Re: [PATCH 00/14] alpha: fix entry bugs and consolidate generic helpers

[Matt Turner]

On Tue, Nov 3, 2026 at 4:43 PM Magnus Lindholm <linmag7@xxxxxxxxx> wrote:
> With GENERIC_ENTRY upstream for Alpha, this series moves syscall dispatch
> and ordinary interrupt/exception entry and exit into C using generic
> helpers, and removes redundant fork, core-dump, audit and ptrace code.

Each patch builds, and the whole series boots and runs fine on my ES47
(EV7) with both CPUs enabled and lockdep on. seccomp_bpf passes 142/142
as both a normal user and root, and the ptrace syscall_info and
peeksiginfo selftests pass. No lockdep reports or warnings in dmesg.

I also wrote a small test that reads 512 bytes from /dev/zero in a loop
while a 1ms itimer fires SIGALRM. Without the series, 188 of 200000 reads
came back wrong; with it, none did. So patch 4 fixes a real problem.

Two small things, neither worth a respin on its own:

Patch 2: the comment in regs_return_value() says a3 is an error flag
only at syscall exit, but the code negates r0 for any user-mode regs
with r19 set. That's the same as MIPS, and nothing on alpha calls it
outside syscall exit today (we have no uprobes), so the code is fine.
The comment just reads as though there's a check that isn't there.

Patches 3/5/6: the ALPHA_SYSCALL_META_* bits come in as 0x10, 0x20,
then 0x8, and the RESULT define has an extra tab. Cosmetic.

For the series:

Tested-by: Matt Turner <mattst88@xxxxxxxxx>
Reviewed-by: Matt Turner <mattst88@xxxxxxxxx>