Add macros simplifying the fake preload setup

[mhorne]

Summary

This could have been done in r359673, but I didn't think I would return
to this function so soon. Lift these macros from arm64 so we don't need
to worry about the size when inserting new elements.

[markj]

sys/riscv/riscv/machdep.c

740 ↗

(On Diff #71974)

I guess this is ok since in practice you don't have to worry about unaligned stores?

765 ↗

(On Diff #71974)

Why did the second term change from KERNBASE + KERNENTRY to KERNBASE?

[mhorne]

sys/riscv/riscv/machdep.c

740 ↗

(On Diff #71974)

Yeah, safe in practice but it wouldn't hurt to round-up here as well.

765 ↗

(On Diff #71974)

I discovered that KERNENTRY is useless, but forgot to remove it as I intended. Will update this shortly.

[mhorne]

Remove KERNENTRY entirely. Call roundup in PRELOAD_PUSH_VALUE.

[mhorne]

sys/riscv/riscv/machdep.c

770 ↗

(On Diff #72080)

BTW, anyone have a clue how this code works on arm/arm64? It would be good to enable or just remove it here as appropriate.

sys/arm/include/elf.h has the following hint, but it's not clear enough to me:

/*
 * Magic number for the elf trampoline, chosen wisely to be an immediate
 * value.
 */
#define MAGIC_TRAMP_NUMBER     0x5c000003

[jrtc27]

That's a holdover from the a.out days. See D21099.

[markj]

Seems ok to me aside from the comment.

sys/riscv/riscv/machdep.c

740 ↗

(On Diff #71974)

I just forgot that riscv permits misaligned memory accesses. I'm not sure that the added roundup by 4 really helps anything?

[jrtc27]

sys/riscv/riscv/machdep.c

740 ↗

(On Diff #71974)

It permits it, but it may be emulated by M-mode if the hardware doesn't, so it should be avoided where possible as it will be orders of magnitude slower.

[mhorne]

In D24910#549236, @jrtc27 wrote:

That's a holdover from the a.out days. See D21099.

Neat, did you find that somehow or were you already aware of its usage? I'll remove that section in a follow-up.

[jrtc27]

In D24910#549248, @mhorne wrote:

In D24910#549236, @jrtc27 wrote:

That's a holdover from the a.out days. See D21099.

Neat, did you find that somehow or were you already aware of its usage? I'll remove that section in a follow-up.

git blame on the definition in the header to find its introduction, then git log on elf_trampoline.c to find when that got GC'ed.

[mhorne]

sys/riscv/riscv/machdep.c

740 ↗

(On Diff #71974)

I dropped the rounding in the committed version, it did not change anything for the current set of preload data, and I intend on adding anything smaller than a uint32_t in there.