powerpc: Add HWCAP3/HWCAP4 data to TCB for Power Architecture.
Checks
Commit Message
This patch adds a new feature for powerpc. In order to get faster
access to the HWCAP3/HWCAP4 masks, similar to HWCAP/HWCAP2 (i.e. for
implementing __builtin_cpu_supports() in GCC) without the overhead of
reading them from the auxiliary vector, we now reserve space for them
in the TCB.
Versioned symbol '__parse_hwcap_3_4_and_convert_at_platform' advertises
the availability of the new HWCAP3/HWCAP4 data in the TCB.
This is an ABI change for GLIBC 2.39.
Suggested-by: Peter Bergner <bergner@linux.ibm.com>
Comments
On 11/30/23 10:07, Manjunath Matti wrote:
> This patch adds a new feature for powerpc. In order to get faster
> access to the HWCAP3/HWCAP4 masks, similar to HWCAP/HWCAP2 (i.e. for
> implementing __builtin_cpu_supports() in GCC) without the overhead of
> reading them from the auxiliary vector, we now reserve space for them
> in the TCB.
>
> Versioned symbol '__parse_hwcap_3_4_and_convert_at_platform' advertises
> the availability of the new HWCAP3/HWCAP4 data in the TCB.
>
> This is an ABI change for GLIBC 2.39.
>
> Suggested-by: Peter Bergner <bergner@linux.ibm.com>
Please note this patch fails pre-commit CI.
It doesn't apply.
https://patchwork.sourceware.org/project/glibc/patch/20231130150740.1195563-1-mmatti@linux.ibm.com/
If a patch depends on anoher patch is should be sent as a series.
> diff --git a/elf/dl-support.c b/elf/dl-support.c
> index 837fa1c836..ff9560ce72 100644
> --- a/elf/dl-support.c
> +++ b/elf/dl-support.c
> @@ -158,6 +158,8 @@ const ElfW(Phdr) *_dl_phdr;
> size_t _dl_phnum;
> uint64_t _dl_hwcap;
> uint64_t _dl_hwcap2;
> +uint64_t _dl_hwcap3;
> +uint64_t _dl_hwcap4;
>
> enum dso_sort_algorithm _dl_dso_sort_algo;
>
> diff --git a/elf/elf.h b/elf/elf.h
> index 5c1c1972d1..4f7cb7385f 100644
> --- a/elf/elf.h
> +++ b/elf/elf.h
> @@ -1234,6 +1234,10 @@ typedef struct
> #define AT_RSEQ_FEATURE_SIZE 27 /* rseq supported feature size. */
> #define AT_RSEQ_ALIGN 28 /* rseq allocation alignment. */
>
> +/* More machine-dependent hints about processor capabilities. */
> +#define AT_HWCAP3 29 /* extension of AT_HWCAP. */
> +#define AT_HWCAP4 30 /* extension of AT_HWCAP. */
> +
> #define AT_EXECFN 31 /* Filename of executable. */
>
> /* Pointer to the global system page used for system calls and other
> diff --git a/sysdeps/generic/ldsodefs.h b/sysdeps/generic/ldsodefs.h
> index 9b50ddd09f..4c7b60d8e1 100644
> --- a/sysdeps/generic/ldsodefs.h
> +++ b/sysdeps/generic/ldsodefs.h
> @@ -646,6 +646,8 @@ struct rtld_global_ro
> /* Mask for more hardware capabilities that are available on some
> platforms. */
> EXTERN uint64_t _dl_hwcap2;
> + EXTERN uint64_t _dl_hwcap3;
> + EXTERN uint64_t _dl_hwcap4;
>
> EXTERN enum dso_sort_algorithm _dl_dso_sort_algo;
>
> diff --git a/sysdeps/powerpc/Versions b/sysdeps/powerpc/Versions
> index cca8fd2fc5..575f7317bf 100644
> --- a/sysdeps/powerpc/Versions
> +++ b/sysdeps/powerpc/Versions
> @@ -24,4 +24,9 @@ ld {
> # and AT_PLATFORM data should be stored into the TCB.
> __parse_hwcap_and_convert_at_platform;
> }
> + GLIBC_2.39 {
> + # Symbol used to version control when the ABI started to specify that
> + # HWCAP3 and HWCAP4 are stored in the TCB.
