[coreboot-gerrit] New patch to review for coreboot: a880fe0 Revert "coreboot table: use memrange library"

Aaron Durbin (adurbin@google.com) gerrit at coreboot.org
Tue Mar 26 18:07:32 CET 2013


Aaron Durbin (adurbin at google.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/2915

-gerrit

commit a880fe0408697f535beef47ad23b8bb4238362c5
Author: Aaron Durbin <adurbin at google.com>
Date:   Tue Mar 26 18:07:32 2013 +0100

    Revert "coreboot table: use memrange library"
    
    This reverts commit 56075eaefcd7ef51464206166b24a0a47a59147f
    
    Change-Id: I8a37ce1f5ce36e4a120941ec264140abc9447ff5
---
 src/lib/coreboot_table.c | 271 ++++++++++++++++++++++++++++++++++++++---------
 1 file changed, 218 insertions(+), 53 deletions(-)

diff --git a/src/lib/coreboot_table.c b/src/lib/coreboot_table.c
index 840470b..84eb5a6 100644
--- a/src/lib/coreboot_table.c
+++ b/src/lib/coreboot_table.c
@@ -29,7 +29,6 @@
 #include <stdlib.h>
 #include <cbfs.h>
 #include <cbmem.h>
-#include <memrange.h>
 #if CONFIG_USE_OPTION_TABLE
 #include <option_table.h>
 #endif
@@ -357,7 +356,54 @@ static struct lb_forward *lb_forward(struct lb_header *header, struct lb_header
 	return forward;
 }
 
-static unsigned long lb_table_fini(struct lb_header *head)
+static void lb_memory_range(struct lb_memory *mem,
+	uint32_t type, uint64_t start, uint64_t size)
+{
+	int entries;
+	entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
+	mem->map[entries].start = pack_lb64(start);
+	mem->map[entries].size = pack_lb64(size);
+	mem->map[entries].type = type;
+	mem->size += sizeof(mem->map[0]);
+}
+
+static void lb_reserve_table_memory(struct lb_header *head)
+{
+/* Dynamic cbmem has already reserved the memory where the coreboot tables
+ * reside. Therefore, there is nothing to fix up. */
+#if !CONFIG_DYNAMIC_CBMEM
+	struct lb_record *last_rec;
+	struct lb_memory *mem;
+	uint64_t start;
+	uint64_t end;
+	int i, entries;
+
+	last_rec = lb_last_record(head);
+	mem = get_lb_mem();
+	start = (unsigned long)head;
+	end = (unsigned long)last_rec;
+	entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
+	/* Resize the right two memory areas so this table is in
+	 * a reserved area of memory.  Everything has been carefully
+	 * setup so that is all we need to do.
+	 */
+	for(i = 0; i < entries; i++ ) {
+		uint64_t map_start = unpack_lb64(mem->map[i].start);
+		uint64_t map_end = map_start + unpack_lb64(mem->map[i].size);
+		/* Does this area need to be expanded? */
+		if (map_end == start) {
+			mem->map[i].size = pack_lb64(end - map_start);
+		}
+		/* Does this area need to be contracted? */
+		else if (map_start == start) {
+			mem->map[i].start = pack_lb64(end);
+			mem->map[i].size = pack_lb64(map_end - end);
+		}
+	}
+#endif
+}
+
+static unsigned long lb_table_fini(struct lb_header *head, int fixup)
 {
 	struct lb_record *rec, *first_rec;
 	rec = lb_last_record(head);
@@ -365,6 +411,9 @@ static unsigned long lb_table_fini(struct lb_header *head)
 		head->table_bytes += rec->size;
 	}
 
+	if (fixup)
+		lb_reserve_table_memory(head);
+
 	first_rec = lb_first_record(head);
 	head->table_checksum = compute_ip_checksum(first_rec, head->table_bytes);
 	head->header_checksum = 0;
@@ -375,53 +424,128 @@ static unsigned long lb_table_fini(struct lb_header *head)
 	return (unsigned long)rec + rec->size;
 }
 
-/* Routines to extract part so the coreboot table or
- * information from the coreboot table after we have written it.
- * Currently get_lb_mem relies on a global we can change the
- * implementaiton.
- */
-static struct lb_memory *mem_ranges = NULL;
-
-struct lb_memory *get_lb_mem(void)
+static void lb_cleanup_memory_ranges(struct lb_memory *mem)
 {
-	return mem_ranges;
-}
+	int entries;
+	int i, j;
+	entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
+
+	/* Sort the lb memory ranges */
+	for(i = 0; i < entries; i++) {
+		uint64_t entry_start = unpack_lb64(mem->map[i].start);
+		for(j = i + 1; j < entries; j++) {
+			uint64_t temp_start = unpack_lb64(mem->map[j].start);
+			if (temp_start < entry_start) {
+				struct lb_memory_range tmp;
+				tmp = mem->map[i];
+				mem->map[i] = mem->map[j];
+				mem->map[j] = tmp;
+			}
+		}
+	}
 
