Re: [PATCH v3 1/2] s390/ebcdic: Add character classes for the invariant subset of EBCDIC

[Heiko Carstens <hca@linux.ibm.com>]

On Mon, Aug 31, 2026 at 02:14:19PM +0200, Ilya Leoshkevich wrote:
> > > +const __u8 _ebctype_inv[256] = {
> > > +	_C,     _C,    _C,    _C,    _C,    _C|_S, _C,    _C,	/* 0x00-0x07 */
> > > +	_C,     _C,    _C,    _C|_S, _C|_S, _C|_S, _C,    _C,	/* 0x08-0x0f */
> > > +	_C,     _C,    _C,    _C,    _C,    _C|_S, _C,    _C,	/* 0x10-0x17 */
> > > +	_C,     _C,    _C,    _C,    _C,    _C,    _C,    _C,	/* 0x18-0x1f */
> > > +	_C,     _C,    _C,    _C,    _C,    _C|_S, _C,    _C,	/* 0x20-0x27 */
> > > +	_C,     _C,    _C,    _C,    _C,    _C,    _C,    _C,	/* 0x28-0x2f */
> > > +	0,      0,     _C,    _C,    _C,    _C,    _C,    _C,	/* 0x30-0x37 */
> > > +	_C,     _C,    _C,    _C,    _C,    _C,    0,     _C,	/* 0x38-0x3f */
> > > +	_S|_SP, 0,     0,     0,     0,     0,     0,     0,	/* 0x40-0x47 */
> > > +	0,      0,     _P,    _P,    _P,    _P,    _P,    _P,	/* 0x48-0x4f */
> > 
> > [Severity: Medium]
> > Does this array incorrectly include variant EBCDIC characters?
> > 
> > 0x4A is mapped as _P (invariant punctuation) here, but isn't it a variant
> > character (e.g., '¢' in US 037, '[' in German 273)?

I guess this is indeed wrong, and 0x4a should be mapped to "0"?

> > > +	_P,     _U|_X, _U|_X, _U|_X, _U|_X, _U|_X, _U|_X, _U,	/* 0xc0-0xc7 */
> > 
> > [Severity: Medium]
> > Is 0xC0 also a variant character?
> > 
> > It is classified as _P here, but it can represent '{' in US 037 or 'Ä'
> > in German 273.
> > 
> > This seems to violate the definition of the invariant subset and could lead
> > to missed warnings for garbled boot parameters.
> 
> This is a naming and not a functional issue. The IBM037 punctuation
> marks must be present in the table for the series to make sense.
> 
> I guess a 100% proper name here would be something along the lines of
> "invariant with IBM037 punctuation" or "IBM037 without weird
> characters", but I don't think it's that important: one can look up
> the Wikipedia reference in the commit message to get to the exact table
> I used here.
> 
> I will defer it to the maintainers, whether I need to change comments,
> names, or the commit message here.

For 0xc0 I don't see a problem - it matches what we have in lib/ctype.c, which
also maps "{" to _P.

That only leaves the question above about 0x4a.