Lubomir Rintel has uploaded this change for review.
vx900: skip remap of high memory ranges if unnecessary
If the DRAM does *not* actually overlap the PCI space, the remap code
notices it is the case, but for some reason proceeds with the remapping,
attempting to remap a negatively sized chunk. Bummer.
With a single 1024M (two ranks of 512M) module:
Nothing to remap
Mem remapping enabled
Remapstart 5120(MB)
Remapend 6144(MB)
Top of RAM 1024MB
New top of RAM 2560MB
Wrote remap map a0101
Mem remapping enabled
Remapstart 4096(MB)
Remapend 2560(MB)
New top of memory is at 2560MB
Needless to say, subsequent ram_resource() ruins the memory map for the
OS -- Linux won't boot without a mem= argument and memtest quickly.
TEST=memtest and Linux boot on HP t5550 with 1024M of memory
Change-Id: Ic221723a26c5d1a03bf34c7722b0abe115f456ba
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
---
M src/northbridge/via/vx900/northbridge.c
1 file changed, 3 insertions(+), 1 deletion(-)
git pull ssh://review.coreboot.org:29418/coreboot refs/changes/71/22271/1
diff --git a/src/northbridge/via/vx900/northbridge.c b/src/northbridge/via/vx900/northbridge.c
index 3638293..774f744 100644
--- a/src/northbridge/via/vx900/northbridge.c
+++ b/src/northbridge/via/vx900/northbridge.c
@@ -120,6 +120,7 @@
*/
if (tolm >= vx900_get_top_of_ram(mcu)) {
printk(BIOS_DEBUG, "Nothing to remap\n");
+ return 0;
}
/* This is how the Vendor BIOS. Keep it for comparison for now */
@@ -275,7 +276,8 @@
uma_memory_size >> 20);
/* FIXME: How do we handle remapping above 4G? */
u64 tor = vx900_remap_above_4g(mcu, pci_tolm);
- ram_resource(dev, idx++, RAM_4GB >> 10, (tor - RAM_4GB) >> 10);
+ if (tor)
+ ram_resource(dev, idx++, RAM_4GB >> 10, (tor - RAM_4GB) >> 10);
set_late_cbmem_top(tolmk << 10);
To view, visit change 22271. To unsubscribe, visit settings.