Reserve any memory above 768MB to prevent u-boot from relocating fdt or initrd data into memory that Linux cannot reference during early boot. Code taken (with modifications) from the powerpc bootm.c.
Signed-off-by: David A. Long dave.long@linaro.org --- arch/arm/lib/bootm.c | 24 +++++++++++++++++++++++- 1 files changed, 23 insertions(+), 1 deletions(-)
diff --git a/arch/arm/lib/bootm.c b/arch/arm/lib/bootm.c index 802e833..437ef35 100644 --- a/arch/arm/lib/bootm.c +++ b/arch/arm/lib/bootm.c @@ -32,6 +32,10 @@
DECLARE_GLOBAL_DATA_PTR;
+#ifndef CONFIG_SYS_LINUX_LOWMEM_MAX_SIZE +#define CONFIG_SYS_LINUX_LOWMEM_MAX_SIZE (768*1024*1024) +#endif + #if defined (CONFIG_SETUP_MEMORY_TAGS) || \ defined (CONFIG_CMDLINE_TAG) || \ defined (CONFIG_INITRD_TAG) || \ @@ -60,7 +64,25 @@ static int bootm_linux_fdt(int machid, bootm_headers_t *images);
void arch_lmb_reserve(struct lmb *lmb) { - ulong sp; + phys_size_t bootm_size; + ulong size, sp, bootmap_base; + + bootmap_base = getenv_bootm_low(); + bootm_size = getenv_bootm_size(); + +#ifdef DEBUG + if (((u64)bootmap_base + bootm_size) > + (CONFIG_SYS_SDRAM_BASE + (u64) gd->ram_size)) + puts("WARNING: bootm_low + bootm_size exceed total memory\n"); +#endif + + size = min(bootm_size, CONFIG_SYS_LINUX_LOWMEM_MAX_SIZE); + + if (size < bootm_size) { + ulong base = bootmap_base + size; + printf("WARNING: adjusting available memory to %lx\n", size); + lmb_reserve(lmb, base, bootm_size - size); + }
/* * Booting a (Linux) kernel image
Hi David,
Le 19/08/2011 19:57, David Long a écrit :
What exactly prevents ARM Linux from booting when FDT or initrd are above 768MB? Can this limitation not be lifter on the Linux side?
Also:
I'd tend to think that if no max size was defined, then we should not limit.
Amicalement,
On Sat, 2011-08-20 at 08:01 +0200, Albert ARIBAUD wrote:
What exactly prevents ARM Linux from booting when FDT or initrd are above 768MB? Can this limitation not be lifter on the Linux side?
I don't think it reasonably can be.
I have to start by saying that I am in no way a Linux VM expert. But...
With Linux HIGHMEM configured the last 1/4GB of the kernel virtual address space is reserved for a few things, including explicitly mapping in parts of the last 1/4GB of physical memory as needed. I just don't think the VM subsystem is initialized enough to do this remapping so early, even if one wanted to make the calls. Indeed, I don't see how it could be initialized this early since the FDT can contain information about physical memory.
Some other platforms limit u-boot to accessing the first 8MB of RAM using CONFIG_SYS_BOOTMAPSZ, but that resulted in a failure to boot in my test. I did not chase down exactly why that approach failed.
I was just going with the example already in place for powerpc. I'm reluctant to change this without having more background on why it's this way now.
-dl
-
Albert ARIBAUD -
David Long