Re: Booting the Linux/ppc64 kernel without Open Firmware HOWTO (v3)

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Tue, 2005-05-24 at 14:34 +1000, Benjamin Herrenschmidt wrote:
> Hi !
> 
> Here's revision 3 of the spec for the booting of linux/ppc64 with a
> flattened device-tree. The novelty is that I added a new more compact
> format. A followup mail will have the kernel patches to add support to
> this new format, I'll submit them upstream for after 2.6.12 I think.
> 
> David and I are still working on sample code & tools. We have a
> prototype of a device-tree "compiler" that can build the flattened blob
> from a textual representation. We'll release that soon, hopefully this
> week.

And here is the patch:

Index: linux-work/arch/ppc64/kernel/prom_init.c
===================================================================
--- linux-work.orig/arch/ppc64/kernel/prom_init.c	2005-05-23 18:15:28.000000000 +1000
+++ linux-work/arch/ppc64/kernel/prom_init.c	2005-05-24 14:10:00.000000000 +1000
@@ -1514,7 +1514,14 @@
 	return 0;
 }
 
-static void __init scan_dt_build_strings(phandle node, unsigned long *mem_start,
+/*
+ * The Open Firmware 1275 specification states properties must be 31 bytes or
+ * less, however not all firmwares obey this. Make it 64 bytes to be safe.
+ */
+#define MAX_PROPERTY_NAME 64
+
+static void __init scan_dt_build_strings(phandle node,
+					 unsigned long *mem_start,
 					 unsigned long *mem_end)
 {
 	unsigned long offset = reloc_offset();
@@ -1527,14 +1534,19 @@
 	/* get and store all property names */
 	prev_name = RELOC("");
 	for (;;) {
-		
-		/* 32 is max len of name including nul. */
-		namep = make_room(mem_start, mem_end, 32, 1);
+		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
 		if (call_prom("nextprop", 3, 1, node, prev_name, namep) <= 0) {
 			/* No more nodes: unwind alloc */
 			*mem_start = (unsigned long)namep;
 			break;
 		}
+
+		/* skip "name" */
+		if (strcmp(namep, RELOC("name")) == 0) {
+			*mem_start = (unsigned long)namep;
+			prev_name = RELOC("name");
+			continue;
+		}
 		soff = dt_find_string(namep);
 		if (soff != 0) {
 			*mem_start = (unsigned long)namep;
@@ -1555,12 +1567,6 @@
 	}
 }
 
-/*
- * The Open Firmware 1275 specification states properties must be 31 bytes or
- * less, however not all firmwares obey this. Make it 64 bytes to be safe.
- */
-#define MAX_PROPERTY_NAME 64
-
 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
 					unsigned long *mem_end)
 {
@@ -1570,10 +1576,8 @@
 	unsigned long soff;
 	unsigned char *valp;
 	unsigned long offset = reloc_offset();
-	char pname[MAX_PROPERTY_NAME];
-	char *path;
-
-	path = RELOC(prom_scratch);
+	static char pname[MAX_PROPERTY_NAME] __initdata;
+	char *path, *p, *lp;
 
 	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
 
@@ -1588,10 +1592,20 @@
 			call_prom("package-to-path", 3, 1, node, namep, l);
 		}
 		namep[l] = '\0';
-		*mem_start = _ALIGN(((unsigned long) namep) + strlen(namep) + 1, 4);
+		/* now try to find the unit name in that mess */
+		p = namep;
+		lp = NULL;
+		for (lp = NULL; *p; p++)
+			if (*p == '/')
+				lp = p + 1;
+		if (lp != NULL)
+			memmove(namep, lp, strlen(lp) + 1);
+		*mem_start = _ALIGN(((unsigned long) namep) +
+				    strlen(namep) + 1, 4);
 	}
 
 	/* get it again for debugging */
+	path = RELOC(prom_scratch);
 	memset(path, 0, PROM_SCRATCH_SIZE);
 	call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
 
