Re: [PATCH v1 1/1] lib/vsprintf: refactor duplicate code to xnumber()

[Andy Shevchenko]

On Tue, 2015-12-29 at 00:20 +0200, Andy Shevchenko wrote:
> On Mon, Dec 28, 2015 at 11:42 PM, Rasmus Villemoes
> <linux@xxxxxxxxxxxxxxxxxx> wrote:
> > On Mon, Dec 28 2015, Joe Perches <joe@xxxxxxxxxxx> wrote:
> > 
> > > On Mon, 2015-12-28 at 20:18 +0200, Andy Shevchenko wrote:
> > > > xnumber() is a special helper to print a fixed size type in a
> > > > hex format with
> > > > '0x' prefix with padding and reduced size. In the module we
> > > > have already
> > > > several copies of such code. Consolidate them under xnumber()
> > > > helper.
> > > > 
> > > > There are couple of differences though.
> > > > 
> > > > It seems nobody cared about the output in case of
> > > > CONFIG_KALLSYMS=n when
> > > > printing symbol address because the asked width is not enough
> > > > to care either
> > > > prefix or last byte. Fixed here.
> > 
> > ok, though I'm curious what 'last byte' refers to here?
> 
> The last byte ('78') as it appears in the string carrying the number
> '0x12345678'. Yeah, might be confusing, I'm open for suggestion how
> to
> phrase it.
> 
> > 
> > > > The %pNF specifier used to be allowed with a specific field
> > > > width, though there
> > > > is neither any user of it nor mention in the documentation.
> > > > 
> > > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxx
> > > > om>
> > > > ---
> > > > Âlib/vsprintf.c | 43 +++++++++++++++---------------------------
> > > > -
> > > > Â1 file changed, 15 insertions(+), 28 deletions(-)
> > > > 
> > > > diff --git a/lib/vsprintf.c b/lib/vsprintf.c
> > > > index dcf5646..e971549 100644
> > > > --- a/lib/vsprintf.c
> > > > +++ b/lib/vsprintf.c
> > > > @@ -514,6 +514,16 @@ char *number(char *buf, char *end,
> > > > unsigned long long num,
> > > > ÂÂÂÂÂreturn buf;
> > > > Â}
> > > > 
> > > > +static noinline_for_stack
> > > > +char *xnumber(char *buf, char *end, unsigned long long value,
> > > > unsigned int type,
> > > > +ÂÂÂÂÂÂÂÂÂÂstruct printf_spec spec)
> > 
> > Is there any aspect of the passed-through printf_spec which isn't
> > overridden in xnumber? The users are/will be various %p extensions,
> > which probably means that no-one passes a non-default precision
> > (gcc
> > complains about %.*p), and the remaining possible flags (PLUS,
> > LEFT,
> > SPACE) are useless and/or impossible to pass to %p without gcc
> > complaining. In other words, why pass the spec at all instead of
> > just
> > building it inside xnumber?
> 
> Wow, good catch!
> I slightly suspected something like that, but didn't made up my mind
> to check this.
> 
> > 
> > > xnumber isn't a great name.
> > 
> > Maybe 'hexnumber'.
> 
> We already have similar for %*ph. And as you noticed belowâ
> 
> > That's a bit further away from 'number', and 'x'
> > might stand for something other than hex.
> 
> âisn't only about hex. I don't know how to play on words the all
> three
> flags including 16 base.
> 
> > 
> > > unsigned int type should probably be size_t size
> > 
> > Compromise: 'unsigned int size'. The name should be size since it's
> > supposed to be the size of the actual type being printed. But the
> > type
> > carrying that information need not be 8 bytes wide on 64bits.
> 
> Exactly, the result anyway as for now only 8 bits as a part of 
> unsigned int.

Oops, 24 bits of signed int. Incorrectly caught wrong line.
So, I will change this to be int size then.

> 
> > 
> > > > static noinline_for_stack
> > > > Âchar *address_val(char *buf, char *end, const void *addr,
> > > > ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂstruct printf_spec spec, const char *fmt)
> > > > Â{
> > > > -ÂÂÂÂunsigned long long num;
> > > > -
> > > > -ÂÂÂÂspec.flags |= SPECIAL | SMALL | ZEROPAD;
> > > > -ÂÂÂÂspec.base = 16;
> > > > -
> > > > ÂÂÂÂÂswitch (fmt[1]) {
> > > > ÂÂÂÂÂcase 'd':
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂnum = *(const dma_addr_t *)addr;
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂspec.field_width = sizeof(dma_addr_t) * 2 + 2;
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂbreak;
> > > > +ÂÂÂÂÂÂÂÂÂÂÂÂreturn xnumber(buf, end, *(const dma_addr_t
> > > > *)addr, sizeof(dma_addr_t), spec);
> > > > ÂÂÂÂÂcase 'p':
> > > > ÂÂÂÂÂdefault:
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂnum = *(const phys_addr_t *)addr;
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂspec.field_width = sizeof(phys_addr_t) * 2 + 2;
> > > > -ÂÂÂÂÂÂÂÂÂÂÂÂbreak;
> > > > +ÂÂÂÂÂÂÂÂÂÂÂÂreturn xnumber(buf, end, *(const phys_addr_t
> > > > *)addr, sizeof(phys_addr_t), spec);
> > > > ÂÂÂÂÂ}
> > > > -
> > > > -ÂÂÂÂreturn number(buf, end, num, spec);
> > > > Â}
> > 
> > Nit: I think it would be a bit easier to read if the
> > cast+dereference
> > are kept outside the function calls. I'd suggest just introducing
> > 'unsigned int size', assign the appropriate value in the two cases,
> > and
> > fall through to a common 'xnumber(buf, end, num, size);'. It'll
> > even
> > line up nicely ;-)
> 
> Will try that.
> 
> > 
> > num = *(const dma_addr_t *)addr;
> > size = sizeof(dma_addr_t);
> 
> Thanks, Rasmus, for review.
> 

-- 
Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
Intel Finland Oy

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/