lp64: fix undefined reference compilation breakage #9130
Conversation
__udivdi3 isn't defined on LP64, don't export it. Fixes: a306792 ("llext: add missing symbol exports") Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
|
fixes #9129 |
| return ~crc; | ||
| } | ||
|
|
||
| #if !CONFIG_SOC_MIMX9352_A55 |
There was a problem hiding this comment.
CONFIG_SOC_MIMX9352_A55 != 64 bits
There was a problem hiding this comment.
@andyross can you help here?
EDIT: also tagging @keith-packard
There was a problem hiding this comment.
This is a libgcc symbol implementing a long divide (exactly which operation I'd have to look up, there are lots of variants). The bug here is presumably that some platforms (presumably only the NXP listed) are aarch64 devices with have a hardware divide, and so libgcc.a on those platforms doesn't include it. So trying to EXPORT_SYMBOL() ends up hitting a link failure trying to extract an address (because it doesn't exist). This seems not unreasonable.
I think if one wanted to nitpick:
- It should probably check something more generic, like CONFIG_64BIT, CONFIG_SOC_FAMILY_ARM64 (also X86_64 if we ever want to do 64 bit fuzzing!), or a GCC arch symbol like x86_64 or __ARM_ARCH_8A.
- It's not a SOF symbol to begin with , and this EXPORT_SYMBOL() belongs upstream in Zephyr llext somewhere.
|
Ping? Would be nice to get this solved somehow as this is now showing up as a fail on every PR. |
andyross
left a comment
andyross
left a comment
There was a problem hiding this comment.
Realize I never gave this a +1. Still think this would be better expressed with an expression that correctly detects more "platforms that don't have __udivdi3" than just that one NXP ARM core, but no reason not to merge.
|
Thanks all, merging this to make PR CI pass. Please follow-up with improvements discussed. |
|
Better fix submitted in: |
5 participants
__udivdi3 isn't defined on LP64, don't export it.
Fixes: a306792 ("llext: add missing symbol exports")