[igt-dev] [PATCH i-g-t v2 2/2] build: provide stub implementation for memfd_create

Rodrigo Vivi rodrigo.vivi at intel.com
Wed Jul 25 22:56:40 UTC 2018


On Wed, Jul 25, 2018 at 02:28:31PM -0700, Antonio Argenziano wrote:
> 
> 
> On 25/07/18 12:57, De Marchi, Lucas wrote:
> > On Wed, 2018-07-25 at 09:46 -0700, Antonio Argenziano wrote:
> > > 
> > > On 24/07/18 15:20, Lucas De Marchi wrote:
> > > > When libc misses memfd_create(), provide a stub implementation to go
> > > > through the syscall() route. Syscall numbers are provided for platforms
> > > > currently supported by i-g-t only.
> > > > 
> > > > v2: add support to autotools
> > > > 
> > > > Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
> > > > ---
> > > >    configure.ac                  |  3 +++
> > > >    lib/stubs/syscalls/sys/mman.h | 37 +++++++++++++++++++++++++++++++++++
> > > >    meson.build                   |  3 +++
> > > 
> > > I thought you were going to remove "#include <linux/memfd.h>" from
> > > gem_userptr_blits and include the stub instead.
> > 
> > I decided not to. To summarize:
> > 
> > In order to get the memfd_create() function definition one needs to include
> > sys/mman.h. The man page is wrong and I sent a patch.
> > 
> > linux/memfd.h is the kernel header (i.e. not from libc). It's useful to have
> > it if we are using or going to use new flags not present in the libc headers.
> > Or harmless. This is their content for me:
> > 
> > kernel header (linux/memfd.h):
> > /* flags for memfd_create(2) (unsigned int) */
> > #define MFD_CLOEXEC             0x0001U
> > #define MFD_ALLOW_SEALING       0x0002U
> > #define MFD_HUGETLB             0x0004U
> > 
> > libc header (bits/mman-shared.h which is down the include chain for
> > sys/mman.h):
> > 
> > #ifdef __USE_GNU
> > /* Flags for memfd_create.  */
> > # ifndef MFD_CLOEXEC
> > #  define MFD_CLOEXEC 1U
> > #  define MFD_ALLOW_SEALING 2U
> > #  define MFD_HUGETLB 4U
> > # endif
> > ...
> > 
> > We could also check if linux/memfd.h is available or just remove the include
> > since we are not using any flag not available on libc, but IMO it's not
> > necessary or at least not for this patch.
> 
> Fair enough, time for me to modernize a little anyway :).
> 
> For both patches:
> Acked-by: Antonio Argenziano <antonio.argenziano at intel.com>

pushed, thanks

> 
> > 
> > 
> > Lucas De Marchi
> > 
> > > 
> > > Antonio.
> > > 
> > > >    3 files changed, 43 insertions(+)
> > > >    create mode 100644 lib/stubs/syscalls/sys/mman.h
> > > > 
> > > > diff --git a/configure.ac b/configure.ac
> > > > index 0a5b0425..416a3240 100644
> > > > --- a/configure.ac
> > > > +++ b/configure.ac
> > > > @@ -81,6 +81,9 @@ AC_CHECK_FUNCS(timer_create, [], [
> > > >    ])
> > > >    AC_SUBST(TIMER_LIBS)
> > > > +dnl Check for memfd_create
> > > > +AC_CHECK_FUNCS(memfd_create)
> > > > +
> > > >    dnl Check for CPUID
> > > >    cpuid="yes"
> > > >    AC_TRY_LINK([
> > > > diff --git a/lib/stubs/syscalls/sys/mman.h b/lib/stubs/syscalls/sys/mman.h
> > > > new file mode 100644
> > > > index 00000000..2ac2da6a
> > > > --- /dev/null
> > > > +++ b/lib/stubs/syscalls/sys/mman.h
> > > > @@ -0,0 +1,37 @@
> > > > +/* SPDX-License-Identifier: MIT */
> > > > +
> > > > +#pragma once
> > > > +
> > > > +#include_next <sys/mman.h>
> > > > +
> > > > +#if !defined(HAVE_MEMFD_CREATE) || !HAVE_MEMFD_CREATE
> > > > +#include <errno.h>
> > > > +#include <sys/syscall.h>
> > > > +#include <sys/types.h>
> > > > +#include <unistd.h>
> > > > +
> > > > +#ifndef __NR_memfd_create
> > > > +#if defined __x86_64__
> > > > +#define __NR_memfd_create 319
> > > > +#elif defined __i386__
> > > > +#define __NR_memfd_create 356
> > > > +#elif defined __arm__
> > > > +#define __NR_memfd_create 385
> > > > +#else
> > > > +#warning "__NR_memfd_create unknown for your architecture"
> > > > +#endif
> > > > +#endif
> > > > +
> > > > +static inline int missing_memfd_create(const char *name, unsigned int
> > > > flags)
> > > > +{
> > > > +#ifdef __NR_memfd_create
> > > > +	return syscall(__NR_memfd_create, name, flags);
> > > > +#else
> > > > +	errno = ENOSYS;
> > > > +	return -1;
> > > > +#endif
> > > > +}
> > > > +
> > > > +#define memfd_create missing_memfd_create
> > > > +
> > > > +#endif
> > > > diff --git a/meson.build b/meson.build
> > > > index 22118828..63ac6292 100644
> > > > --- a/meson.build
> > > > +++ b/meson.build
> > > > @@ -196,6 +196,9 @@ if cc.has_member('struct sysinfo', 'totalram',
> > > >    	config.set('HAVE_STRUCT_SYSINFO_TOTALRAM', 1)
> > > >    endif
> > > > +have = cc.has_function('memfd_create', prefix : '''#include
> > > > <sys/mman.h>''', args : '-D_GNU_SOURCE')
> > > > +config.set10('HAVE_MEMFD_CREATE', have)
> > > > +
> > > >    add_project_arguments('-D_GNU_SOURCE', language : 'c')
> > > >    add_project_arguments('-include', 'config.h', language : 'c')
> > > > 
> _______________________________________________
> igt-dev mailing list
> igt-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/igt-dev


More information about the igt-dev mailing list