Hello Simon,
On Wed, 18 Nov 2015 18:05:07 -0700, Simon Glass sjg@chromium.org wrote:
Hi Albert,
early malloc() area
"Arena", and "malloc arena", are established designations for the memory space used by malloc implementations; and I prefer to use this more specific term, as people may use it as a search keyword when looking for malloc related stuff.
Arena is OK, but can you please mention 'early' each time? It's confusing otherwise. I think we should have a clear distinction between the early malloc() and full malloc().
Good point, will do.
If you can have it so that the stack top equals the global_data bottom, then we should be OK.
Will do that in v6.
Note, however, that it will not simplify assembly code: it will turn a subtraction from sp into an assignment to sp, which is not simpler, and it will add an assignment to whatever register represents the first argument, since we'll turn a (void) function into a (ulong top) function, so all in all, it will add one assembly instruction with respect to the 'ulong board_init_f_get_reserve_size(void)' approach.
True, but now we are just passing values around rather than doing arithmetic in assembler.
Well, at the C level an addition may be more complex than an assignment, but at the assembly level, that's pretty much equally simple. Anyway, the change is ok by me.
gd works at this point, and I want to avoid any aliasing issue.
I don't really understand that, but if you want to use gd I think it would be worth a one-line comment.
I believe it is already present in v5:
+ * ALSO IMPORTANT: + * + * On some architectures, gd will only be set once arch_setup_gd() is + * called. Therefore: + * + * Do not use 'gd->' until arch_setup_gd() has been called!
I will expand that comment a bit, and add a one-liner in the code too.
Thanks again for your comments!
Amicalement,