Hi Albert,
On Fri, Feb 17, 2012 at 3:08 AM, Albert ARIBAUD albert.u.boot@aribaud.net wrote:
Hi Simon,
Le 21/11/2011 21:49, Simon Glass a écrit :
- .text :
- {
- __text_start = .;
This assignment to __text_start does not exist in any of the existing u-boot.lds files. What is the point of it?
I would like to have a single symbol for the start of the text region across all architectures. ARM has _start but this is not present elsewhere. Also I think it should be defined by the link script.
- .u_boot_cmd : { *(.u_boot_cmd) }
- __u_boot_cmd_end = .;
- . = ALIGN(4);
- __image_copy_end = .;
Ditto for __image_copy_end.
This is the end of the region that needs to be copied when U-Boot is relocated. The symbol exists in ARMv7 so I have reproduced it here. If I read the SPL code correctly then it seems to need this.
These two changes are unexplained in the commit message. Mind you, the one about CPUDIR and start.S isn't either... and it should, because once the commit is in, there is no indication left that it is part of a set, so readers will have a difficulty spotting the changes introduced.
OK I will take a look at these problems.
But what bothers me most is that the patch set produces u-boot.bin files which are not binary identical to those produced without the patch set; if I remove the two assignments, then binary identity is preserved.
Note: I check for binary identity by diff'ing hex dumps of u-boot.bin files produced with and without the patch set. If the only difference is the build version and date, I deem the files binary identical. The dump is done with 'od -t x1z -A x u-boot.bin'.
So unless there is a compelling and strictly unavoidable reason for these assignements, please drop them in the V3 patch set.
The binary will need to include values for these symbols in the dynsym area, so adding a link symbol cannot produce an identical binary. But the change should be harmless - it is just an extra symbol.
I do not want to drop these - in fact I want to standardize on these (or something similar that we agree) across all architectures since it makes the generic relocation code easier (at present it uses #ifdef to work out what to do in the two different cases (SPL and non-SPL). Ultimately we can look towards more uniformity across architectures in the .lds files.
I will take another look at this series.
Amicalement,
Albert.
Regards, Simon