Re: [PATCH] x86/HVM: shrink (pack) struct periodic_time

[Jan Beulich <jbeulich@xxxxxxxx>]

On 29.10.2026 17:23, Andrew Cooper wrote:
> On 29/10/2026 1:56 pm, Jan Beulich wrote:
>> Except for warned_timeout_too_short, which is used with
>> test_and_set_bool(), the booleans there don't need to occupy entire bytes.
>> With them converted to bitfields, pending_intr_nr can move up into the
>> vacated 4-byte slot.
>>
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> test_and_?_bool() are a source of multiple problems, not least because
> they have no business living in xen/types.h
> 
> Next, most users of them really ought to be using the bit forms, not the
> whole boolean form.
> 
> Next next, this particular warned_timeout_too_short has no business
> being an atomic at all.  It's a printk_once() in disguise.  Worse, it's
> a gdprink() so we're doing this atomic work for a no-op in release
> builds.  Also the debug message is tantamount to useless as it fails to
> identify the domain/vcpu which produced the overly short period.
> 
> 
> At a minimum, I think you ought to convert warned_timeout_too_short to
> be like the rest, rather than leaving it like this.  Or remove it; I
> really do question its utility.
> 
> But, if you were to expand the test_and_?_bool() macros to finally
> remove them from the code base, leaving behind just the xchg
> expressions, that would also be very much appreciated.

A lot of largely orthogonal (to this patch) work, which may indeed want
doing, yet surely not right here. There's no problem at all for someone
to do that work on top of this change, assuming it may go in.

Jan