addr_cells,
+                                                        &usm);
+                               size = dt_mem_next_cell(dt_root_size_cells,
+                                                        &usm);
+                       }
+                       if (iommu_is_off) {
+                               if (base >= 0x80000000ul)
+                                       continue;
+                               if ((base + size) > 0x80000000ul)
+                                       size = 0x80000000ul - base;
+                       }
+                       lmb_add(base, size);
+               } while (--rngs);
        }
        lmb_dump_all();
        return 0;
@@ -947,9 +986,9 @@ static int __init early_init_dt_scan_memory(unsigned long node,
        } else if (strcmp(type, "memory") != 0)
                return 0;
 
-       reg = (cell_t *)of_get_flat_dt_prop(node, "linux,usable-memory", &l);
+       reg = of_get_flat_dt_prop(node, "linux,usable-memory", &l);
        if (reg == NULL)
-               reg = (cell_t *)of_get_flat_dt_prop(node, "reg", &l);
+               reg = of_get_flat_dt_prop(node, "reg", &l);
        if (reg == NULL)
                return 0;
 
@@ -977,7 +1016,10 @@ static int __init early_init_dt_scan_memory(unsigned long node,
                }
 #endif
                lmb_add(base, size);
+
+               memstart_addr = min((u64)memstart_addr, base);
        }
+
        return 0;
 }
 
@@ -1033,15 +1075,93 @@ static void __init early_reserve_mem(void)
                DBG("reserving: %llx -> %llx\n", base, size);
                lmb_reserve(base, size);
        }
+}
 
-#if 0
-       DBG("memory reserved, lmbs :\n");
-       lmb_dump_all();
-#endif
+#ifdef CONFIG_PHYP_DUMP
+/**
+ * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
+ *
+ * Function to find the largest size we need to reserve
+ * during early boot process.
+ *
+ * It either looks for boot param and returns that OR
+ * returns larger of 256 or 5% rounded down to multiples of 256MB.
+ *
+ */
+static inline unsigned long phyp_dump_calculate_reserve_size(void)
+{
+       unsigned long tmp;
+
+       if (phyp_dump_info->reserve_bootvar)
+               return phyp_dump_info->reserve_bootvar;
+
+       /* divide by 20 to get 5% of value */
+       tmp = lmb_end_of_DRAM();
+       do_div(tmp, 20);
+
+       /* round it down in multiples of 256 */
+       tmp = tmp & ~0x0FFFFFFFUL;
+
+       return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
 }
 
+/**
+ * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
+ *
+ * This routine may reserve memory regions in the kernel only
+ * if the system is supported and a dump was taken in last
+ * boot instance or if the hardware is supported and the
+ * scratch area needs to be setup. In other instances it returns
+ * without reserving anything. The memory in case of dump being
+ * active is freed when the dump is collected (by userland tools).
+ */
+static void __init phyp_dump_reserve_mem(void)
+{
+       unsigned long base, size;
+       unsigned long variable_reserve_size;
+
+       if (!phyp_dump_info->phyp_dump_configured) {
+               printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
+               return;
+       }
+
+       if (!phyp_dump_info->phyp_dump_at_boot) {
+               printk(KERN_INFO "Phyp-dump disabled at boot time\n");
+               return;
+       }
+
+       variable_reserve_size = phyp_dump_calculate_reserve_size();
+
+       if (phyp_dump_info->phyp_dump_is_active) {
+               /* Reserve *everything* above RMR.Area freed by userland tools*/
+               base = variable_reserve_size;
+               size = lmb_end_of_DRAM() - base;
+
+               /* XXX crashed_ram_end is wrong, since it may be beyond
+                * the memory_limit, it will need to be adjusted. */
+               lmb_reserve(base, size);
+
+               phyp_dump_info->init_reserve_start = base;
+               phyp_dump_info->init_reserve_size = size;
+       } else {
+               size = phyp_dump_info->cpu_state_size +
+                       phyp_dump_info->hpte_region_size +
+                       variable_reserve_size;
+               base = lmb_end_of_DRAM() - size;
+               lmb_reserve(base, size);
+               phyp_dump_info->init_reserve_start = base;
+               phyp_dump_info->init_reserve_size = size;
+       }
+}
+#else
+static inline void __init phyp_dump_reserve_mem(void) {}
+#endif /* CONFIG_PHYP_DUMP  && CONFIG_PPC_RTAS */
+
+
 void __init early_init_devtree(void *params)
 {
+       phys_addr_t limit;
+
        DBG(" -> early_init_devtree(%p)\n", params);
 
        /* Setup flat device-tree pointer */
@@ -1052,6 +1172,11 @@ void __init early_init_devtree(void *params)
        of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
 #endif
 
+#ifdef CONFIG_PHYP_DUMP
+       /* scan tree to see if dump occured during last boot */
+       of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
+#endif
+
        /* Retrieve various informations from the /chosen node of the
         * device-tree, including the platform type, initrd location and
         * size, TCE reserve, and more ...
@@ -1069,14 +1194,31 @@ void __init early_init_devtree(void *params)
 
        /* Reserve LMB regions used by kernel, initrd, dt, etc... */
        lmb_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
+       /* If relocatable, reserve first 32k for interrupt vectors etc. */
+       if (PHYSICAL_START > MEMORY_START)
+               lmb_reserve(MEMORY_START, 0x8000);
        reserve_kdump_trampoline();
        reserve_crashkernel();
        early_reserve_mem();
+       phyp_dump_reserve_mem();
+
+       limit = memory_limit;
+       if (! limit) {
+               phys_addr_t memsize;
+
+               /* Ensure that total memory size is page-aligned, because
+                * otherwise mark_bootmem() gets upset. */
+               lmb_analyze();
+               memsize = lmb_phys_mem_size();
+               if ((memsize & PAGE_MASK) != memsize)
+                       limit = memsize & PAGE_MASK;
+       }
+       lmb_enforce_memory_limit(limit);
 
