On Tue, 17 Mar 2026 14:56:47 +0000 David Laight <[email protected]> wrote:
> On Tue, 17 Mar 2026 00:07:57 +0100 > "Arnd Bergmann" <[email protected]> wrote: > > > On Mon, Mar 16, 2026, at 23:51, Matt Roper wrote: > > > On Mon, Mar 16, 2026 at 11:43:46PM +0100, Arnd Bergmann wrote: > > >> From: Arnd Bergmann <[email protected]> > > >> > > >> ptrdiff_t, size_t and long are the same size on all supported > > >> architectures, > > >> but gcc requires the use of the %zx modifier instead of %lx. On 32-bit > > >> > > > > > > Nathan Chancellor just sent a fix here: > > > > > > https://lore.kernel.org/all/20260316-drm-xe-fix-32-bit-wformat-ptrdiff-v1-1-0108b10b2...@kernel.org/ > > > > > > His solution uses %tx rather than %zx, which according to > > > Documentation/core-api/printk-formats.rst sounds like it's probably the > > > most accurate format string for this case? > > > > > > > Right, Nathan's version is correct. I wasn't aware that size_t > > and ptrdiff_t have different modifiers. > > Consider: > void foo(size_t len) > { > char *p = malloc(len); > char *end = p + len; > ptrdiff_t diff = p - end; > printf("len %zu, diff %td\n", len, diff); > free(p); > } > Clearly foo(5) output "5 -5". > But what about foo(5u << 29) on a 32bit system with a large unlimit and > a big enough hole in the address space. > More likely would be something running on an x86 in real mode. > There malloc(65000) might be reasonable, so size_t would be unsigned int > but ptrdiff_t would need to be a signed 32bit type. > > Of course, in linux, they are all 'long'. I should have said 'the same size as long'. > > David > > > > > > Arnd > > > >
