Patchwork [U-Boot] arm imx35 timer failure

login
register
mail settings
Submitter Laurent Joye
Date March 29, 2011, 2:24 p.m.
Message ID <1301408699.3605.24.camel@hrwks7001-l.haslerrail.net>
Download mbox | patch
Permalink /patch/88787/
State Changes Requested
Headers show

Comments

Laurent Joye - March 29, 2011, 2:24 p.m.
Hello,
I'm developing a product based on the imx35 cpu with NOR and NAND flash.
The SBC I use is the pcm-043 from Phytec.
I use the FEC to communicate over the ethernet.

I've made my own BSP for Das U-boot based on the release tagged
v2011.03-rc2 on the git://git.denx.de/u-boot.git master repository.

I've encountered one problem in uboot so that my board can work
properly:

This concerns the FEC autonegotiation timeout.
I had some MDIO read an write and autonegotiation wait timeout failures.
By debugging the problem I've found that the changes I've brought to the
timer.c file (see patch below for arch/arm/cpu/arm1136/timer.c) of the
imx35 cpu solves the problem.
It seems that the return value of the "get_timer_masked" function is not
given with the same unit as the input value for the "get_timer"
function. It seems also to me that this value should be dependent of the
"CONFIG_SYS_HZ" macro.

Did anybody encountered this problem?
Could you please tell me if there is really something wrong in timer.c
file?

Laurent Joye
HaslerRail AG
Bern, Switzerland

Related patch:
Stefano Babic - April 5, 2011, 1:48 p.m.
On 03/29/2011 04:24 PM, Laurent Joye wrote:
> Hello,
> I'm developing a product based on the imx35 cpu with NOR and NAND flash.
> The SBC I use is the pcm-043 from Phytec.
> I use the FEC to communicate over the ethernet.

Hi Laurent,

> It seems that the return value of the "get_timer_masked" function is not
> given with the same unit as the input value for the "get_timer"
> function. It seems also to me that this value should be dependent of the
> "CONFIG_SYS_HZ" macro.

I think you are right. However, examining your problem I checked that
the i.MX35 timer functions are quite different compared to  other i.MX
processors (mX31 and MX5). There is no good reason to make things
different and I think we should refactor this code containing the timer
functions for the i.MX35 (and better having one single file with the
i.MX31 code).

Best regards,
Stefano Babic

Patch

diff --git a/arch/arm/cpu/arm1136/mx35/timer.c
b/arch/arm/cpu/arm1136/mx35/timer.c
index db1e2c9..00ca3e6 100644
--- a/arch/arm/cpu/arm1136/mx35/timer.c
+++ b/arch/arm/cpu/arm1136/mx35/timer.c
@@ -70,7 +70,7 @@  inline ulong get_timer_masked(void)
        struct gpt_regs *gpt = (struct gpt_regs *)GPT1_BASE_ADDR;
        ulong val = readl(&gpt->counter);
 
-       return val;
+       return val / CONFIG_SYS_HZ;
 }
 
 void reset_timer(void)
@@ -84,12 +84,14 @@  ulong get_timer(ulong base)
 
        tmp = get_timer_masked();
 
-       if (tmp <= (base * 1000)) {
+       if (tmp < base) {
                /* Overflow */
-               tmp += (0xffffffff -  base);
+               tmp = (0xffffffff - (base - tmp));
+       } else {
+               tmp = tmp -base;
        }
 
-       return (tmp / 1000) - base;
+       return tmp;
 }
 
 void set_timer(ulong t)
@@ -103,6 +105,7 @@  void set_timer(ulong t)
 void __udelay(unsigned long usec)
 {
        ulong tmp;
+       usec /= CONFIG_SYS_HZ;
 
        tmp = get_timer_masked();       /* get current timestamp */