-       lmb_enforce_memory_limit(memory_limit);
        lmb_analyze();
+       lmb_dump_all();
 
-       DBG("Phys. mem: %lx\n", lmb_phys_mem_size());
+       DBG("Phys. mem: %llx\n", lmb_phys_mem_size());
 
        /* We may need to relocate the flat tree, do it now.
         * FIXME .. and the initrd too? */
@@ -1144,6 +1286,37 @@ struct device_node *of_find_node_by_phandle(phandle handle)
 EXPORT_SYMBOL(of_find_node_by_phandle);
 
 /**
+ *     of_find_next_cache_node - Find a node's subsidiary cache
+ *     @np:    node of type "cpu" or "cache"
+ *
+ *     Returns a node pointer with refcount incremented, use
+ *     of_node_put() on it when done.  Caller should hold a reference
+ *     to np.
+ */
+struct device_node *of_find_next_cache_node(struct device_node *np)
+{
+       struct device_node *child;
+       const phandle *handle;
+
+       handle = of_get_property(np, "l2-cache", NULL);
+       if (!handle)
+               handle = of_get_property(np, "next-level-cache", NULL);
+
+       if (handle)
+               return of_find_node_by_phandle(*handle);
+
+       /* OF on pmac has nodes instead of properties named "l2-cache"
+        * beneath CPU nodes.
+        */
+       if (!strcmp(np->type, "cpu"))
+               for_each_child_of_node(np, child)
+                       if (!strcmp(child->type, "cache"))
+                               return child;
+
+       return NULL;
+}
+
+/**
  *     of_find_all_nodes - Get next node in global list
  *     @prev:  Previous node or NULL to start iteration
  *             of_node_put() will be called on it
@@ -1244,12 +1417,14 @@ EXPORT_SYMBOL(of_node_put);
  */
 void of_attach_node(struct device_node *np)
 {
-       write_lock(&devtree_lock);
+       unsigned long flags;
+
+       write_lock_irqsave(&devtree_lock, flags);
        np->sibling = np->parent->child;
        np->allnext = allnodes;
        np->parent->child = np;
        allnodes = np;
-       write_unlock(&devtree_lock);
+       write_unlock_irqrestore(&devtree_lock, flags);
 }
 
 /*
@@ -1260,8 +1435,9 @@ void of_attach_node(struct device_node *np)
 void of_detach_node(struct device_node *np)
 {
        struct device_node *parent;
+       unsigned long flags;
 
-       write_lock(&devtree_lock);
+       write_lock_irqsave(&devtree_lock, flags);
 
        parent = np->parent;
        if (!parent)
@@ -1292,7 +1468,7 @@ void of_detach_node(struct device_node *np)
        of_node_set_flag(np, OF_DETACHED);
 
 out_unlock:
-       write_unlock(&devtree_lock);
+       write_unlock_irqrestore(&devtree_lock, flags);
 }
 
 #ifdef CONFIG_PPC_PSERIES
@@ -1373,20 +1549,21 @@ __initcall(prom_reconfig_setup);
 int prom_add_property(struct device_node* np, struct property* prop)
 {
        struct property **next;
+       unsigned long flags;
 
        prop->next = NULL;      
-       write_lock(&devtree_lock);
+       write_lock_irqsave(&devtree_lock, flags);
        next = &np->properties;
        while (*next) {
                if (strcmp(prop->name, (*next)->name) == 0) {
                        /* duplicate ! don't insert it */
-                       write_unlock(&devtree_lock);
+                       write_unlock_irqrestore(&devtree_lock, flags);
                        return -1;
                }
                next = &(*next)->next;
        }
        *next = prop;
-       write_unlock(&devtree_lock);
+       write_unlock_irqrestore(&devtree_lock, flags);
 
 #ifdef CONFIG_PROC_DEVICETREE
        /* try to add to proc as well if it was initialized */
@@ -1406,9 +1583,10 @@ int prom_add_property(struct device_node* np, struct property* prop)
 int prom_remove_property(struct device_node *np, struct property *prop)
 {
        struct property **next;
+       unsigned long flags;
        int found = 0;
 
-       write_lock(&devtree_lock);
+       write_lock_irqsave(&devtree_lock, flags);
        next = &np->properties;
        while (*next) {
                if (*next == prop) {
@@ -1421,7 +1599,7 @@ int prom_remove_property(struct device_node *np, struct property *prop)
                }
                next = &(*next)->next;
        }
-       write_unlock(&devtree_lock);
+       write_unlock_irqrestore(&devtree_lock, flags);
 
        if (!found)
                return -ENODEV;
@@ -1447,9 +1625,10 @@ int prom_update_property(struct device_node *np,
                         struct property *oldprop)
 {
        struct property **next;
+       unsigned long flags;
        int found = 0;
 
-       write_lock(&devtree_lock);
+       write_lock_irqsave(&devtree_lock, flags);
        next = &np->properties;
        while (*next) {
                if (*next == oldprop) {
@@ -1463,7 +1642,7 @@ int prom_update_property(struct device_node *np,
                }
                next = &(*next)->next;
        }
-       write_unlock(&devtree_lock);
+       write_unlock_irqrestore(&devtree_lock, flags);
 
        if (!found)
                return -ENODEV;