Skip to contentSkip to conversation

v2015.10-rc2: timer not initialized on Kirkwood

U-Boot Simon Guinot 2 participants 3 messages Sep 2, 2015 – Sep 3, 2015

lore
Share & tools
lore
#0 lore raw

Hi,

While testing U-Boot v2015.10-rc2 on the Kirkwood-based LaCie boards I noticed that the autoboot counter is not decrementing. It stays stuck at '3' endlessly. After some digging, I found out that this regression is due to the commit: ade741b3896b1a3872ff74437f04b50762d05849 "arm: mvebu: Call timer_init early before PHY and DDR init".

With this commit it appears that the timer initialization is skipped on Kirkwood boards. As a consequence the timer is not ticking and then all the features relying on the timer are most likely broken.

On the Kirkwood boards, the timer_init function is only called from from the ARM init_sequence. SPL support is disabled. The problem is that the patch introduces a static init_done variable (to prevent multiple timer initializations). But while debugging the timer_init function (via JTAG), I noticed that the init_done initial value is not zero. So the function exists without initializing the timer. A possible explanation is that timer_init is called before the U-Boot relocation, when the BSS segment is still not available...

Maybe we should use an initialized variable instead ?

Regards,

Simon
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20150902/a0726b0b/attachment.sig>

#1 lore raw

Hi Simon,

Quoted 5 lines · Simon Guinot

While testing U-Boot v2015.10-rc2 on the Kirkwood-based LaCie boards I noticed that the autoboot counter is not decrementing. It stays stuck at '3' endlessly. After some digging, I found out that this regression is due to the commit: ade741b3896b1a3872ff74437f04b50762d05849 "arm: mvebu: Call timer_init early before PHY and DDR init".

Sorry about that. Unfortunately I don't have a Kirkwood target here. So I can't test any of the MVEBU changes on this platform.

Quoted 4 lines

With this commit it appears that the timer initialization is skipped on Kirkwood boards. As a consequence the timer is not ticking and then all the features relying on the timer are most likely broken.

Quoted 8 lines

On the Kirkwood boards, the timer_init function is only called from from the ARM init_sequence. SPL support is disabled. The problem is that the patch introduces a static init_done variable (to prevent multiple timer initializations). But while debugging the timer_init function (via JTAG), I noticed that the init_done initial value is not zero. So the function exists without initializing the timer. A possible explanation is that timer_init is called before the U-Boot relocation, when the BSS segment is still not available...

Hmmm. I thought that BSS should be available and cleared at that time. But this does not seem to be the case.

Quoted 1 line

Maybe we should use an initialized variable instead ?

Could you please test the small attached patch? If this fixes the problem for you on your Kirkwood board?

If yes, I'll submit it officially to the list for inclusion.

Thanks,
Stefan

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-arm-mvebu-timer.c-Explicitly-move-init_done-var-to-d.patch
Type: text/x-diff
Size: 1177 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20150903/e93aa0bb/attachment.patch>

#2 lore raw

Hi Stefan,

Quoted 25 lines · Stefan Roese

Hi Simon,

Quoted 5 lines · Simon Guinot

While testing U-Boot v2015.10-rc2 on the Kirkwood-based LaCie boards I noticed that the autoboot counter is not decrementing. It stays stuck at '3' endlessly. After some digging, I found out that this regression is due to the commit: ade741b3896b1a3872ff74437f04b50762d05849 "arm: mvebu: Call timer_init early before PHY and DDR init".

Sorry about that. Unfortunately I don't have a Kirkwood target here. So I can't test any of the MVEBU changes on this platform.

Quoted 4 lines

With this commit it appears that the timer initialization is skipped on Kirkwood boards. As a consequence the timer is not ticking and then all the features relying on the timer are most likely broken.

Quoted 8 lines

On the Kirkwood boards, the timer_init function is only called from from the ARM init_sequence. SPL support is disabled. The problem is that the patch introduces a static init_done variable (to prevent multiple timer initializations). But while debugging the timer_init function (via JTAG), I noticed that the init_done initial value is not zero. So the function exists without initializing the timer. A possible explanation is that timer_init is called before the U-Boot relocation, when the BSS segment is still not available...

Quoted 7 lines

Hmmm. I thought that BSS should be available and cleared at that time. But this does not seem to be the case.

Quoted 1 line

Maybe we should use an initialized variable instead ?

Could you please test the small attached patch? If this fixes the problem for you on your Kirkwood board?

I can confirm that the attached patch does fix the issue.

Thanks for the quick reply.

Simon

Quoted 42 lines

If yes, I'll submit it officially to the list for inclusion.

Thanks,
Stefan

Quoted 1 line

From 9525d96509e36ae534a15f55afd2eae30f9e000a Mon Sep 17 00:00:00 2001

From: Stefan Roese <[email protected]>
Date: Thu, 3 Sep 2015 07:22:42 +0200
Subject: [PATCH] arm: mvebu: timer.c: Explicitly move "init_done" var to data
section

As reported by Simon Guinot, commit ade741b3
"arm: mvebu: Call timer_init early before PHY and DDR init" breaks Kirkwood platforms. As the static variable "init_done" is not available at that early boot time. This patch moves it to explicitly to the data section, making it available at that time.

---
 arch/arm/mach-mvebu/timer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-mvebu/timer.c +1 -1
index c516c41..5449a89 100644
@@ -41,7 +41,7 @@
 #define timestamp			gd->arch.tbl
 #define lastdec				gd->arch.lastinc
 
-static int init_done;
+static int init_done __attribute__((section(".data"))) = 0;
 
 /* Timer reload and current value registers */
 struct kwtmr_val {
Signature · 1 line
2.5.1

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20150903/9eac5742/attachment.sig>