@@ -1599,20 +1613,24 @@
 	prev_name = RELOC("");
 	sstart = (char *)RELOC(dt_string_start);
 	for (;;) {
-		if (call_prom("nextprop", 3, 1, node, prev_name, pname) <= 0)
+		if (call_prom("nextprop", 3, 1, node, prev_name,
+			      RELOC(pname)) <= 0)
 			break;
-
+		if (strcmp(RELOC(pname), RELOC("name")) == 0) {
+			prev_name = RELOC("name");
+			continue;
+		}
 		/* find string offset */
-		soff = dt_find_string(pname);
+		soff = dt_find_string(RELOC(pname));
 		if (soff == 0) {
-			prom_printf("WARNING: Can't find string index for <%s>, node %s\n",
-				    pname, path);
+			prom_printf("WARNING: Can't find string index "
+				    "for <%s>, node %s\n", RELOC(pname), path);
 			break;
 		}
 		prev_name = sstart + soff;
 
 		/* get length */
-		l = call_prom("getproplen", 2, 1, node, pname);
+		l = call_prom("getproplen", 2, 1, node, RELOC(pname));
 
 		/* sanity checks */
 		if (l < 0)
@@ -1621,7 +1639,7 @@
 			prom_printf("WARNING: ignoring large property ");
 			/* It seems OF doesn't null-terminate the path :-( */
 			prom_printf("[%s] ", path);
-			prom_printf("%s length 0x%x\n", pname, l);
+			prom_printf("%s length 0x%x\n", RELOC(pname), l);
 			continue;
 		}
 
@@ -1633,15 +1651,15 @@
 		/* push property content */
 		align = (l >= 8) ? 8 : 4;
 		valp = make_room(mem_start, mem_end, l, align);
-		call_prom("getprop", 4, 1, node, pname, valp, l);
+		call_prom("getprop", 4, 1, node, RELOC(pname), valp, l);
 		*mem_start = _ALIGN(*mem_start, 4);
 	}
 
 	/* Add a "linux,phandle" property. */
 	soff = dt_find_string(RELOC("linux,phandle"));
 	if (soff == 0)
-		prom_printf("WARNING: Can't find string index for <linux-phandle>"
-			    " node %s\n", path);
+		prom_printf("WARNING: Can't find string index for"
+			    " <linux-phandle> node %s\n", path);
 	else {
 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
 		dt_push_token(4, mem_start, mem_end);
@@ -1691,7 +1709,8 @@
 
 	/* Build header and make room for mem rsv map */ 
 	mem_start = _ALIGN(mem_start, 4);
-	hdr = make_room(&mem_start, &mem_end, sizeof(struct boot_param_header), 4);
+	hdr = make_room(&mem_start, &mem_end,
+			sizeof(struct boot_param_header), 4);
 	RELOC(dt_header_start) = (unsigned long)hdr;
 	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
 
@@ -1704,11 +1723,11 @@
 	namep = make_room(&mem_start, &mem_end, 16, 1);
 	strcpy(namep, RELOC("linux,phandle"));
 	mem_start = (unsigned long)namep + strlen(namep) + 1;
-	RELOC(dt_string_end) = mem_start;
 
 	/* Build string array */
 	prom_printf("Building dt strings...\n"); 
 	scan_dt_build_strings(root, &mem_start, &mem_end);
+	RELOC(dt_string_end) = mem_start;
 
 	/* Build structure */
 	mem_start = PAGE_ALIGN(mem_start);
@@ -1723,9 +1742,11 @@
 	hdr->totalsize = RELOC(dt_struct_end) - RELOC(dt_header_start);
 	hdr->off_dt_struct = RELOC(dt_struct_start) - RELOC(dt_header_start);
 	hdr->off_dt_strings = RELOC(dt_string_start) - RELOC(dt_header_start);
+	hdr->dt_strings_size = RELOC(dt_string_end) - RELOC(dt_string_start);
 	hdr->off_mem_rsvmap = ((unsigned long)rsvmap) - RELOC(dt_header_start);
 	hdr->version = OF_DT_VERSION;
-	hdr->last_comp_version = 1;
+	/* Version 16 is not backward compatible */
+	hdr->last_comp_version = 0x10;
 
 	/* Reserve the whole thing and copy the reserve map in, we
 	 * also bump mem_reserve_cnt to cause further reservations to
Index: linux-work/arch/ppc64/kernel/setup.c
===================================================================
--- linux-work.orig/arch/ppc64/kernel/setup.c	2005-05-23 18:15:28.000000000 +1000
+++ linux-work/arch/ppc64/kernel/setup.c	2005-05-23 18:16:09.000000000 +1000
@@ -10,7 +10,7 @@
  *      2 of the License, or (at your option) any later version.
  */
 