> + __parse_hwcap_3_4_and_convert_at_platform;
> + }
> }
> diff --git a/sysdeps/powerpc/dl-procinfo.c b/sysdeps/powerpc/dl-procinfo.c
> index 6b9dad24d8..376a4b6118 100644
> --- a/sysdeps/powerpc/dl-procinfo.c
> +++ b/sysdeps/powerpc/dl-procinfo.c
> @@ -38,6 +38,10 @@
> needed.
> */
>
> +/* The total number of available bits (including those prior to
> + _DL_HWCAP_FIRST). Some of these bits might not be used. */
> +#define _DL_HWCAP_COUNT 128
> +
> #ifndef PROCINFO_CLASS
> # define PROCINFO_CLASS
> #endif
> @@ -61,7 +65,7 @@ PROCINFO_CLASS struct cpu_features _dl_powerpc_cpu_features
> #if !defined PROCINFO_DECL && defined SHARED
> ._dl_powerpc_cap_flags
> #else
> -PROCINFO_CLASS const char _dl_powerpc_cap_flags[64][15]
> +PROCINFO_CLASS const char _dl_powerpc_cap_flags[_DL_HWCAP_COUNT][15]
> #endif
> #ifndef PROCINFO_DECL
> = {
> diff --git a/sysdeps/powerpc/dl-procinfo.h b/sysdeps/powerpc/dl-procinfo.h
> index 641eb54380..f8c2a52c31 100644
> --- a/sysdeps/powerpc/dl-procinfo.h
> +++ b/sysdeps/powerpc/dl-procinfo.h
> @@ -22,16 +22,17 @@
> #include <ldsodefs.h>
> #include <sysdep.h> /* This defines the PPC_FEATURE[2]_* macros. */
>
> -/* The total number of available bits (including those prior to
> - _DL_HWCAP_FIRST). Some of these bits might not be used. */
> -#define _DL_HWCAP_COUNT 64
> +/* Feature masks are all 32-bits in size. */
> +#define _DL_HWCAP_SIZE 32
>
> -/* Features started at bit 31 and decremented as new features were added. */
> -#define _DL_HWCAP_LAST 31
> +/* AT_HWCAP2 feature strings follow the AT_HWCAP feature strings. */
> +#define _DL_HWCAP2_OFFSET _DL_HWCAP_SIZE
>
> -/* AT_HWCAP2 features started at bit 31 and decremented as new features were
> - added. HWCAP2 feature bits start at bit 0. */
> -#define _DL_HWCAP2_LAST 31
> +/* AT_HWCAP3 feature strings follow the AT_HWCAP2 feature strings. */
> +#define _DL_HWCAP3_OFFSET (_DL_HWCAP2_OFFSET + _DL_HWCAP_SIZE)
> +
> +/* AT_HWCAP4 feature strings follow the AT_HWCAP3 feature strings. */
> +#define _DL_HWCAP4_OFFSET (_DL_HWCAP3_OFFSET + _DL_HWCAP_SIZE)
>
> /* These bits influence library search. */
> #define HWCAP_IMPORTANT (PPC_FEATURE_HAS_ALTIVEC \
> @@ -187,21 +188,42 @@ _dl_procinfo (unsigned int type, unsigned long int word)
> case AT_HWCAP:
> _dl_printf ("AT_HWCAP: ");
>
> - for (int i = 0; i <= _DL_HWCAP_LAST; ++i)
> + for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
> if (word & (1 << i))
> _dl_printf (" %s", _dl_hwcap_string (i));
> break;
> case AT_HWCAP2:
> {
> - unsigned int offset = _DL_HWCAP_LAST + 1;
>
> _dl_printf ("AT_HWCAP2: ");
>
> - /* We have to go through them all because the kernel added the
> - AT_HWCAP2 features starting with the high bits. */
> - for (int i = 0; i <= _DL_HWCAP2_LAST; ++i)
> - if (word & (1 << i))
> - _dl_printf (" %s", _dl_hwcap_string (offset + i));
> + /* We have to go through them all because the kernel added the
> + AT_HWCAP2 features starting with the high bits. */
> + for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
> + if (word & (1 << i))
> + _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP2_OFFSET + i));
> + break;
> + }
> + case AT_HWCAP3:
> + {
> + _dl_printf ("AT_HWCAP3: ");
