summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--src/lib/coreboot_table.c271
1 files changed, 53 insertions, 218 deletions
diff --git a/src/lib/coreboot_table.c b/src/lib/coreboot_table.c
index 84eb5a65a9..840470b9bc 100644
--- a/src/lib/coreboot_table.c
+++ b/src/lib/coreboot_table.c
@@ -29,6 +29,7 @@
#include <stdlib.h>
#include <cbfs.h>
#include <cbmem.h>
+#include <memrange.h>
#if CONFIG_USE_OPTION_TABLE
#include <option_table.h>
#endif
@@ -356,54 +357,7 @@ static struct lb_forward *lb_forward(struct lb_header *header, struct lb_header
return forward;
}
-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)
+static unsigned long lb_table_fini(struct lb_header *head)
{
struct lb_record *rec, *first_rec;
rec = lb_last_record(head);
@@ -411,9 +365,6 @@ static unsigned long lb_table_fini(struct lb_header *head, int fixup)
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;
@@ -424,128 +375,53 @@ static unsigned long lb_table_fini(struct lb_header *head, int fixup)
return (unsigned long)rec + rec->size;
}
-static void lb_cleanup_memory_ranges(struct lb_memory *mem)
-{
- 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;
- }
- }
- }
+/* 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;
- /* 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;
- }
- }
+struct lb_memory *get_lb_mem(void)
+{
+ return mem_ranges;
}
-static void lb_remove_memory_range(struct lb_memory *mem,
- uint64_t start, uint64_t size)
+/* This structure keeps track of the coreboot table memory ranges. */
+static struct memory_ranges lb_ranges;
+
+static struct lb_memory *build_lb_mem(struct lb_header *head)
{
- uint64_t end;
- int entries;
- int i;
+ struct lb_memory *mem;
- 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);
- }
- }
-}
+ /* Record where the lb memory ranges will live */
+ mem = lb_memory(head);
+ mem_ranges = 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);
+ /* 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);
+
+ return mem;
}
-static void lb_dump_memory_ranges(struct lb_memory *mem)
+static void commit_lb_memory(struct lb_memory *mem)
{
- int entries;
+ struct memory_range *r;
+ struct lb_memory_range *lb_r;
int i;
- entries = (mem->size - sizeof(*mem))/sizeof(mem->map[0]);
- 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);
+ lb_r = &mem->map[0];
+ i = 0;
+
+ memory_ranges_each_entry(r, &lb_ranges) {
const char *entry_type;
- switch (mem->map[i].type) {
+ switch (memory_range_tag(r)) {
case LB_MEM_RAM: entry_type="RAM"; break;
case LB_MEM_RESERVED: entry_type="RESERVED"; break;
case LB_MEM_ACPI: entry_type="ACPI"; break;
@@ -557,58 +433,25 @@ static void lb_dump_memory_ranges(struct lb_memory *mem)
}
printk(BIOS_DEBUG, "%2d. %016llx-%016llx: %s\n",
- i, entry_start, entry_start+entry_size-1, entry_type);
+ i, memory_range_base(r), memory_range_end(r)-1,
+ entry_type);
- }
-}
-
-/* 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)
-{
- return mem_ranges;
-}
+ 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);
-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;
+ i++;
+ lb_r++;
+ mem->size += sizeof(struct lb_memory_range);
+ }
}
-static void lb_add_rsvd_range(void *gp, struct device *dev, struct resource *res)
+void lb_add_memory_range(struct lb_memory *mem,
+ uint32_t type, uint64_t start, uint64_t size)
{
- struct lb_memory *mem = gp;
- lb_add_memory_range(mem, LB_MEM_RESERVED, res->base, res->size);
+ insert_memory_range(&lb_ranges, start, size, type);
}
-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,
@@ -623,7 +466,7 @@ 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, 0);
+ low_table_end = (unsigned long) lb_table_fini(head);
printk(BIOS_DEBUG, "Table forward entry ends at 0x%08lx.\n",
low_table_end);
low_table_end = ALIGN(low_table_end, 4096);
@@ -682,17 +525,9 @@ unsigned long write_coreboot_table(
high_tables_base, high_tables_size);
#endif /* CONFIG_DYNAMIC_CBMEM */
- /* 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.
- */
+ /* 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);
/* Record our motherboard */
lb_mainboard(head);
@@ -721,5 +556,5 @@ unsigned long write_coreboot_table(
add_cbmem_pointers(head);
/* Remember where my valid memory ranges are */
- return lb_table_fini(head, 1);
+ return lb_table_fini(head);
}