lib/libc: implement C23 memalignment()
ClosedPublic
Actions

Authored by fuz on Mon, Nov 10, 5:57 PM.

Details

Reviewers

markj
kib
brooks
jrtc27
imp
ziaee

Group Reviewers

manpages
cheri

Commits

rG6c57e368eb17: lib/libc: implement C23 memalignment()

Summary

This new function computes the alignment of a pointer.
If the pointer is a null pointer, null is returned.
I have tried to write an implementation that can cope
with traditional address-based architectures, even if
size_t and uintptr_t are of different length. Adjustments
may be needed for CHERI though.

A man page is provided, too. No unit test for now.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

fuz created this revision.Mon, Nov 10, 5:57 PM

Herald added a subscriber: ziaee. · View Herald TranscriptMon, Nov 10, 5:57 PM

Harbormaster completed remote builds in B68516: Diff 166146.Mon, Nov 10, 5:57 PM

fuz requested review of this revision.Mon, Nov 10, 5:57 PM

Herald added a reviewer: cheri. · View Herald TranscriptMon, Nov 10, 5:57 PM

ziaee added inline comments.Mon, Nov 10, 6:02 PM

lib/libc/stdlib/memalignment.3
2	The hyphen was for a license parser that Warner abandoned years ago.

fuz added inline comments.Mon, Nov 10, 6:12 PM

lib/libc/stdlib/memalignment.3
2	The hyphen follows style(9). Are you sure I should diverge from it?

jrtc27 added inline comments.Mon, Nov 10, 6:35 PM

lib/libc/stdlib/memalignment.c
20	Um, shouldn’t this be ~’ed? Whether as ~(align - 1) or ~align + 1 (or just -align in two’s complement).

fuz planned changes to this revision.Mon, Nov 10, 6:37 PM

fuz added inline comments.

lib/libc/stdlib/memalignment.c
20	Ah... brain fart time ...

correct alignment computation

Harbormaster completed remote builds in B68517: Diff 166150.Mon, Nov 10, 6:39 PM

ziaee added inline comments.Mon, Nov 10, 7:18 PM

lib/libc/stdlib/memalignment.3
2	We took the hyphen out of style(9) in February.

This looks good to my eye.

include/stdlib.h
174	I'm guessing this is too new for POSIX alignment, eh?

This revision is now accepted and ready to land.Mon, Nov 10, 8:10 PM

kib accepted this revision.Tue, Nov 11, 8:52 AM

kib added inline comments.

lib/libc/stdlib/memalignment.c
20	This would deserve a comment that we expect 2-complement. BTW, I think we must compile libc with -fwrapv, if we don't already.

fuz added inline comments.Tue, Nov 11, 10:14 AM

include/stdlib.h
174	Yeah, issue 8 doesn't seem to have it.
lib/libc/stdlib/memalignment.c
20	`uintptr_t` is an unsigned type, so neither is necessary. What may be possible is removing lines 15 and 16. I'll have to check if null pointers always become zero when cast to integer.

kib added inline comments.Tue, Nov 11, 10:17 AM

lib/libc/stdlib/memalignment.c
20	`uintptr_t` is an unsigned type, so neither is necessary. What it means -a for unsigned then? What may be possible is removing lines 15 and 16. I'll have to check if null pointers always become zero when cast to integer. I did not said that -fwrapv is necessary for this commit, I said that we should do this for libc in general (and rtld/libsys/libtr but there I would expect it less significant).

fuz added inline comments.Tue, Nov 11, 10:27 AM

lib/libc/stdlib/memalignment.c
20	What it means -a for unsigned then? The standard just says (ISO/IEC 9899:2024 § 6.5.4.3 ¶ 3): The result of the unary - operator is the negative of its (promoted) operand. The integer promotions are performed on the operand, and the result has the promoted type. The negative of an unsigned number is the same as subtracting that number from zero, which gives `~a + 1`. Please also have a look at D53657 and descendent commits if you have some time.