This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 1/2] Hoist ZVA check out of the function
- From: Adhemerval Zanella <adhemerval dot zanella at linaro dot org>
- To: libc-alpha at sourceware dot org
- Date: Thu, 28 Sep 2017 12:12:15 -0700
- Subject: Re: [PATCH 1/2] Hoist ZVA check out of the function
- Authentication-results: sourceware.org; auth=none
- References: <1505834596-21641-1-git-send-email-siddhesh@sourceware.org>
On 19/09/2017 08:23, Siddhesh Poyarekar wrote:
> The DZP bit in the dczid_el0 register does not change dynamically, so
> it is safe to read once during program startup. Hoist the zva check
> into an ifunc resolver and store the result into a static variable,
> which can be read in case of non-standard zva sizes. This effectively
> adds 3 ifunc variants for memset - one for cases where zva is
> disabled, one for 64 byte zva and another for 128 byte zva. I have
> retained the older memset as __memset_generic for internal libc.so use
> so that the change impact is minimal. We should eventually have a
> discussion on what is more expensive, reading dczid_el0 on every
> memset invocation or the indirection due to PLT.
>
> The gains due to this are significant for falkor, with gains as high
> as 80% in some cases. Likewise for mustang, although the numbers are
> slightly lower. Here's a sample from the falkor tests:
>
I would use a more compact ChangeLog entry as:
* sysdeps/aarch64/memset.S (do_no_zva, do_zva_64,
do_zva_128, do_zva_default): New macro.
Same for the other entries where it applies.
> (MEMSET): Use the new macros.
> (MEMSET)[INTERNAL_MEMSET]: Retain old memset.
> (MEMSET)[!INTERNAL_MEMSET]: Remove zva check.
> * sysdeps/aarch64/multiarch/Makefile (sysdep_routines):
> Add memset_generic, memset_nozva and memset_zva.
> * sysdeps/aarch64/multiarch/ifunc-impl-list.c
> (__libc_ifunc_impl_list): Add memset ifuncs.
> * sysdeps/aarch64/multiarch/init-arch.h (INIT_ARCH): New
> static variable __aarch64_zva_size and local variable
> zva_size.
> * sysdeps/aarch64/multiarch/memset.c: New file.
> * sysdeps/aarch64/multiarch/memset_generic.S: New file.
> * sysdeps/aarch64/multiarch/memset_nozva.S: New file.
> * sysdeps/aarch64/multiarch/memset_zva.S: New file.
> * sysdeps/unix/sysv/linux/aarch64/cpu-features.c
> (DCZID_DZP_MASK): New macro.
> (DCZID_BS_MASK): Likewise.
> (init_cpu_features): Read and set zva_size.
> * sysdeps/unix/sysv/linux/aarch64/cpu-features.h
> (struct cpu_features): New member zva_size.
>
>
> diff --git a/sysdeps/aarch64/memset.S b/sysdeps/aarch64/memset.S
> index 110fd22..8cff3a4 100644
> --- a/sysdeps/aarch64/memset.S
> +++ b/sysdeps/aarch64/memset.S
> @@ -37,7 +37,108 @@
> #define zva_len x7
> #define zva_lenw w7
>
> -ENTRY_ALIGN (__memset, 6)
> +/* Macros that do the critical loops for either no zva or zva of 64 bytes, 128
> + bytes and higher sizes. */
> +
> +#ifndef ZVA_MACROS
> +# define ZVA_MACROS
> +/* No ZVA. */
> +.macro do_no_zva
> + sub count, dstend, dst /* Count is 16 too large. */
> + add dst, dst, 16
