On Thu, Aug 22, 2019 at 01:14:23PM -0700, Andrey Smirnov wrote: > On Thu, Aug 22, 2019 at 5:52 AM Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> wrote: > > > > get_runtime_offset shall return the offset between the address we are > > running at and the address we are linked at. This value obviously > > changes when we relocate the binary. cf3b09737b tried to avoid using > > R_AARCH64_RELATIVE relocations, but in fact this is exactly what the > > function needs to work. Consider barebox starting at 0x10000000 > > when we are linked at 0x0 then get_runtime_offset() should return > > 0x10000000 before relocate_to_current_adr(), but afterwards it should > > return 0x0. > > > > This patch brings back the previously removed "a" flag. Since gcc5 > > doesn't put the values of R_AARCH64_RELATIVE fixup'd relocations > > into the binary but zeroes instead, we help ourselves by basing > > get_runtime_offset on an address which actually is zero. With > > CONFIG_RELOCATABLE=y the binary is always linked to 0x0, so _text > > is initially zero. > > > > This paragraph might be worth putting in the comment as well to > document some cleverness that is still required to deal with GCC5. Makes sence. Added this: /* * With older gcc versions (gcc5) function pointers will not be filled * into the binary during compile time and instead rely on relocation * during runtime. In the binary we'll always have 0x0 here. We deliberately * use _text here since that is 0x0 and is correct without relocation. */ Sascha -- Pengutronix e.K. | | Industrial Linux Solutions | http://www.pengutronix.de/ | Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 | _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
