On Fri, Sep 21, 2018 at 1:43 PM Wolfgang Denk wd@denx.de wrote:
Dear Alex,
In message 20180918081013.26660-1-alex.kiernan@gmail.com you wrote:
When the image which bootm is given can't be booted, call panic with the error message rather than printf/hang so that we can recover from broken images via a bootcount mechanism. If hang on failure is still required then CONFIG_PANIC_HANG can still be enabled.
I wonder if the failing of the FDT creation is a reason to panic / reboot at all? The point of no return is architecture dependent - for example, on Power architecture it is usuallywhen the kernel image gests decompressed/written to RAM, as then the exception vectors get overwritten, so no return to U-Boot is possible.
But the FDT is always (I think ?) just written to an area of RAM that is not critical for the execution of U-Boot itself - so when this fails, why can we not simply return to U-Boot with an eroor indication?
I feel we should use panic/hang only as a last resort, when no other recovery is possible?
Thinking it through, that sounds reasonable to me... the reason I ended up looking at this code was because I deployed a broken FIT image, then got stuck here. But if bootm had returned with a failure I could have done useful recovery actions without needing a reboot.