> + sub count, count, 64 + 16 /* Adjust count and bias for loop. */
> +1: stp q0, q0, [dst], 64
> + stp q0, q0, [dst, -32]
> + subs count, count, 64
> + b.hi 1b
> + stp q0, q0, [dstend, -64]
> + stp q0, q0, [dstend, -32]
> + ret
> +.endm
> +
> +/* Write the first and last 64 byte aligned block using stp rather
> + than using DC ZVA. This is faster on some cores. */
> +.macro do_zva_64
> + str q0, [dst, 16]
> + stp q0, q0, [dst, 32]
> + bic dst, dst, 63
> + stp q0, q0, [dst, 64]
> + stp q0, q0, [dst, 96]
> + sub count, dstend, dst /* Count is now 128 too large. */
> + sub count, count, 128+64+64 /* Adjust count and bias for loop. */
> + add dst, dst, 128
> + nop
> +1: dc zva, dst
> + add dst, dst, 64
> + subs count, count, 64
> + b.hi 1b
> + stp q0, q0, [dst, 0]
> + stp q0, q0, [dst, 32]
> + stp q0, q0, [dstend, -64]
> + stp q0, q0, [dstend, -32]
> + ret
> +.endm
> +
> +/* ZVA size of 128 bytes. */
> +.macro do_zva_128
> + str q0, [dst, 16]
> + stp q0, q0, [dst, 32]
> + stp q0, q0, [dst, 64]
> + stp q0, q0, [dst, 96]
> + bic dst, dst, 127
> + sub count, dstend, dst /* Count is now 128 too large. */
> + sub count, count, 128+128 /* Adjust count and bias for loop. */
> + add dst, dst, 128
> +1: dc zva, dst
> + add dst, dst, 128
> + subs count, count, 128
> + b.hi 1b
> + stp q0, q0, [dstend, -128]
> + stp q0, q0, [dstend, -96]
> + stp q0, q0, [dstend, -64]
> + stp q0, q0, [dstend, -32]
> + ret
> +.endm
> +
> +/* ZVA size of more than 128 bytes. */
> +.macro do_zva_default
> + add tmp1, zva_len, 64 /* Max alignment bytes written. */
> + cmp count, tmp1
> + blo MEMSET_L(no_zva)
> +
> + sub tmp2, zva_len, 1
> + add tmp1, dst, zva_len
> + add dst, dst, 16
> + subs count, tmp1, dst /* Actual alignment bytes to write. */
> + bic tmp1, tmp1, tmp2 /* Aligned dc zva start address. */
> + beq 2f
> +1: stp q0, q0, [dst], 64
> + stp q0, q0, [dst, -32]
> + subs count, count, 64
> + b.hi 1b
> +2: mov dst, tmp1
> + sub count, dstend, tmp1 /* Remaining bytes to write. */
> + subs count, count, zva_len
> + b.lo 4f
> +3: dc zva, dst
> + add dst, dst, zva_len
> + subs count, count, zva_len
> + b.hs 3b
> +4: add count, count, zva_len
> + subs count, count, 64
> + b.ls 6f
> +5: stp q0, q0, [dst], 64
> + stp q0, q0, [dst, -32]
> + subs count, count, 64
> + b.hi 5b
> +6: stp q0, q0, [dstend, -64]
> + stp q0, q0, [dstend, -32]
> + ret
> +.endm
> +#endif
> +
> +/* Memset entry point. */
> +ENTRY_ALIGN (MEMSET, 6)
>
> DELOUSE (0)
> DELOUSE (2)
> @@ -46,9 +147,9 @@ ENTRY_ALIGN (__memset, 6)
> add dstend, dstin, count
>
> cmp count, 96
> - b.hi L(set_long)
> + b.hi MEMSET_L(set_long)
> cmp count, 16
> - b.hs L(set_medium)
> + b.hs MEMSET_L(set_medium)
> mov val, v0.D[0]
>
> /* Set 0..15 bytes. */
> @@ -68,9 +169,9 @@ ENTRY_ALIGN (__memset, 6)
> 3: ret
>
> /* Set 17..96 bytes. */
> -L(set_medium):
> +MEMSET_L(set_medium):
> str q0, [dstin]
> - tbnz count, 6, L(set96)
> + tbnz count, 6, MEMSET_L(set96)
> str q0, [dstend, -16]
> tbz count, 5, 1f
> str q0, [dstin, 16]
> @@ -80,7 +181,7 @@ L(set_medium):
> .p2align 4
