Re: [PATCH 1/3] arm: socfpga: Enable ECC of L2 and OCRAM on startup.

On 10/04/2014 11:18 PM, Masami Hiramatsu wrote:

(2014/10/04 6:42), Dinh Nguyen wrote:

quoted

On 10/03/2014 04:51 AM, Masami Hiramatsu wrote:

quoted

Hi Dinh,

(2014/10/02 20:38), Dinh Nguyen wrote:

quoted

On 10/1/14, 5:18 PM, Thor Thayer wrote:

quoted

On 10/01/2014 04:10 PM, Dinh Nguyen wrote:

quoted

On 10/1/14, 4:07 PM, Thor Thayer wrote:

quoted

On 10/01/2014 12:13 PM, Dinh Nguyen wrote:

quoted

On 10/1/14, 11:31 AM,tthayer-yzvPICuk2ABMcg4IHK0kFoH6Mc4MB0Vx@public.gmane.org  wrote:

quoted

From: Thor Thayer<redacted>

[...]

quoted

quoted

quoted

      static void socfpga_cyclone5_restart(enum reboot_mode mode, const
char *cmd)

@@ -98,6 +101,13 @@ static void socfpga_cyclone5_restart(enum

reboot_mode mode, const char *cmd)
        writel(temp, rst_manager_base_addr + SOCFPGA_RSTMGR_CTRL);
    }
    +static void __init socfpga_cyclone5_init(void)
+{
+    of_platform_populate(NULL, of_default_bus_match_table,
+                 NULL, NULL);

Why is this needed?

Dinh

Hi Dinh.

Are you asking why the of_platform_populate() is needed? If so, it is
used to kick off discovery of devices at the root of the tree.

I was asking about of_platform populate(). This was removed in this
commit:

8b5c18f05 ARM: l2c: socfpga: convert to generic l2c OF initialisation

Just trying to understand what's the need to add it back?

Dinh

It is used to kick off discovery of devices at the root of the tree.
This is needed when .init_machine was defined because we're not using
the default implementation (which called this function).

Then, can you please add a separate patch to fix up the removal of the call?

It seems that this patch does enough reasonable thing.

To initialize ECC on OCRAM, socfpga_init_ocram_ecc() must be invoked
while booting. In that case, he might need this function.

8b5c18f05 is to replace l2x0_of_init() with .l2c_aux_* fields, and
this patch has a different reason to call another init function in
.init_machine. In this case, it is natural to define .init_machine
handler in this patch.

8b5c18f05 also removes the need for .init_machine because
of_platform_populate() is now called by common arm code.

quoted

IOW, both use same socfpga_cyclone5_init() for the function name,
but those have different purpose. I think it is not reviving the
old function, but just adding a new function which has same name.

So I'm just wondering if the call to socfpga_init_ocram_ecc() can be
done in socfpga_init_irq() along with the call to socfpga_init_l2_ecc(),
so that you don't have to bring back the custom init_machine callback.

Indeed, it will be better and simpler.

Thank you,

Hi. I didn't explain this well in my previous comments.

The call to socfpga_init_ocram_ecc() needs to be in .init_machine 
because socfpga_init_ocram_ecc() needs the device tree structure to be 
populated to dereference the sram handle. It is in the .init_machine 
function that the device tree discovery/population process is kicked off 
so it makes sense to add it here.

I originally tried adding socfpga_init_ocram_ecc() in the 
socfpga_init_irq() but the sram phandle couldn't be parsed.

I'm moving the OCRAM addition into a separate patch as requested by Dinh.

Thor

--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html