On Thu, Dec 06, 2018 at 04:45:08PM +0100, Stefan Agner wrote:
On 06.12.2018 16:13, Tom Rini wrote:
On Thu, Dec 06, 2018 at 02:58:52PM +0100, Stefan Agner wrote:
From: Stefan Agner stefan.agner@toradex.com
Drop the statically allocated get_contents_vfatname_block and dynamically allocate a buffer only if required. This saves 64KiB of memory.
Signed-off-by: Stefan Agner stefan.agner@toradex.com
Is there a particular issue you're running into here, or just that what we're doing is just generally not a great idea? I ask since I need to poke git history to make sure that the thing at the back of my mind about us needing to do this for some annoying boot case is intentional.
I was in the process checking large global variables since I had issues on a target with limited SRAM (used pre-relocation). I only after writing this patch I realized that the space for the .bss section is not required pre-relocation, so the patch did not help in my particular case.
That said, in git history I saw that get_contents_vfatname_block has been used more broadly in earlier FAT implementations. Now it is only used to handle a corner case: when loading from a file with an offset... I guess the .bss section is cleared, so we probably safe 64KiB of zeroing out... I feel it is still worth doing the change, but I don't have a really strong feeling.
Thanks for checking. Yes, this is probably a good thing to do overall and whatever historical reason we needed it like this isn't applicable now if we're doing offset loading only in this case which isn't an SPL thing anyhow.