-/* This structure keeps track of the coreboot table memory ranges. */
-static struct memory_ranges lb_ranges;
+	/* Merge adjacent entries */
+	for(i = 0; (i + 1) < entries; i++) {
+		uint64_t start, end, nstart, nend;
+		if (mem->map[i].type != mem->map[i + 1].type) {
+			continue;
+		}
+		start  = unpack_lb64(mem->map[i].start);
+		end    = start + unpack_lb64(mem->map[i].size);
+		nstart = unpack_lb64(mem->map[i + 1].start);
+		nend   = nstart + unpack_lb64(mem->map[i + 1].size);
+		if ((start <= nstart) && (end >= nstart)) {
+			if (start > nstart) {
+				start = nstart;
+			}
+			if (end < nend) {
+				end = nend;
+			}
+			/* Record the new region size */
+			mem->map[i].start = pack_lb64(start);
+			mem->map[i].size  = pack_lb64(end - start);
+
+			/* Delete the entry I have merged with */
+			memmove(&mem->map[i + 1], &mem->map[i + 2],
+				((entries - i - 2) * sizeof(mem->map[0])));
+			mem->size -= sizeof(mem->map[0]);
+			entries -= 1;
+			/* See if I can merge with the next entry as well */
+			i -= 1;
+		}
+	}
+}
 
-static struct lb_memory *build_lb_mem(struct lb_header *head)
+static void lb_remove_memory_range(struct lb_memory *mem,
+	uint64_t start, uint64_t size)
 {
-	struct lb_memory *mem;
-
-	/* Record where the lb memory ranges will live */
-	mem = lb_memory(head);
-	mem_ranges = mem;
+	uint64_t end;
+	int entries;
+	int i;
 
-	/* Fill the memory map out. The order of operations is important in
-	 * that each overlapping range will take over the next. Therefore,
-	 * add cacheable resources as RAM then add the reserved resources. */
-	memory_ranges_init(&lb_ranges, IORESOURCE_CACHEABLE,
-	                   IORESOURCE_CACHEABLE, LB_MEM_RAM);
-	memory_ranges_add_resources(&lb_ranges, IORESOURCE_RESERVE,
-	                            IORESOURCE_RESERVE, LB_MEM_RESERVED);
+	end = start + size;
+	entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
+
+	/* Remove a reserved area from the memory map */
+	for(i = 0; i < entries; i++) {
+		uint64_t map_start = unpack_lb64(mem->map[i].start);
+		uint64_t map_end   = map_start + unpack_lb64(mem->map[i].size);
+		if ((start <= map_start) && (end >= map_end)) {
+			/* Remove the completely covered range */
+			memmove(&mem->map[i], &mem->map[i + 1],
+				((entries - i - 1) * sizeof(mem->map[0])));
+			mem->size -= sizeof(mem->map[0]);
+			entries -= 1;
+			/* Since the index will disappear revisit what will appear here */
+			i -= 1;
+		}
+		else if ((start > map_start) && (end < map_end)) {
+			/* Split the memory range */
+			memmove(&mem->map[i + 1], &mem->map[i],
+				((entries - i) * sizeof(mem->map[0])));
+			mem->size += sizeof(mem->map[0]);
+			entries += 1;
+			/* Update the first map entry */
+			mem->map[i].size = pack_lb64(start - map_start);
+			/* Update the second map entry */
+			mem->map[i + 1].start = pack_lb64(end);
+			mem->map[i + 1].size  = pack_lb64(map_end - end);
+			/* Don't bother with this map entry again */
+			i += 1;
+		}
+		else if ((start <= map_start) && (end > map_start)) {
+			/* Shrink the start of the memory range */
+			mem->map[i].start = pack_lb64(end);
+			mem->map[i].size  = pack_lb64(map_end - end);
+		}
+		else if ((start < map_end) && (start > map_start)) {
+			/* Shrink the end of the memory range */
+			mem->map[i].size = pack_lb64(start - map_start);
+		}
+	}
+}
 
-	return mem;
+void lb_add_memory_range(struct lb_memory *mem,
+	uint32_t type, uint64_t start, uint64_t size)
+{
+	lb_remove_memory_range(mem, start, size);
+	lb_memory_range(mem, type, start, size);
+	lb_cleanup_memory_ranges(mem);
 }
 
-static void commit_lb_memory(struct lb_memory *mem)
+static void lb_dump_memory_ranges(struct lb_memory *mem)
 {
-	struct memory_range *r;
-	struct lb_memory_range *lb_r;
+	int entries;
 	int i;
+	entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
 
