On Mon, Dec 7, 2020 at 11:29 AM Ard Biesheuvel ardb@kernel.org wrote:
On Mon, 7 Dec 2020 at 19:08, dann frazier dann.frazier@canonical.com wrote:
On Mon, Dec 7, 2020 at 10:50 AM Ard Biesheuvel ardb@kernel.org wrote:
On Mon, 7 Dec 2020 at 18:26, dann frazier dann.frazier@canonical.com wrote:
This reverts commit c042dd600f4e89b6e7bdffa00aea4d1d3c1e9686.
This caused the build to emit ADR_PREL_PG_HI21 relocations in the sha{1,2}_ce modules. This relocation type is not supported by the linux-4.4.y kernel module loader when CONFIG_ARM64_ERRATUM_843419=y, which we have enabled, so these modules now fail to load:
[ 37.866250] module sha1_ce: unsupported RELA relocation: 275
This issue does not exist with the backport to 4.9+. Bisection shows that this is due to those kernels also having a backport of commit 41c066f ("arm64: assembler: make adr_l work in modules under KASLR")
Hi Dann,
Would it be an option to backport 41c066f as well?
Hi Ard,
That was attempted before, but caused a build failure which would still happen today: https://www.spinics.net/lists/stable/msg179709.html Specifically, head.S still has a 3 argument usage of adr_l. I'm not sure how to safely fix that up myself.
Given that the original reason for reverting the backport of 41c066f no longer holds (as there are other users of adr_l in v4.4 now), I think the best solution is to backport it again, but with the hunk below folded in. (This just replaces the macro invocation with its output when called with the 3 arguments in question, so the generated code is identical)
--- a/arch/arm64/kernel/head.S +++ b/arch/arm64/kernel/head.S @@ -424,7 +424,8 @@ __mmap_switched: str xzr, [x6], #8 // Clear BSS b 1b 2:
adr_l sp, initial_sp, x4
adrp x4, initial_sp
add sp, x4, :lo12:initial_sp str_l x21, __fdt_pointer, x5 // Save FDT pointer str_l x24, memstart_addr, x6 // Save PHYS_OFFSET mov x29, #0
Thanks Ard - that works. I'll follow-up with a backport patch.
-dann