> +
> + /* We have to go through them all because the kernel added the
> + AT_HWCAP3 features starting with the high bits. */
> + for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
> + if (word & (1 << i))
> + _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP3_OFFSET + i));
> + break;
> + }
> + case AT_HWCAP4:
> + {
> + _dl_printf ("AT_HWCAP4: ");
> +
> + /* We have to go through them all because the kernel added the
> + AT_HWCAP4 features starting with the high bits. */
> + for (int i = 0; i <= _DL_HWCAP_SIZE; ++i)
> + if (word & (1 << i))
> + _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP4_OFFSET + i));
> break;
> }
> case AT_L1I_CACHEGEOMETRY:
> diff --git a/sysdeps/powerpc/hwcapinfo.c b/sysdeps/powerpc/hwcapinfo.c
> index bf46a4c84d..9db9fa2c18 100644
> --- a/sysdeps/powerpc/hwcapinfo.c
> +++ b/sysdeps/powerpc/hwcapinfo.c
> @@ -31,7 +31,7 @@ void
> __tcb_parse_hwcap_and_convert_at_platform (void)
> {
>
> - uint64_t h1, h2;
> + uint64_t h1, h2, h3, h4;
>
> /* Read AT_PLATFORM string from auxv and convert it to a number. */
> __tcb.at_platform = _dl_string_platform (GLRO (dl_platform));
> @@ -39,6 +39,8 @@ __tcb_parse_hwcap_and_convert_at_platform (void)
> /* Read HWCAP and HWCAP2 from auxv. */
> h1 = GLRO (dl_hwcap);
> h2 = GLRO (dl_hwcap2);
> + h3 = GLRO (dl_hwcap3);
> + h4 = GLRO (dl_hwcap4);
>
> /* hwcap contains only the latest supported ISA, the code checks which is
> and fills the previous supported ones. */
> @@ -64,18 +66,23 @@ __tcb_parse_hwcap_and_convert_at_platform (void)
> else if (h1 & PPC_FEATURE_POWER5)
> h1 |= PPC_FEATURE_POWER4;
>
> - uint64_t array_hwcaps[] = { h1, h2 };
> + uint64_t array_hwcaps[] = { h1, h2, h3, h4 };
> init_cpu_features (&GLRO(dl_powerpc_cpu_features), array_hwcaps);
>
> /* Consolidate both HWCAP and HWCAP2 into a single doubleword so that
> we can read both in a single load later. */
> __tcb.hwcap = (h1 << 32) | (h2 & 0xffffffff);
> - __tcb.hwcap_extn = 0x0;
> +
> + /* Consolidate both HWCAP3 and HWCAP4 into a single doubleword so that
> + we can read both in a single load later. */
> + __tcb.hwcap_extn = (h3 << 32) | (h4 & 0xffffffff);
>
> }
> #if IS_IN (rtld)
> versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
> __parse_hwcap_and_convert_at_platform, GLIBC_2_23);
> +versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
> + __parse_hwcap_3_4_and_convert_at_platform, GLIBC_2_39);
> #endif
>
> /* Export __parse_hwcap_and_convert_at_platform in libc.a. This is used by
> @@ -84,4 +91,6 @@ versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
> #ifndef SHARED
> weak_alias (__tcb_parse_hwcap_and_convert_at_platform, \
> __parse_hwcap_and_convert_at_platform);
> +weak_alias (__tcb_parse_hwcap_and_convert_at_platform, \
> + __parse_hwcap_3_4_and_convert_at_platform);
> #endif
> diff --git a/sysdeps/unix/sysv/linux/dl-parse_auxv.h b/sysdeps/unix/sysv/linux/dl-parse_auxv.h
> index cf5e81bf2c..74e95814cd 100644
> --- a/sysdeps/unix/sysv/linux/dl-parse_auxv.h
> +++ b/sysdeps/unix/sysv/linux/dl-parse_auxv.h
> @@ -47,6 +47,8 @@ void _dl_parse_auxv (ElfW(auxv_t) *av, dl_parse_auxv_t auxv_values)
> GLRO(dl_platform) = (void *) auxv_values[AT_PLATFORM];
> GLRO(dl_hwcap) = auxv_values[AT_HWCAP];
> GLRO(dl_hwcap2) = auxv_values[AT_HWCAP2];
> + GLRO(dl_hwcap3) = auxv_values[AT_HWCAP3];
> + GLRO(dl_hwcap4) = auxv_values[AT_HWCAP4];
> GLRO(dl_clktck) = auxv_values[AT_CLKTCK];
> GLRO(dl_fpu_control) = auxv_values[AT_FPUCW];
> _dl_random = (void *) auxv_values[AT_RANDOM];
> diff --git a/sysdeps/unix/sysv/linux/dl-sysdep.c b/sysdeps/unix/sysv/linux/dl-sysdep.c
> index 1b3dd869b5..e497206602 100644
> --- a/sysdeps/unix/sysv/linux/dl-sysdep.c
> +++ b/sysdeps/unix/sysv/linux/dl-sysdep.c
> @@ -197,6 +197,8 @@ _dl_show_auxv (void)
> [AT_SYSINFO_EHDR - 2] = { "SYSINFO_EHDR: 0x", hex },
> [AT_RANDOM - 2] = { "RANDOM: 0x", hex },
> [AT_HWCAP2 - 2] = { "HWCAP2: 0x", hex },
> + [AT_HWCAP3 - 2] = { "HWCAP3: 0x", hex },
> + [AT_HWCAP4 - 2] = { "HWCAP4: 0x", hex },
> [AT_MINSIGSTKSZ - 2] = { "MINSIGSTKSZ: ", dec },
> [AT_L1I_CACHESIZE - 2] = { "L1I_CACHESIZE: ", dec },
> [AT_L1I_CACHEGEOMETRY - 2] = { "L1I_CACHEGEOMETRY: 0x", hex },
> diff --git a/sysdeps/unix/sysv/linux/powerpc/cpu-features.c b/sysdeps/unix/sysv/linux/powerpc/cpu-features.c
> index 7c6e20e702..7c39e97cc6 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/cpu-features.c
> +++ b/sysdeps/unix/sysv/linux/powerpc/cpu-features.c
> @@ -113,6 +113,8 @@ init_cpu_features (struct cpu_features *cpu_features, uint64_t hwcaps[])
> which are set by __tcb_parse_hwcap_and_convert_at_platform. */
> cpu_features->hwcap = hwcaps[0];
> cpu_features->hwcap2 = hwcaps[1];
> + cpu_features->hwcap3 = hwcaps[2];
> + cpu_features->hwcap4 = hwcaps[3];
> /* Default is to use aligned memory access on optimized function unless
> tunables is enable, since for this case user can explicit disable
> unaligned optimizations. */
> diff --git a/sysdeps/unix/sysv/linux/powerpc/cpu-features.h b/sysdeps/unix/sysv/linux/powerpc/cpu-features.h
> index e5fce88e5e..e7bdca9994 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/cpu-features.h
> +++ b/sysdeps/unix/sysv/linux/powerpc/cpu-features.h
> @@ -26,6 +26,8 @@ struct cpu_features
> bool use_cached_memopt;
> unsigned long int hwcap;
> unsigned long int hwcap2;
> + unsigned long int hwcap3;
> + unsigned long int hwcap4;
> };
>
> static const char hwcap_names[] = {
> diff --git a/sysdeps/unix/sysv/linux/powerpc/libc-start.c b/sysdeps/unix/sysv/linux/powerpc/libc-start.c
> index b6aec4615d..34b9bc7c8f 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/libc-start.c
> +++ b/sysdeps/unix/sysv/linux/powerpc/libc-start.c
> @@ -87,6 +87,12 @@ __libc_start_main_impl (int argc, char **argv,
> case AT_HWCAP2:
> _dl_hwcap2 = (unsigned long int) av->a_un.a_val;
> break;
> + case AT_HWCAP3:
> + _dl_hwcap3 = (unsigned long int) av->a_un.a_val;
> + break;
> + case AT_HWCAP4:
> + _dl_hwcap4 = (unsigned long int) av->a_un.a_val;
> + break;
> case AT_PLATFORM:
> _dl_platform = (void *) av->a_un.a_val;
> break;
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> index 5a68aeb9ee..b1073f0942 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/ld.abilist
> @@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
> GLIBC_2.35 __rseq_flags D 0x4
> GLIBC_2.35 __rseq_offset D 0x4
> GLIBC_2.35 __rseq_size D 0x4
