| Message ID | 1454578864-1136-1-git-send-email-ard.biesheuvel@linaro.org |
|---|---|
| State | New |
| Headers | show |
diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S index 8b60fde5ce48..931ebd0696ea 100644 --- a/arch/arm/kernel/vmlinux.lds.S +++ b/arch/arm/kernel/vmlinux.lds.S @@ -293,11 +293,11 @@ SECTIONS */ .text_itcm ITCM_OFFSET : AT(__itcm_start) { - __sitcm_text = .; + __sitcm_text = ABSOLUTE(.); *(.tcm.text) *(.tcm.rodata) . = ALIGN(4); - __eitcm_text = .; + __eitcm_text = ABSOLUTE(.); } /* @@ -314,10 +314,10 @@ SECTIONS .data_dtcm DTCM_OFFSET : AT(__dtcm_start) { . = ALIGN(4); - __sdtcm_data = .; + __sdtcm_data = ABSOLUTE(.); *(.tcm.data) . = ALIGN(4); - __edtcm_data = .; + __edtcm_data = ABSOLUTE(.); } /* Reset the dot pointer or the linker gets confused */
The new base relative kallsyms implementation that is queued for inclusion in v4.6 assumes that kernel symbols are within 2 GB of each other, and hence splits the 32-bit kallsyms symbol address space into 2 GB of absolute addresses and 2 GB of relative values. This triggers a pathological case on ARM, where, if CONFIG_KALLSYMS_ALL, CONFIG_HAVE_TCM and CONFIG_VMSPLIT_1G are all enabled, the distance between kernel symbols exceeds this 2 GB, and the kallsyms table generation fails. Since both VMSPLIT_1G and HAVE_TCM are niche features, and there is no code in mainline that actually populates the TCM code and data regions, we can work around the issue by hiding the TCM section markers from kallsyms, by emitting them as absolute symbols. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> --- arch/arm/kernel/vmlinux.lds.S | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) -- 2.5.0 _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel