On Sun, Apr 14, 2019 at 12:46:25PM +0200, Soeren Moch wrote:
Hi Stefano,
On 14.04.19 11:52, stefano babic wrote:
Hallo Soeren,
after pulling from Tom and merging u-boot-imx, - next, your board is not buildable anymore. Better: it is buildable, but size exceeds:
arm: + tbs2910+u-boot.imx exceeds file size limit:
- limit: 392192 bytes
- actual: 396288 bytes
- excess: 4096 bytes
Just dropping CONFIG_EFI_PARTITION (I supposed EFI is not used on this board), solves the issue. I can do myself in this way or let me know what is your preferred way.
Hm, the full name of this board is "TBS2910 Matrix ARM mini PC", so it is supposed to provide a full PC environment. It supports booting from SATA, and such disk may as well be partitioned with EFI. It is really terrible to drop user visible features, just to support a new "driver model" that is supposed to ease life for developers, but in fact is much work and a real pain for maintainers of existing boards, and brings exactly nothing for the board's users.
So much for that. Now to your question.
Would be OK for me to drop CONFIG_EFI_PARTITION for now since it is more important to get this series merged, because other boards also depend on the SATA patches. We can readd EFI_PARTITION support later. For sure there will be more patches in this development cycle. But if you can wait a few hours, I will look for a better solution, or send a formal patch for this CONFIG_EFI_PARTITION removal.
Now that I'm back from vacation, I do want to chime in here. So, yes, this presents a difficult challenge. I don't want to encourage people to take out functionality just to make the binary fit. And I assume the binary size limit is so that we don't overwrite something else, so we can't just grow it either. So what options do we have? Taking out CONFIG_DM_WARN and CONFIG_MMC_VERBOSE gives us about 2KB. Changing the loglevel from 4 to 3 saves another 2KB. Dropping the memtest command (which is not helpful and discouraged) gives another 1KB.
But these are also, well, stopgaps. Is it all going to be worth it? Well, I hope so. We can for example gain another 1KB back by disabling block cache, which does improve read speed from filesystems and thus something end users see. But it could have been written without DM, it's just a new feature and "carrot" to move to DM.
I don't even see any obvious-to-me cases of including big hunks of data that should be dropped, so that brings us to maybe-not-used functionality like bootelf/bootvx/nfs and so on. Which I don't like.
So, I'm trying to keep a harder still eye on size growth when pulling in functionality. And I kinda wish LTO was easier to bring in as that might really save us some space.