> +GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> index 21f472e674..40942a2cc6 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/be/ld.abilist
> @@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
> GLIBC_2.35 __rseq_flags D 0x4
> GLIBC_2.35 __rseq_offset D 0x8
> GLIBC_2.35 __rseq_size D 0x4
> +GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F
> diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> index 9c9c40450d..01f2694a4d 100644
> --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ld.abilist
> @@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
> GLIBC_2.35 __rseq_flags D 0x4
> GLIBC_2.35 __rseq_offset D 0x8
> GLIBC_2.35 __rseq_size D 0x4
> +GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F
@@ -158,6 +158,8 @@ const ElfW(Phdr) *_dl_phdr;
size_t _dl_phnum;
uint64_t _dl_hwcap;
uint64_t _dl_hwcap2;
+uint64_t _dl_hwcap3;
+uint64_t _dl_hwcap4;
enum dso_sort_algorithm _dl_dso_sort_algo;
@@ -1234,6 +1234,10 @@ typedef struct
#define AT_RSEQ_FEATURE_SIZE 27 /* rseq supported feature size. */
#define AT_RSEQ_ALIGN 28 /* rseq allocation alignment. */
+/* More machine-dependent hints about processor capabilities. */
+#define AT_HWCAP3 29 /* extension of AT_HWCAP. */
+#define AT_HWCAP4 30 /* extension of AT_HWCAP. */
+
#define AT_EXECFN 31 /* Filename of executable. */
/* Pointer to the global system page used for system calls and other
@@ -646,6 +646,8 @@ struct rtld_global_ro
/* Mask for more hardware capabilities that are available on some
platforms. */
EXTERN uint64_t _dl_hwcap2;
+ EXTERN uint64_t _dl_hwcap3;
+ EXTERN uint64_t _dl_hwcap4;
EXTERN enum dso_sort_algorithm _dl_dso_sort_algo;
@@ -24,4 +24,9 @@ ld {
# and AT_PLATFORM data should be stored into the TCB.
__parse_hwcap_and_convert_at_platform;
}
+ GLIBC_2.39 {
+ # Symbol used to version control when the ABI started to specify that
+ # HWCAP3 and HWCAP4 are stored in the TCB.
+ __parse_hwcap_3_4_and_convert_at_platform;
+ }
}
@@ -38,6 +38,10 @@
needed.
*/
+/* The total number of available bits (including those prior to
+ _DL_HWCAP_FIRST). Some of these bits might not be used. */
+#define _DL_HWCAP_COUNT 128
+
#ifndef PROCINFO_CLASS
# define PROCINFO_CLASS
#endif
@@ -61,7 +65,7 @@ PROCINFO_CLASS struct cpu_features _dl_powerpc_cpu_features
#if !defined PROCINFO_DECL && defined SHARED
._dl_powerpc_cap_flags
#else
-PROCINFO_CLASS const char _dl_powerpc_cap_flags[64][15]
+PROCINFO_CLASS const char _dl_powerpc_cap_flags[_DL_HWCAP_COUNT][15]
#endif
#ifndef PROCINFO_DECL
= {
@@ -22,16 +22,17 @@
#include <ldsodefs.h>
#include <sysdep.h> /* This defines the PPC_FEATURE[2]_* macros. */
-/* The total number of available bits (including those prior to
- _DL_HWCAP_FIRST). Some of these bits might not be used. */
-#define _DL_HWCAP_COUNT 64
+/* Feature masks are all 32-bits in size. */
+#define _DL_HWCAP_SIZE 32
-/* Features started at bit 31 and decremented as new features were added. */
-#define _DL_HWCAP_LAST 31
+/* AT_HWCAP2 feature strings follow the AT_HWCAP feature strings. */
+#define _DL_HWCAP2_OFFSET _DL_HWCAP_SIZE
-/* AT_HWCAP2 features started at bit 31 and decremented as new features were
- added. HWCAP2 feature bits start at bit 0. */
