Hi Stefan,
On Thu, Jun 27, 2013 at 9:12 PM, Stefan Roese sr@denx.de wrote:
Hi Simon,
On 06/28/2013 04:14 AM, Simon Glass wrote:
I can create something with or without compression and I cannot see the failure. Tom is seeing it though. I will see if I can get another board to try. If you have any ideas please let me know.
Sorry, but I don't have any ideas. I also checked the parameters of the gunzip call. And they were identical in the working (before this patch) and non-working version (with this patch). gunzip() just didn't return.
One idea: It might have something to do with the interrupt disabling, as powerpc compresses to 0. And thats the location of the interrupt vectors (timer?). Just an idea. Do you remember shuffling the disable_interrupts() call around?
I'll check myself in a few hours.
Thanks. Apparently it happens on an ARM board also.
I am writing some more sandbox tests to increase the test coverage and hopefully spot something. Please let me know if you have any ideas on this. It seems like (at least in the ARM case) it must be returning BOOTM_ERR_OVERLAP from bootm_load_os(). In your case it may be dying before then by overwriting memory. So the addresses used in that function may be the key to it. However you say they are the same, so I am not sure.
I am also doing a careful diff against the Chromium tree, which seems to work, to try to see if there is a difference somewhere.
Regards, Simon