Re: [PATCH 0/2] xen/common: llc-coloring parser fixes

[Mykola Kvach <xakep.amatop@xxxxxxxxx>]

On Mon, May 18, 2026 at 10:06 AM Jan Beulich <jbeulich@xxxxxxxx> wrote:
>
> On 18.05.2026 09:01, Mykola Kvach wrote:
> > On Mon, May 18, 2026 at 9:52 AM Jan Beulich <jbeulich@xxxxxxxx> wrote:
> >>
> >> On 18.05.2026 08:42, Mykola Kvach wrote:
> >>> On Mon, May 18, 2026 at 9:20 AM Jan Beulich <jbeulich@xxxxxxxx> wrote:
> >>>> On 16.05.2026 17:03, Mykola Kvach wrote:
> >>>>> This small series fixes two issues in parse_color_config().
> >>>>>
> >>>>> The first patch makes parse failures leave the caller-visible color 
> >>>>> count
> >>>>> at zero.  This prevents a rejected command-line value from leaving a
> >>>>> partially parsed configuration behind for later init paths to consume.
> >>>>>
> >>>>> The second patch rejects empty color tokens.  Previously, delimiters in
> >>>>> places where a color value was expected could be interpreted as color 0,
> >>>>> because simple_strtoul() returns zero without advancing the input 
> >>>>> pointer.
> >>>>> The patch checks that each parsed color value consumed input.  It also
> >>>>> adds the missing newline to the DT color parsing error message.
> >>>>>
> >>>>> Mykola Kvach (2):
> >>>>>   xen/common: llc-coloring: clear color count on parse failure
> >>>>>   xen/common: llc-coloring: reject empty color tokens
> >>>>
> >>>> For both of these, a question which isn't even considered in the 
> >>>> reasoning
> >>>> is whether the present behavior may be intentional. Especially for the 
> >>>> 2nd
> >>>> ISTR Stefano(?) not so long ago indicating that the behavior is indeed
> >>>> intended to be this way. That may have been somewhere on Matrix rather 
> >>>> than
> >>>> on the list, though.
> >>>
> >>> Thank you for pointing this out.
> >>>
> >>> For the first patch, my reasoning was that returning an error while
> >>> leaving a partially parsed caller-visible color count behind looks
> >>> surprising. If the value is rejected, I think later init paths should
> >>> not be able to consume the partially parsed state.
> >>>
> >>> For the second patch, my reasoning was that the current behavior looks
> >>> accidental rather than an intentional extension of the syntax.
> >>
> >> That was my impression as well, hence why I had raised the question back
> >> then.
> >>
> >>> The parser comment says:
> >>>
> >>> COLOR_CONFIGURATION ::= COLOR | RANGE,...,COLOR | RANGE
> >>> RANGE ::= COLOR-COLOR
> >>>
> >>> The user guide also describes this as a comma-separated list of colors
> >>> or ranges, with ranges written as hyphen-separated inclusive intervals.
> >>> I don't see an empty-token production there.
> >>
> >> What you quote is insufficient to determine: COLOR may be allowed to be
> >> <nothing>. Iirc the reasoning went in particular towards a range with
> >> merely the upper end specified being something (halfway) meaningful.
> >
> > Right, I see your point.
> >
> > The grammar does not define COLOR explicitly, so it does not by itself
> > prove that an empty token is invalid. I was implicitly reading COLOR as
> > a numeric color value, partly because all examples seem to use numeric
> > values, but I agree that this is not stated there.
> >
> > If this behavior is intentional, then I agree the second patch should
> > not go in as-is.
>
> Just to mention: Something like "-5" won't be interpreted as "0-5" even
> right now. Instead it's taken as a single color with value -5U, afaict.

I don't think that is what happens with the current parser.

I tested this without the patches from this series:

(XEN) Command line: dom0_mem=2048M console=dtuart dtuart=serial0
(XEN) loglvl=all console_timestamps=boot llc-coloring=on
(XEN) xen-llc-colors=-5
...
(XEN) LLC coloring info:
(XEN) Number of LLC colors supported: 32
(XEN) Xen LLC colors (6): { 0-5 }

So "-5" is currently interpreted as "0-5", not as a single color with
value -5U.

That seems to happen because simple_strtoul() does not consume the
leading '-', so start remains 0 and the parser then takes the range
path.

Best regards,
Mykola