> /* Set 64..96 bytes. Write 64 bytes from the start and
> 32 bytes from the end. */
> -L(set96):
> +MEMSET_L(set96):
> str q0, [dstin, 16]
> stp q0, q0, [dstin, 32]
> stp q0, q0, [dstend, -32]
> @@ -88,108 +189,63 @@ L(set96):
>
> .p2align 3
> nop
> -L(set_long):
> +MEMSET_L(set_long):
> +#ifdef INTERNAL_MEMSET
> and valw, valw, 255
> bic dst, dstin, 15
> str q0, [dstin]
> cmp count, 256
> ccmp valw, 0, 0, cs
> - b.eq L(try_zva)
> -L(no_zva):
> - sub count, dstend, dst /* Count is 16 too large. */
> - add dst, dst, 16
> - sub count, count, 64 + 16 /* Adjust count and bias for loop. */
> -1: stp q0, q0, [dst], 64
> - stp q0, q0, [dst, -32]
> -L(tail64):
> - subs count, count, 64
> - b.hi 1b
> -2: stp q0, q0, [dstend, -64]
> - stp q0, q0, [dstend, -32]
> - ret
> + b.eq MEMSET_L(try_zva)
>
> - .p2align 3
> -L(try_zva):
> +MEMSET_L(no_zva):
> + do_no_zva
> +
> + .p2align 4
> +MEMSET_L(try_zva):
> mrs tmp1, dczid_el0
> - tbnz tmp1w, 4, L(no_zva)
> and tmp1w, tmp1w, 15
> cmp tmp1w, 4 /* ZVA size is 64 bytes. */
> - b.ne L(zva_128)
> + b.ne MEMSET_L(zva_128)
> + do_zva_64
>
> - /* Write the first and last 64 byte aligned block using stp rather
> - than using DC ZVA. This is faster on some cores.
> - */
> -L(zva_64):
> - str q0, [dst, 16]
> - stp q0, q0, [dst, 32]
> - bic dst, dst, 63
> - stp q0, q0, [dst, 64]
> - stp q0, q0, [dst, 96]
> - sub count, dstend, dst /* Count is now 128 too large. */
> - sub count, count, 128+64+64 /* Adjust count and bias for loop. */
> - add dst, dst, 128
> - nop
> -1: dc zva, dst
> - add dst, dst, 64
> - subs count, count, 64
> - b.hi 1b
> - stp q0, q0, [dst, 0]
> - stp q0, q0, [dst, 32]
> - stp q0, q0, [dstend, -64]
> - stp q0, q0, [dstend, -32]
> - ret
> -
> - .p2align 3
> -L(zva_128):
> +MEMSET_L(zva_128):
> cmp tmp1w, 5 /* ZVA size is 128 bytes. */
> - b.ne L(zva_other)
> + b.ne MEMSET_L(zva_other)
> + do_zva_128
>
> - str q0, [dst, 16]
> - stp q0, q0, [dst, 32]
> - stp q0, q0, [dst, 64]
> - stp q0, q0, [dst, 96]
> - bic dst, dst, 127
> - sub count, dstend, dst /* Count is now 128 too large. */
> - sub count, count, 128+128 /* Adjust count and bias for loop. */
> - add dst, dst, 128
> -1: dc zva, dst
> - add dst, dst, 128
> - subs count, count, 128
> - b.hi 1b
> - stp q0, q0, [dstend, -128]
> - stp q0, q0, [dstend, -96]
> - stp q0, q0, [dstend, -64]
> - stp q0, q0, [dstend, -32]
> - ret
> -
> -L(zva_other):
> +MEMSET_L(zva_other):
> mov tmp2w, 4
> lsl zva_lenw, tmp2w, tmp1w
> - add tmp1, zva_len, 64 /* Max alignment bytes written. */
> - cmp count, tmp1
> - blo L(no_zva)
> + do_zva_default
> +#else
> + /* Memset called through PLT, so we need only one of the ZVA
> + variants. */
> +# ifdef MEMSET_ZVA
> + and valw, valw, 255
> +# endif
> + bic dst, dstin, 15
> + str q0, [dstin]
> +# ifdef MEMSET_ZVA
> + cmp count, 256
> + ccmp valw, 0, 0, cs
> + b.eq MEMSET_L(try_zva)
> +# endif
> +MEMSET_L(no_zva):
> + do_no_zva
> +# if defined MEMSET_ZVA
> +MEMSET_L(try_zva):
> +# if MEMSET_ZVA == 64
> + do_zva_64
> +# elif MEMSET_ZVA == 128
> + do_zva_128
