Re: Set PSTATE.PRIV for user process

[Sad Clouds <cryintothebluesky%gmail.com@localhost>]

On Fri, 23 Oct 2026 13:03:28 +0100
Sad Clouds <cryintothebluesky%gmail.com@localhost> wrote:

> Something is really whacky going on here. It feels like during
> save/restore another trap occurs (text or data fault?), causing partial
> register save/restore and messing up their values.
>
 
Another interesting thing I notice is the alignment of window registers
save area on the stack. If I examine the locations in the core dump:

(gdb) x/16xg $sp + 2047
0xffffffffffffbfd0:     0x0000000066666665     l0
0xffffffffffffbfd8:     0x0000000066666663     l1
0xffffffffffffbfe0:     0x0000000040828eb0     l2
0xffffffffffffbfe8:     0x0000000066666662     l3
0xffffffffffffbff0:     0x0000000000000000     l4
0xffffffffffffbff8:     0x6666666666666663     l5
0xffffffffffffc000:     0x5555555555555556     l6  <-- starts on 8 KiB page boundary
0xffffffffffffc008:     0x5555555555555557     l7

0xffffffffffffc010:     0x0000000040a17e80     i0
0xffffffffffffc018:     0x0000000000001400     i1
0xffffffffffffc020:     0x6666666666666662     i2
0xffffffffffffc028:     0x6666666666666663     i3
0xffffffffffffc030:     0x6666666666666664     i4
0xffffffffffffc038:     0xffffffffffffb911     i5
0xffffffffffffc040:     0x6666666666666666     i6
0xffffffffffffc048:     0x6666666666666667     i7

NetBSD sparc64 uses 8 KiB pages. Register %l6 address starts on 8 KiB
page boundary.

Registers %l0-%l5 on one page are saved and restored with correct
values, but registers %l6-%l7 and %i0-%i7 on the next pages are not
saved correctly to those memory addresses.

Could be a coincidence or a smoking gun.