-#define _DL_HWCAP2_LAST 31
+/* AT_HWCAP3 feature strings follow the AT_HWCAP2 feature strings. */
+#define _DL_HWCAP3_OFFSET (_DL_HWCAP2_OFFSET + _DL_HWCAP_SIZE)
+
+/* AT_HWCAP4 feature strings follow the AT_HWCAP3 feature strings. */
+#define _DL_HWCAP4_OFFSET (_DL_HWCAP3_OFFSET + _DL_HWCAP_SIZE)
/* These bits influence library search. */
#define HWCAP_IMPORTANT (PPC_FEATURE_HAS_ALTIVEC \
@@ -187,21 +188,42 @@ _dl_procinfo (unsigned int type, unsigned long int word)
case AT_HWCAP:
_dl_printf ("AT_HWCAP: ");
- for (int i = 0; i <= _DL_HWCAP_LAST; ++i)
+ for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
if (word & (1 << i))
_dl_printf (" %s", _dl_hwcap_string (i));
break;
case AT_HWCAP2:
{
- unsigned int offset = _DL_HWCAP_LAST + 1;
_dl_printf ("AT_HWCAP2: ");
- /* We have to go through them all because the kernel added the
- AT_HWCAP2 features starting with the high bits. */
- for (int i = 0; i <= _DL_HWCAP2_LAST; ++i)
- if (word & (1 << i))
- _dl_printf (" %s", _dl_hwcap_string (offset + i));
+ /* We have to go through them all because the kernel added the
+ AT_HWCAP2 features starting with the high bits. */
+ for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
+ if (word & (1 << i))
+ _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP2_OFFSET + i));
+ break;
+ }
+ case AT_HWCAP3:
+ {
+ _dl_printf ("AT_HWCAP3: ");
+
+ /* We have to go through them all because the kernel added the
+ AT_HWCAP3 features starting with the high bits. */
+ for (int i = 0; i < _DL_HWCAP_SIZE; ++i)
+ if (word & (1 << i))
+ _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP3_OFFSET + i));
+ break;
+ }
+ case AT_HWCAP4:
+ {
+ _dl_printf ("AT_HWCAP4: ");
+
+ /* We have to go through them all because the kernel added the
+ AT_HWCAP4 features starting with the high bits. */
+ for (int i = 0; i <= _DL_HWCAP_SIZE; ++i)
+ if (word & (1 << i))
+ _dl_printf (" %s", _dl_hwcap_string (_DL_HWCAP4_OFFSET + i));
break;
}
case AT_L1I_CACHEGEOMETRY:
@@ -31,7 +31,7 @@ void
__tcb_parse_hwcap_and_convert_at_platform (void)
{
- uint64_t h1, h2;
+ uint64_t h1, h2, h3, h4;
/* Read AT_PLATFORM string from auxv and convert it to a number. */
__tcb.at_platform = _dl_string_platform (GLRO (dl_platform));
@@ -39,6 +39,8 @@ __tcb_parse_hwcap_and_convert_at_platform (void)
/* Read HWCAP and HWCAP2 from auxv. */
h1 = GLRO (dl_hwcap);
h2 = GLRO (dl_hwcap2);
+ h3 = GLRO (dl_hwcap3);
+ h4 = GLRO (dl_hwcap4);
/* hwcap contains only the latest supported ISA, the code checks which is
and fills the previous supported ones. */
@@ -64,18 +66,23 @@ __tcb_parse_hwcap_and_convert_at_platform (void)
else if (h1 & PPC_FEATURE_POWER5)
h1 |= PPC_FEATURE_POWER4;
- uint64_t array_hwcaps[] = { h1, h2 };
+ uint64_t array_hwcaps[] = { h1, h2, h3, h4 };
init_cpu_features (&GLRO(dl_powerpc_cpu_features), array_hwcaps);
/* Consolidate both HWCAP and HWCAP2 into a single doubleword so that
we can read both in a single load later. */
__tcb.hwcap = (h1 << 32) | (h2 & 0xffffffff);
- __tcb.hwcap_extn = 0x0;
+
+ /* Consolidate both HWCAP3 and HWCAP4 into a single doubleword so that
+ we can read both in a single load later. */
