On 21/03/2025 6:52 am, Jan Beulich wrote: > On 20.03.2025 18:56, Andrew Cooper wrote: >> On 18/03/2025 3:26 pm, Jan Beulich wrote: >>> The sole dependency is percpu_init_areas(), >>> which runs - as even visible from patch context here - ahead of >>> early_traps_init(). >> percpu_init_areas() isn't even a dependency. Notice how it only starts >> from CPU1. CPU0 (which is what this_cpu() refers to in both of these >> functions) is correct (i.e. 0) at build time. > Oh, right. > >> I still intend to remove percpu_init_areas(). It's all build-time >> constant; it just needs doing by the assembler, rather than the >> compiler, as (-(long)__per_cpu_start) is not an integer constant expression. > Except that this value also cannot be expressed by a relocation, afaict, > so the assembler would choke when trying to write the object file. (Same > for x86'es alternative initializer.)
Yeah, I found that while experimenting. However, the PROVIDE_HIDDEN() trick I used for IDT autogen does work. There are a couple of nasty thing. While we have ASM_INT() now, we don't have ASM_PTR() or equivalent, which is needed because this is a common file and we can't just use .quad. Also, because this is an asm() block in a C file, we can't use any of linkage.h; opencoding it isn't too bad, but it also appears that __read_mostly doesn't honour -fdata-sections. ~Andrew