-	lb_r = &mem->map[0];
-	i = 0;
-
-	memory_ranges_each_entry(r, &lb_ranges) {
+	printk(BIOS_DEBUG, "coreboot memory table:\n");
+	for(i = 0; i < entries; i++) {
+		uint64_t entry_start = unpack_lb64(mem->map[i].start);
+		uint64_t entry_size = unpack_lb64(mem->map[i].size);
 		const char *entry_type;
 
-		switch (memory_range_tag(r)) {
+		switch (mem->map[i].type) {
 		case LB_MEM_RAM: entry_type="RAM"; break;
 		case LB_MEM_RESERVED: entry_type="RESERVED"; break;
 		case LB_MEM_ACPI: entry_type="ACPI"; break;
@@ -433,25 +557,58 @@ static void commit_lb_memory(struct lb_memory *mem)
 		}
 
 		printk(BIOS_DEBUG, "%2d. %016llx-%016llx: %s\n",
-			i, memory_range_base(r), memory_range_end(r)-1,
-			entry_type);
-
-		lb_r->start = pack_lb64(memory_range_base(r));
-		lb_r->size = pack_lb64(memory_range_size(r));
-		lb_r->type = memory_range_tag(r);
+			i, entry_start, entry_start+entry_size-1, entry_type);
 
-		i++;
-		lb_r++;
-		mem->size += sizeof(struct lb_memory_range);
 	}
 }
 
-void lb_add_memory_range(struct lb_memory *mem,
-	uint32_t type, uint64_t start, uint64_t size)
+/* Routines to extract part so the coreboot table or
+ * information from the coreboot table after we have written it.
+ * Currently get_lb_mem relies on a global we can change the
+ * implementaiton.
+ */
+static struct lb_memory *mem_ranges = NULL;
+
+struct lb_memory *get_lb_mem(void)
 {
-	insert_memory_range(&lb_ranges, start, size, type);
+	return mem_ranges;
 }
 
+static void build_lb_mem_range(void *gp, struct device *dev, struct resource *res)
+{
+	struct lb_memory *mem = gp;
+	lb_memory_range(mem, LB_MEM_RAM, res->base, res->size);
+}
+
+static struct lb_memory *build_lb_mem(struct lb_header *head)
+{
+	struct lb_memory *mem;
+
+	/* Record where the lb memory ranges will live */
+	mem = lb_memory(head);
+	mem_ranges = mem;
+
+	/* Build the raw table of memory */
+	search_global_resources(
+		IORESOURCE_MEM | IORESOURCE_CACHEABLE, IORESOURCE_MEM | IORESOURCE_CACHEABLE,
+		build_lb_mem_range, mem);
+	lb_cleanup_memory_ranges(mem);
+	return mem;
+}
+
+static void lb_add_rsvd_range(void *gp, struct device *dev, struct resource *res)
+{
+	struct lb_memory *mem = gp;
+	lb_add_memory_range(mem, LB_MEM_RESERVED, res->base, res->size);
+}
+
+static void add_lb_reserved(struct lb_memory *mem)
+{
+	/* Add reserved ranges */
+	search_global_resources(
+		IORESOURCE_MEM | IORESOURCE_RESERVE, IORESOURCE_MEM | IORESOURCE_RESERVE,
+		lb_add_rsvd_range, mem);
+}
 
 unsigned long write_coreboot_table(
 	unsigned long low_table_start, unsigned long low_table_end,
@@ -466,7 +623,7 @@ unsigned long write_coreboot_table(
 		head = lb_table_init(low_table_end);
 		lb_forward(head, (struct lb_header*)rom_table_end);
 
-		low_table_end = (unsigned long) lb_table_fini(head);
+		low_table_end = (unsigned long) lb_table_fini(head, 0);
 		printk(BIOS_DEBUG, "Table forward entry ends at 0x%08lx.\n",
 			low_table_end);
 		low_table_end = ALIGN(low_table_end, 4096);
@@ -525,9 +682,17 @@ unsigned long write_coreboot_table(
 		high_tables_base, high_tables_size);
 #endif /* CONFIG_DYNAMIC_CBMEM */
 
-	/* No other memory areas can be added after the memory table has been
-	 * committed as the entries won't show up in the serialize mem table. */
-	commit_lb_memory(mem);
+	/* Add reserved regions */
+	add_lb_reserved(mem);
+
+	lb_dump_memory_ranges(mem);
+
+	/* Note:
+	 * I assume that there is always memory at immediately after
+	 * the low_table_end.  This means that after I setup the coreboot table.
+	 * I can trivially fixup the reserved memory ranges to hold the correct
+	 * size of the coreboot table.
+	 */
 
 	/* Record our motherboard */
 	lb_mainboard(head);
@@ -556,5 +721,5 @@ unsigned long write_coreboot_table(
 	add_cbmem_pointers(head);
 
 	/* Remember where my valid memory ranges are */
-	return lb_table_fini(head);
+	return lb_table_fini(head, 1);
 }



More information about the coreboot-gerrit mailing list