@@ -162,11 +162,16 @@ void watchdog_init(void)
*/
u32 omap_sdram_size(void)
{
- u32 section, i, total_size = 0, size, addr;
+ u32 section, i, valid;
+ u64 sdram_start = 0, sdram_end = 0, addr,
+ size, total_size = 0, trap_size = 0;
for (i = 0; i < 4; i++) {
section = __raw_readl(DMM_BASE + i*4);
+ valid = (section & EMIF_SDRC_ADDRSPC_MASK) >>
+ (EMIF_SDRC_ADDRSPC_SHIFT);
addr = section & EMIF_SYS_ADDR_MASK;
+
/* See if the address is valid */
if ((addr >= DRAM_ADDR_SPACE_START) &&
(addr < DRAM_ADDR_SPACE_END)) {
@@ -174,9 +179,20 @@ u32 omap_sdram_size(void)
EMIF_SYS_SIZE_SHIFT);
size = 1 << size;
size *= SZ_16M;
- total_size += size;
+
+ if (valid != DMM_SDRC_ADDR_SPC_INVALID) {
+ if (!sdram_start || (addr < sdram_start))
+ sdram_start = addr;
+ if (!sdram_end || ((addr + size) > sdram_end))
+ sdram_end = addr + size;
+ } else {
+ trap_size = size;
+ }
+
}
+
}
+ total_size = (sdram_end - sdram_start) - (trap_size);
return total_size;
}
The DMM sections can be overlapping with each other, with sections 3 to 0 having the highest to lowest priority in that order. There could also be a section that is used trap the unmapped Tiler entries and this trap section could be overlapping with the actual sdram area. So take care of the above scenarios while calculating the size of the actual ram. Signed-off-by: R Sricharan <r.sricharan@ti.com> --- arch/arm/cpu/armv7/omap-common/hwinit-common.c | 20 ++++++++++++++++++-- 1 files changed, 18 insertions(+), 2 deletions(-)