On 22.04.2018 07:46, Maxime Villard wrote: > Le 22/04/2018 à 01:25, Joerg Sonnenberger a écrit : >> Module Name: src >> Committed By: joerg >> Date: Sat Apr 21 23:25:01 UTC 2018 >> >> Modified Files: >> src/sys/arch/amd64/amd64: locore.S >> >> Log Message: >> Do not use movq for loading arbitrary 64bit immediates. The ISA >> restricts it to 32bit immediates. >> >> >> To generate a diff of this commit: >> cvs rdiff -u -r1.163 -r1.164 src/sys/arch/amd64/amd64/locore.S >> >> Please note that diffs are not public domain; they are subject to the >> copyright notices on the relevant files. > > Mmh. Is there a compiler where this makes a difference? On NetBSD/GGG it > doesn't (because if it did, SVS would never have worked), but I see that > on MacOS the instruction indeed makes a difference, the encoding becomes: > > movq 0x0, %rax > > Which is obviously not what we expect. > > Is this the problem you were having a few weeks ago? That is to say, the > kernel that was crashing at boot time, did you compile it on another > system/compiler that generated a "movq 0x0,%rax"? > > Anyway your change seems correct. > > Thanks, > Maxime There are reports that the SVS kernel built by Clang doesn't work.
Attachment:
signature.asc
Description: OpenPGP digital signature