+ __tcb.hwcap_extn = (h3 << 32) | (h4 & 0xffffffff);
}
#if IS_IN (rtld)
versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
__parse_hwcap_and_convert_at_platform, GLIBC_2_23);
+versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
+ __parse_hwcap_3_4_and_convert_at_platform, GLIBC_2_39);
#endif
/* Export __parse_hwcap_and_convert_at_platform in libc.a. This is used by
@@ -84,4 +91,6 @@ versioned_symbol (ld, __tcb_parse_hwcap_and_convert_at_platform, \
#ifndef SHARED
weak_alias (__tcb_parse_hwcap_and_convert_at_platform, \
__parse_hwcap_and_convert_at_platform);
+weak_alias (__tcb_parse_hwcap_and_convert_at_platform, \
+ __parse_hwcap_3_4_and_convert_at_platform);
#endif
@@ -47,6 +47,8 @@ void _dl_parse_auxv (ElfW(auxv_t) *av, dl_parse_auxv_t auxv_values)
GLRO(dl_platform) = (void *) auxv_values[AT_PLATFORM];
GLRO(dl_hwcap) = auxv_values[AT_HWCAP];
GLRO(dl_hwcap2) = auxv_values[AT_HWCAP2];
+ GLRO(dl_hwcap3) = auxv_values[AT_HWCAP3];
+ GLRO(dl_hwcap4) = auxv_values[AT_HWCAP4];
GLRO(dl_clktck) = auxv_values[AT_CLKTCK];
GLRO(dl_fpu_control) = auxv_values[AT_FPUCW];
_dl_random = (void *) auxv_values[AT_RANDOM];
@@ -197,6 +197,8 @@ _dl_show_auxv (void)
[AT_SYSINFO_EHDR - 2] = { "SYSINFO_EHDR: 0x", hex },
[AT_RANDOM - 2] = { "RANDOM: 0x", hex },
[AT_HWCAP2 - 2] = { "HWCAP2: 0x", hex },
+ [AT_HWCAP3 - 2] = { "HWCAP3: 0x", hex },
+ [AT_HWCAP4 - 2] = { "HWCAP4: 0x", hex },
[AT_MINSIGSTKSZ - 2] = { "MINSIGSTKSZ: ", dec },
[AT_L1I_CACHESIZE - 2] = { "L1I_CACHESIZE: ", dec },
[AT_L1I_CACHEGEOMETRY - 2] = { "L1I_CACHEGEOMETRY: 0x", hex },
@@ -113,6 +113,8 @@ init_cpu_features (struct cpu_features *cpu_features, uint64_t hwcaps[])
which are set by __tcb_parse_hwcap_and_convert_at_platform. */
cpu_features->hwcap = hwcaps[0];
cpu_features->hwcap2 = hwcaps[1];
+ cpu_features->hwcap3 = hwcaps[2];
+ cpu_features->hwcap4 = hwcaps[3];
/* Default is to use aligned memory access on optimized function unless
tunables is enable, since for this case user can explicit disable
unaligned optimizations. */
@@ -26,6 +26,8 @@ struct cpu_features
bool use_cached_memopt;
unsigned long int hwcap;
unsigned long int hwcap2;
+ unsigned long int hwcap3;
+ unsigned long int hwcap4;
};
static const char hwcap_names[] = {
@@ -87,6 +87,12 @@ __libc_start_main_impl (int argc, char **argv,
case AT_HWCAP2:
_dl_hwcap2 = (unsigned long int) av->a_un.a_val;
break;
+ case AT_HWCAP3:
+ _dl_hwcap3 = (unsigned long int) av->a_un.a_val;
+ break;
+ case AT_HWCAP4:
+ _dl_hwcap4 = (unsigned long int) av->a_un.a_val;
+ break;
case AT_PLATFORM:
_dl_platform = (void *) av->a_un.a_val;
break;
@@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
GLIBC_2.35 __rseq_flags D 0x4
GLIBC_2.35 __rseq_offset D 0x4
GLIBC_2.35 __rseq_size D 0x4
+GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F
@@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
GLIBC_2.35 __rseq_flags D 0x4
GLIBC_2.35 __rseq_offset D 0x8
GLIBC_2.35 __rseq_size D 0x4
+GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F
@@ -8,3 +8,4 @@ GLIBC_2.34 __rtld_version_placeholder F
GLIBC_2.35 __rseq_flags D 0x4
GLIBC_2.35 __rseq_offset D 0x8
GLIBC_2.35 __rseq_size D 0x4
+GLIBC_2.39 __parse_hwcap_3_4_and_convert_at_platform F