Hi Lokesh,
thanks for having a look!
On 02/02/17 05:35, Lokesh Vutla wrote:
On 2/2/2017 5:18 AM, André Przywara wrote:
On 27/01/17 16:39, Andrew F. Davis wrote:
Hi,
spl_init on some boards is called after stack and heap relocation, on some platforms spl_relocate_stack_gd is called to handle setting the limit to its value CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN when simple SPL malloc is enabled during relocation. spl_init should then not re-assign the old pre-relocation limit when this is defined.
I am sorry to say this, but this very patch breaks SPL boot on sunxi boards (tested on the Pine64, the Orangepi PC2 (with a similar ARMv8 SoC) and the OrangePi Zero (ARMv7 Allwinner H3 SoC).
U-Boot SPL 2017.03-rc1-00022-gf77309d (Feb 01 2017 - 23:27:19) DRAM: 1024 MiB Trying to boot from MMC1 MMC Device 0 not found spl: could not find mmc device. error: -19 SPL: failed to boot from all boot devices ### ERROR ### Please RESET the board ###
sunxi boards both use the default CONFIG_SYS_MALLOC_F_LEN=0x400 and CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x100000.
In your case, before this patch malloc_limit is MALLOC_F_LEN(0x400), after this patch malloc_limit is SPL_STACK_R_MALLOC_SIMPLE_LEN(0x100000) . Stack relocated address is 0x4fe00000(is this in DDR?). SPL_STACK_R is mainly targetted for relocating SPL stack to DDR.
Yes, this is at DDR (which starts at 1G).
- Is 1MB available at 0x4fe00000?
Yes, the smallest board I have has 256MB, though I've seen this on boards with one GB as well. But even in the worst case this is 2 MB below end of DDR. The BSS is at 0x4ff80000, but that shouldn't conflict then.
Is this address actually the _starting_ address for the malloc heap? So malloc uses memory above this address, the stack below? Or does malloc use some stack allocation for its purpose?
- Can you reduce the size or change SPL_STACK_R to DDR and verify?
I will check tonight (no boards here at the moment).
Thanks for the explanation and the pointers, this gives me something to debug on.
Cheers, Andre
Thanks and regards, Lokesh
At the moment the code there is beyond me, so I can't really say how this breaks, but it clearly does: reverting this patch makes U-Boot happy again.
Any ideas?
Cheers, Andre.
Signed-off-by: Andrew F. Davis afd@ti.com
common/spl/spl.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/common/spl/spl.c b/common/spl/spl.c index 462c3a2b97..a3808a988b 100644 --- a/common/spl/spl.c +++ b/common/spl/spl.c @@ -183,7 +183,12 @@ int spl_init(void) int ret;
debug("spl_init()\n"); -#if defined(CONFIG_SYS_MALLOC_F_LEN) +/*
- with CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN we set malloc_base and
- malloc_limit in spl_relocate_stack_gd
- */
+#if defined(CONFIG_SYS_MALLOC_F_LEN) && \
- !defined(CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN)
#ifdef CONFIG_MALLOC_F_ADDR gd->malloc_base = CONFIG_MALLOC_F_ADDR; #endif