summaryrefslogtreecommitdiff
path: root/src/arch/armv7/bootblock.inc
diff options
context:
space:
mode:
authorRonald G. Minnich <rminnich@gmail.com>2013-08-16 15:57:56 -0700
committerIsaac Christensen <isaac.christensen@se-eng.com>2014-08-06 22:07:06 +0200
commitd6b16f54b9c34a8095a3eefbaf334150c15cecb5 (patch)
treeb4efe93ba92e56b6d167eb215eeb59355544161a /src/arch/armv7/bootblock.inc
parent7111835f3900aaf2e1a23a62029daa668963fbe1 (diff)
Set armv7 up for cpu_info to work as on x86 (so threads can work)
On x86, cpu_info lives at the top of stack. Make the arm do that as well, as the threading model needs that and so will multicore support. As part of this change, make the stack size a power of 2. Also make it much smaller -- 2048 bytes is PLENTY for ram stage. Note that the small stack size is counterintuitive for rom stage. How can this work in rom stage, which needs a HUGE stack for lzma? The main use of STACK_SIZE has always been in ram stage; since 2002 or so it was to size per-core stacks (see, e.g., src/arch/x86/lib/c_start.S:.space CONFIG_MAX_CPUS*CONFIG_STACK_SIZE and, more recently, thread stacks. So, we define the STACK_TOP for rom and ram stage, but the STACK_SIZE has no real effect on the ROM stage (no hardware red zones on the stack) and hence we're ok with actually defining the "wrong" stack size. In fact, the coreboot_ram ldscript for armv7 sizes the stack by subtracting CONFIG_STACK_BOTTOM from CONFIG_STACK_TOP, so we replicate that arithmetic in bootblock.inc Observed stack usage in ramstage: BS: BS_PAYLOAD_LOAD times (us): entry 1 run 153887 exit 1 Jumping to boot code at 23104044 CPU0: stack: 02072800 - 02073000, lowest used address 020728d4, stack used: 1836 bytes entry = 23104044 Which means we do need 2K, not 1K. Change-Id: I1a21db87081597efe463095bfd33c89eba1d569f Signed-off-by: Ronald G. Minnich <rminnich@gmail.com> Reviewed-on: https://gerrit.chromium.org/gerrit/66135 Reviewed-by: Ronald G. Minnich <rminnich@chromium.org> Tested-by: Ronald G. Minnich <rminnich@chromium.org> Commit-Queue: Ronald G. Minnich <rminnich@chromium.org> (cherry picked from commit f011097e9f2bfb2f4c1109d465be89a79a65ba3e) Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com> Reviewed-on: http://review.coreboot.org/6501 Tested-by: build bot (Jenkins) Reviewed-by: David Hendricks <dhendrix@chromium.org> Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com> Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Diffstat (limited to 'src/arch/armv7/bootblock.inc')
-rw-r--r--src/arch/armv7/bootblock.inc12
1 files changed, 10 insertions, 2 deletions
diff --git a/src/arch/armv7/bootblock.inc b/src/arch/armv7/bootblock.inc
index b2c993af32..c45259db6d 100644
--- a/src/arch/armv7/bootblock.inc
+++ b/src/arch/armv7/bootblock.inc
@@ -82,8 +82,15 @@ init_stack_loop:
/* Set stackpointer in internal RAM to call board_init_f */
call_bootblock:
ldr sp, .Stack /* Set up stack pointer */
- bic sp, sp, #7 /* 8-byte alignment for ABI compliance */
ldr r0,=0x00000000
+ /*
+ * The current design of cpu_info places the
+ * struct at the top of the stack. The number of
+ * words pushed must be at least as large as that
+ * struct.
+ */
+ push {r0-r2}
+ bic sp, sp, #7 /* 8-byte alignment for ABI compliance */
/*
* Use "bl" instead of "b" even though we do not intend to return.
* "bl" gets compiled to "blx" if we're transitioning from ARM to
@@ -104,5 +111,6 @@ wait_for_interrupt:
.Stack:
.word CONFIG_STACK_TOP
.align 2
+/* create this size the same way we do in coreboot_ram.ld: top-bottom */
.Stack_size:
- .word CONFIG_STACK_SIZE
+ .word CONFIG_STACK_TOP - CONFIG_STACK_BOTTOM