linux.git/tools/testing/selftests/vDSO, branch v6.16

RISC-V: vDSO: Wire up getrandom() vDSO implementation

2025-06-05T21:03:09+00:00

Hook up the generic vDSO implementation to the generic vDSO getrandom
implementation by providing the required __arch_chacha20_blocks_nostack
and getrandom_syscall implementations. Also wire up the selftests.

The benchmark result:

	vdso: 25000000 times in 2.466341333 seconds
	libc: 25000000 times in 41.447720005 seconds
	syscall: 25000000 times in 41.043926672 seconds

	vdso: 25000000 x 256 times in 162.286219353 seconds
	libc: 25000000 x 256 times in 2953.855018685 seconds
	syscall: 25000000 x 256 times in 2796.268546000 seconds

[ alex: - Fix dynamic relocation
        - Squash Nathan's fix https://lore.kernel.org/all/20250423-riscv-fix-compat_vdso-lld-v2-1-b7bbbc244501@kernel.org/
	- Add comment from Loongarch ]

Signed-off-by: Xi Ruoyao 
Link: https://lore.kernel.org/r/20250411024600.16045-1-xry111@xry111.site
Tested-by: Alexandre Ghiti 
Signed-off-by: Alexandre Ghiti 
Signed-off-by: Palmer Dabbelt

Merge tag 'timers-vdso-2025-03-23' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2025-03-25T18:30:42+00:00

Pull VDSO infrastructure updates from Thomas Gleixner:

 - Consolidate the VDSO storage

   The VDSO data storage and data layout has been largely architecture
   specific for historical reasons. That increases the maintenance
   effort and causes inconsistencies over and over.

   There is no real technical reason for architecture specific layouts
   and implementations. The architecture specific details can easily be
   integrated into a generic layout, which also reduces the amount of
   duplicated code for managing the mappings.

   Convert all architectures over to a unified layout and common mapping
   infrastructure. This splits the VDSO data layout into subsystem
   specific blocks, timekeeping, random and architecture parts, which
   provides a better structure and allows to improve and update the
   functionalities without conflict and interaction.

 - Rework the timekeeping data storage

   The current implementation is designed for exposing system
   timekeeping accessors, which was good enough at the time when it was
   designed.

   PTP and Time Sensitive Networking (TSN) change that as there are
   requirements to expose independent PTP clocks, which are not related
   to system timekeeping.

   Replace the monolithic data storage by a structured layout, which
   allows to add support for independent PTP clocks on top while reusing
   both the data structures and the time accessor implementations.

* tag 'timers-vdso-2025-03-23' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (55 commits)
  sparc/vdso: Always reject undefined references during linking
  x86/vdso: Always reject undefined references during linking
  vdso: Rework struct vdso_time_data and introduce struct vdso_clock
  vdso: Move architecture related data before basetime data
  powerpc/vdso: Prepare introduction of struct vdso_clock
  arm64/vdso: Prepare introduction of struct vdso_clock
  x86/vdso: Prepare introduction of struct vdso_clock
  time/namespace: Prepare introduction of struct vdso_clock
  vdso/namespace: Rename timens_setup_vdso_data() to reflect new vdso_clock struct
  vdso/vsyscall: Prepare introduction of struct vdso_clock
  vdso/gettimeofday: Prepare helper functions for introduction of struct vdso_clock
  vdso/gettimeofday: Prepare do_coarse_timens() for introduction of struct vdso_clock
  vdso/gettimeofday: Prepare do_coarse() for introduction of struct vdso_clock
  vdso/gettimeofday: Prepare do_hres_timens() for introduction of struct vdso_clock
  vdso/gettimeofday: Prepare do_hres() for introduction of struct vdso_clock
  vdso/gettimeofday: Prepare introduction of struct vdso_clock
  vdso/helpers: Prepare introduction of struct vdso_clock
  vdso/datapage: Define vdso_clock to prepare for multiple PTP clocks
  vdso: Make vdso_time_data cacheline aligned
  arm64: Make asm/cache.h compatible with vDSO
  ...

