On Thu, 06 Jul 2023 16:40:27 PDT (-0700), rdunlap@xxxxxxxxxxxxx wrote:
Hi Palmer,
On 7/6/23 14:46, Palmer Dabbelt wrote:
On Thu, 06 Jul 2023 14:36:05 PDT (-0700), rdunlap@xxxxxxxxxxxxx wrote:
On 7/6/23 14:32, Palmer Dabbelt wrote:
On Thu, 06 Jul 2023 14:27:53 PDT (-0700), rdunlap@xxxxxxxxxxxxx wrote:
On 7/5/23 18:57, Stephen Rothwell wrote:
Hi all,
Please do *not* add any v6.6 related stuff to your linux-next included
branches until after v6.5-rc1 has been released.
Changes since 20230705:
on riscv64:
WARNING: modpost: vmlinux: section mismatch in reference: $xrv64i2p1_m2p0_a2p1_zicsr2p0_zifencei2p0_zihintpause2p0_zmmul1p0+0x14 (section: .text.unlikely.set_bit.constprop.0) -> numa_nodes_parsed (section: .init.data)
Full randconfig file is attached.
Thanks, I'm giving it a look. Do you happen to also have your toolchain version easily availiable? The mapping symbols are new and we've seen some odd stuff happen, something is likely broken somewhere...
I'm using gcc-13.1.0 from https://mirrors.edge.kernel.org/pub/tools/crosstool/
Cool, those are pretty easy to run. I've yet to actually reproduce the
failures, but I'd guess we just want to ignore the mapping symbols with
something like
diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c
index b29b29707f10..2f801469301d 100644
--- a/scripts/mod/modpost.c
+++ b/scripts/mod/modpost.c
@@ -614,6 +614,18 @@ static int ignore_undef_symbol(struct elf_info *info, const char *symname)
/* Expoline thunks are linked on all kernel modules during final link of .ko */
if (strstarts(symname, "__s390_indirect_jump_r"))
return 1;
+
+ /*
+ * RISC-V defines various special symbols that start with "$". The
+ * mapping symbols, which exist to differentiate between incompatible
+ * instruction encodings when disassembling, show up all over the place
+ * and are generally not meant to be treated like other symbols. So
+ * just ignore any of the special symbols.
+ */
+ if (info->hdr->e_machine == EM_RISCV)
+ if (symname[0] == '$')
+ return 1;
+
/* Do not ignore this symbol */
return 0;
}
I haven't even built that, though...
That builds without any errors or warnings, but it doesn't fix the section mismatch
warning.
Turns out there's a hook for mapping symbols, this fixes it for me:
https://lore.kernel.org/r/20230707054007.32591-1-palmer@xxxxxxxxxxxx/
These also trip up backtraces, so we probably need something over there as
well.
I think that might fix the backtraces as well.