-#undef DEBUG
+#define DEBUG
 
 #include <linux/config.h>
 #include <linux/module.h>
Index: linux-work/arch/ppc64/kernel/prom.c
===================================================================
--- linux-work.orig/arch/ppc64/kernel/prom.c	2005-05-23 18:15:28.000000000 +1000
+++ linux-work/arch/ppc64/kernel/prom.c	2005-05-24 14:12:35.000000000 +1000
@@ -15,7 +15,7 @@
  *      2 of the License, or (at your option) any later version.
  */
 
-#undef DEBUG
+#define DEBUG
 
 #include <stdarg.h>
 #include <linux/config.h>
@@ -635,20 +635,23 @@
  * unflatten the tree
  */
 static int __init scan_flat_dt(int (*it)(unsigned long node,
-					 const char *full_path, void *data),
+					 const char *uname, int depth, void *data),
 			       void *data)
 {
 	unsigned long p = ((unsigned long)initial_boot_params) +
 		initial_boot_params->off_dt_struct;
 	int rc = 0;
+	int depth = -1;
 
 	do {
 		u32 tag = *((u32 *)p);
 		char *pathp;
 		
 		p += 4;
-		if (tag == OF_DT_END_NODE)
+		if (tag == OF_DT_END_NODE) {
+			depth --;
 			continue;
+		}
 		if (tag == OF_DT_END)
 			break;
 		if (tag == OF_DT_PROP) {
@@ -664,9 +667,18 @@
 			       " device tree !\n", tag);
 			return -EINVAL;
 		}
+		depth++;
 		pathp = (char *)p;
 		p = _ALIGN(p + strlen(pathp) + 1, 4);
-		rc = it(p, pathp, data);
+		if ((*pathp) == '/') {
+			char *lp, *np;
+			for (lp = NULL, np = pathp; *np; np++)
+				if ((*np) == '/')
+					lp = np+1;
+			if (lp != NULL)
+				pathp = lp;
+		}
+		rc = it(p, pathp, depth, data);
 		if (rc != 0)
 			break;		
 	} while(1);
@@ -727,13 +739,16 @@
 static unsigned long __init unflatten_dt_node(unsigned long mem,
 					      unsigned long *p,
 					      struct device_node *dad,
-					      struct device_node ***allnextpp)
+					      struct device_node ***allnextpp,
+					      unsigned long fpsize)
 {
 	struct device_node *np;
 	struct property *pp, **prev_pp = NULL;
 	char *pathp;
 	u32 tag;
-	unsigned int l;
+	unsigned int l, allocl;
+	int has_name = 0;
+	int new_format = 0;
 
 	tag = *((u32 *)(*p));
 	if (tag != OF_DT_BEGIN_NODE) {
@@ -742,21 +757,56 @@
 	}
 	*p += 4;
 	pathp = (char *)*p;
-	l = strlen(pathp) + 1;
+	l = allocl = strlen(pathp) + 1;
 	*p = _ALIGN(*p + l, 4);
 
-	np = unflatten_dt_alloc(&mem, sizeof(struct device_node) + l,
+	/* version 0x10 has a more compact unit name here instead of the full
+	 * path. we accumulate the full path size using "fpsize", we'll rebuild
+	 * it later. We detect this because the first character of the name is
+	 * not '/'.
+	 */
+	if ((*pathp) != '/') {
+		new_format = 1;
+		if (fpsize == 0) {
+			/* root node: special case. fpsize accounts for path
+			 * plus terminating zero. root node only has '/', so
+			 * fpsize should be 2, but we want to avoid the first
+			 * level nodes to have two '/' so we use fpsize 1 here
+			 */
+			fpsize = 1;
+			allocl = 2;
+		} else {
+			/* account for '/' and path size minus terminal 0
+			 * already in 'l'
+			 */
+			fpsize += l;
+			allocl = fpsize;
+		}
+	}
+
+
+	np = unflatten_dt_alloc(&mem, sizeof(struct device_node) + allocl,
 				__alignof__(struct device_node));
 	if (allnextpp) {
 		memset(np, 0, sizeof(*np));
 		np->full_name = ((char*)np) + sizeof(struct device_node);
-		memcpy(np->full_name, pathp, l);
+		if (new_format) {
+			char *p = np->full_name;
+			/* rebuild full path for new format */
+			if (dad && dad->parent) {
+				strcpy(p, dad->full_name);
+				p += strlen(p);
+			}
+			*(p++) = '/';
+			memcpy(p, pathp, l);
+		} else
+			memcpy(np->full_name, pathp, l);
 		prev_pp = &np->properties;
 		**allnextpp = np;
 		*allnextpp = &np->allnext;
 		if (dad != NULL) {
 			np->parent = dad;
-			/* we temporarily use the `next' field as `last_child'. */
+			/* we temporarily use the next field as `last_child'*/
 			if (dad->next == 0)
 				dad->child = np;
 			else
@@ -782,6 +832,8 @@
 			printk("Can't find property name in list !\n");
 			break;
 		}
+		if (strcmp(pname, "name") == 0)
+			has_name = 1;
 		l = strlen(pname) + 1;
 		pp = unflatten_dt_alloc(&mem, sizeof(struct property),
 					__alignof__(struct property));
@@ -801,6 +853,29 @@
 		}
 		*p = _ALIGN((*p) + sz, 4);
 	}
+	/* with version 0x10 we may not have the name property, recreate
+	 * it here from the unit name if absent
+	 */
+	if (!has_name) {
+		char *p = pathp;
+		int sz;
+
+		DBG("no name, building from %s\n", p);
+		while (*p && (*p) != '@')
+			p++;
+		sz = (p - pathp) + 1;
+		pp = unflatten_dt_alloc(&mem, sizeof(struct property) + sz,
+					__alignof__(struct property));
+		if (allnextpp) {
+			pp->name = "name";
+			pp->length = sz;
+			pp->value = (unsigned char *)(pp + 1);
+			*prev_pp = pp;
+			prev_pp = &pp->next;
+			memcpy(pp->value, pathp, sz - 1);
+			((char *)pp->value)[sz - 1] = 0;
+		}
+	}
 	if (allnextpp) {
 		*prev_pp = NULL;
 		np->name = get_property(np, "name", NULL);
@@ -812,7 +887,7 @@
 			np->type = "<NULL>";
 	}
 	while (tag == OF_DT_BEGIN_NODE) {
-		mem = unflatten_dt_node(mem, p, np, allnextpp);
+		mem = unflatten_dt_node(mem, p, np, allnextpp, fpsize);
 		tag = *((u32 *)(*p));
 	}
 	if (tag != OF_DT_END_NODE) {
@@ -842,7 +917,7 @@
 	/* First pass, scan for size */
 	start = ((unsigned long)initial_boot_params) +
 		initial_boot_params->off_dt_struct;
-	size = unflatten_dt_node(0, &start, NULL, NULL);
+	size = unflatten_dt_node(0, &start, NULL, NULL, 0);
 
 	DBG("  size is %lx, allocating...\n", size);
 
@@ -854,7 +929,7 @@
 	/* Second pass, do actual unflattening */
 	start = ((unsigned long)initial_boot_params) +
 		initial_boot_params->off_dt_struct;
-	unflatten_dt_node(mem, &start, NULL, &allnextp);
+	unflatten_dt_node(mem, &start, NULL, &allnextp, 0);
 	if (*((u32 *)start) != OF_DT_END)
 		printk(KERN_WARNING "Weird tag at end of tree: %x\n", *((u32 *)start));
 	*allnextp = NULL;
@@ -880,7 +955,7 @@
 
 
 static int __init early_init_dt_scan_cpus(unsigned long node,
-					  const char *full_path, void *data)
+					  const char *uname, int depth, void *data)
 {
 	char *type = get_flat_dt_prop(node, "device_type", NULL);
 	u32 *prop;
@@ -933,13 +1008,15 @@
 }
 
 static int __init early_init_dt_scan_chosen(unsigned long node,
-					    const char *full_path, void *data)
+					    const char *uname, int depth, void *data)
 {
 	u32 *prop;
 	u64 *prop64;
 	extern unsigned long memory_limit, tce_alloc_start, tce_alloc_end;
 
-	if (strcmp(full_path, "/chosen") != 0)
+	DBG("search \"chosen\", depth: %d, uname: %s\n", depth, uname);
+
+	if (depth != 1 || strcmp(uname, "chosen") != 0)
 		return 0;
 
 	/* get platform type */
@@ -989,18 +1066,20 @@
 }
 
 static int __init early_init_dt_scan_root(unsigned long node,
-					  const char *full_path, void *data)
+					  const char *uname, int depth, void *data)
 {
 	u32 *prop;
 