selftests/vDSO: Fix GNU hash table entry size for s390x

2025-03-04T16:15:18+00:00

Commit 14be4e6f3522 ("selftests: vDSO: fix ELF hash table entry size for s390x")
changed the type of the ELF hash table entries to 64bit on s390x.
However the *GNU* hash tables entries are always 32bit.
The "bucket" pointer is shared between both hash algorithms.
On s390, this caused the GNU hash algorithm to access its 32-bit entries as if they
were 64-bit, triggering compiler warnings (assignment between "Elf64_Xword *" and
"Elf64_Word *") and runtime crashes.

Introduce a new dedicated "gnu_bucket" pointer which is used by the GNU hash.

Fixes: e0746bde6f82 ("selftests/vDSO: support DT_GNU_HASH")
Reviewed-by: Jens Remus 
Signed-off-by: Thomas Weißschuh 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/r/20250217-selftests-vdso-s390-gnu-hash-v2-1-f6c2532ffe2a@linutronix.de
Signed-off-by: Vasily Gorbik

selftests: vDSO: vdso_standalone_test_x86: Switch to nolibc

2025-03-03T19:00:13+00:00

vdso_standalone_test_x86 provides its own ASM syscall wrappers and
_start() implementation. The in-tree nolibc library already provides
this functionality for multiple architectures. By making use of nolibc,
the standalone testcase can be built from the exact same codebase as the
non-standalone version.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-16-28e14e031ed8@linutronix.de

selftests: vDSO: vdso_test_gettimeofday: Make compatible with nolibc

2025-03-03T19:00:13+00:00

nolibc does not provide sys/time.h and sys/auxv.h,
instead their definitions are available unconditionally.

Guard the includes so they are not attempted on nolibc.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-15-28e14e031ed8@linutronix.de

selftests: vDSO: vdso_test_gettimeofday: Clean up includes

2025-03-03T19:00:13+00:00

Some unnecessary headers are included, remove them.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-14-28e14e031ed8@linutronix.de

selftests: vDSO: parse_vdso: Test __SIZEOF_LONG__ instead of ULONG_MAX

2025-03-03T19:00:13+00:00

According to limits.h(2) ULONG_MAX is only guaranteed to expand to an
expression, not a symbolic constant which can be evaluated by the
preprocessor.

Specifically the definition of ULONG_MAX from nolibc can not be evaluated
by the preprocessor. To provide compatibility with nolibc, check with
__SIZEOF_LONG__ instead, with is provided directly by the preprocessor
and therefore always a symbolic constant.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-13-28e14e031ed8@linutronix.de

selftests: vDSO: parse_vdso: Use UAPI headers instead of libc headers

2025-03-03T19:00:12+00:00

To allow the usage of parse_vdso.c together with a limited libc like
nolibc, use the kernels own elf.h and auxvec.h headers.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-12-28e14e031ed8@linutronix.de

selftests: vDSO: parse_vdso: Drop vdso_init_from_auxv()

2025-03-03T19:00:12+00:00

There are no users left.

This also removes the usage of ElfXX_auxv_t, which is not formally
standardized.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-11-28e14e031ed8@linutronix.de

selftests: vDSO: vdso_standalone_test_x86: Use vdso_init_form_sysinfo_ehdr

2025-03-03T19:00:12+00:00

vdso_standalone_test_x86 is the only user of vdso_init_from_auxv().
Instead of combining the parsing the aux vector with the parsing of the
vDSO, split them apart into getauxval() and the regular
vdso_init_from_sysinfo_ehdr().

The implementation of getauxval() is taken from
tools/include/nolibc/stdlib.h.

Signed-off-by: Thomas Weißschuh 
Signed-off-by: Thomas Gleixner 
Reviewed-by: Vincenzo Frascino 
Acked-by: Shuah Khan 
Link: https://lore.kernel.org/all/20250226-parse_vdso-nolibc-v2-10-28e14e031ed8@linutronix.de