-	if (strcmp(full_path, "/") != 0)
+	if (depth != 0)
 		return 0;
 
 	prop = (u32 *)get_flat_dt_prop(node, "#size-cells", NULL);
 	dt_root_size_cells = (prop == NULL) ? 1 : *prop;
-		
+	DBG("dt_root_size_cells = %x\n", dt_root_size_cells);
+
 	prop = (u32 *)get_flat_dt_prop(node, "#address-cells", NULL);
 	dt_root_addr_cells = (prop == NULL) ? 2 : *prop;
+	DBG("dt_root_addr_cells = %x\n", dt_root_addr_cells);
 	
 	/* break now */
 	return 1;
@@ -1028,7 +1107,7 @@
 
 
 static int __init early_init_dt_scan_memory(unsigned long node,
-					    const char *full_path, void *data)
+					    const char *uname, int depth, void *data)
 {
 	char *type = get_flat_dt_prop(node, "device_type", NULL);
 	cell_t *reg, *endp;
@@ -1044,7 +1123,9 @@
 
 	endp = reg + (l / sizeof(cell_t));
 
-	DBG("memory scan node %s ...\n", full_path);
+	DBG("memory scan node %s ..., reg size %ld, data: %x %x %x %x, ...\n",
+	    uname, l, reg[0], reg[1], reg[2], reg[3]);
+
 	while ((endp - reg) >= (dt_root_addr_cells + dt_root_size_cells)) {
 		unsigned long base, size;
 
@@ -1455,10 +1536,11 @@
 	struct device_node *np = allnodes;
 
 	read_lock(&devtree_lock);
-	for (; np != 0; np = np->allnext)
+	for (; np != 0; np = np->allnext) {
 		if (np->full_name != 0 && strcasecmp(np->full_name, path) == 0
 		    && of_node_get(np))
 			break;
+	}
 	read_unlock(&devtree_lock);
 	return np;
 }
Index: linux-work/include/asm-ppc64/prom.h
===================================================================
--- linux-work.orig/include/asm-ppc64/prom.h	2005-05-23 18:15:28.000000000 +1000
+++ linux-work/include/asm-ppc64/prom.h	2005-05-23 18:16:09.000000000 +1000
@@ -22,13 +22,14 @@
 #define RELOC(x)        (*PTRRELOC(&(x)))
 
 /* Definitions used by the flattened device tree */
-#define OF_DT_HEADER		0xd00dfeed	/* 4: version, 4: total size */
-#define OF_DT_BEGIN_NODE	0x1		/* Start node: full name */
+#define OF_DT_HEADER		0xd00dfeed	/* marker */
+#define OF_DT_BEGIN_NODE	0x1		/* Start of node, full name */
 #define OF_DT_END_NODE		0x2		/* End node */
-#define OF_DT_PROP		0x3		/* Property: name off, size, content */
+#define OF_DT_PROP		0x3		/* Property: name off, size,
+						 * content */
 #define OF_DT_END		0x9
 
-#define OF_DT_VERSION		1
+#define OF_DT_VERSION		0x10
 
 /*
  * This is what gets passed to the kernel by prom_init or kexec
@@ -54,7 +55,9 @@
 	u32	version;		/* format version */
 	u32	last_comp_version;	/* last compatible version */
 	/* version 2 fields below */
-	u32	boot_cpuid_phys;	/* Which physical CPU id we're booting on */
+	u32	boot_cpuid_phys;	/* Physical CPU id we're booting on */
+	/* version 3 fields below */
+	u32	dt_strings_size;	/* size of the DT strings block */
 };
 
 
@@ -169,17 +172,20 @@
  */
 #define HAVE_ARCH_DEVTREE_FIXUPS
 
-static inline void set_node_proc_entry(struct device_node *dn, struct proc_dir_entry *de)
+static inline void set_node_proc_entry(struct device_node *dn,
+				       struct proc_dir_entry *de)
 {
 	dn->pde = de;
 }
 
-static void inline set_node_name_link(struct device_node *dn, struct proc_dir_entry *de)
+static void inline set_node_name_link(struct device_node *dn,
+				      struct proc_dir_entry *de)
 {
 	dn->name_link = de;
 }
 
-static void inline set_node_addr_link(struct device_node *dn, struct proc_dir_entry *de)
+static void inline set_node_addr_link(struct device_node *dn,
+				      struct proc_dir_entry *de)
 {
 	dn->addr_link = de;
 }