> +# else
> + adrp zva_len, __aarch64_zva_size
> + ldr zva_len, [zva_len, #:lo12:__aarch64_zva_size]
> + do_zva_default
> +# endif
> +# endif
> +#endif
>
> - sub tmp2, zva_len, 1
> - add tmp1, dst, zva_len
> - add dst, dst, 16
> - subs count, tmp1, dst /* Actual alignment bytes to write. */
> - bic tmp1, tmp1, tmp2 /* Aligned dc zva start address. */
> - beq 2f
> -1: stp q0, q0, [dst], 64
> - stp q0, q0, [dst, -32]
> - subs count, count, 64
> - b.hi 1b
> -2: mov dst, tmp1
> - sub count, dstend, tmp1 /* Remaining bytes to write. */
> - subs count, count, zva_len
> - b.lo 4f
> -3: dc zva, dst
> - add dst, dst, zva_len
> - subs count, count, zva_len
> - b.hs 3b
> -4: add count, count, zva_len
> - b L(tail64)
> -
> -END (__memset)
> -weak_alias (__memset, memset)
> -libc_hidden_builtin_def (memset)
> +END (MEMSET)
> +libc_hidden_builtin_def (MEMSET)
> diff --git a/sysdeps/aarch64/multiarch/Makefile b/sysdeps/aarch64/multiarch/Makefile
> index 9aa1e79..f611182 100644
> --- a/sysdeps/aarch64/multiarch/Makefile
> +++ b/sysdeps/aarch64/multiarch/Makefile
> @@ -1,4 +1,4 @@
> ifeq ($(subdir),string)
> sysdep_routines += memcpy_generic memcpy_thunderx memcpy_falkor \
> - memmove_falkor
> + memmove_falkor memset_generic memset_nozva memset_zva
> endif
> diff --git a/sysdeps/aarch64/multiarch/ifunc-impl-list.c b/sysdeps/aarch64/multiarch/ifunc-impl-list.c
> index 2cb74d5..29148ac 100644
> --- a/sysdeps/aarch64/multiarch/ifunc-impl-list.c
> +++ b/sysdeps/aarch64/multiarch/ifunc-impl-list.c
> @@ -46,6 +46,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
> IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_thunderx)
> IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_falkor)
> IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_generic))
> + IFUNC_IMPL (i, name, memset,
> + IFUNC_IMPL_ADD (array, i, memset, 1, __memset_nozva)
> + IFUNC_IMPL_ADD (array, i, memset, (zva_size == 64), __memset_zva_64)
> + IFUNC_IMPL_ADD (array, i, memset, (zva_size == 128), __memset_zva_128)
> + IFUNC_IMPL_ADD (array, i, memset, (zva_size > 0), __memset_zva_default)
> + IFUNC_IMPL_ADD (array, i, memset, 1, __memset_generic))
>
> return i;
> }
> diff --git a/sysdeps/aarch64/multiarch/init-arch.h b/sysdeps/aarch64/multiarch/init-arch.h
> index 3af442c..541c27e 100644
> --- a/sysdeps/aarch64/multiarch/init-arch.h
> +++ b/sysdeps/aarch64/multiarch/init-arch.h
> @@ -18,6 +18,9 @@
>
> #include <ldsodefs.h>
>
> -#define INIT_ARCH() \
> - uint64_t __attribute__((unused)) midr = \
> - GLRO(dl_aarch64_cpu_features).midr_el1;
> +#define INIT_ARCH() \
> + uint64_t __attribute__((unused)) midr = \
> + GLRO(dl_aarch64_cpu_features).midr_el1; \
> + extern unsigned __aarch64_zva_size; \
> + unsigned __attribute__((unused)) zva_size = __aarch64_zva_size = \
> + GLRO(dl_aarch64_cpu_features).zva_size;
> diff --git a/sysdeps/aarch64/multiarch/memset.c b/sysdeps/aarch64/multiarch/memset.c
> new file mode 100644
> index 0000000..58e669a
> --- /dev/null
> +++ b/sysdeps/aarch64/multiarch/memset.c
> @@ -0,0 +1,47 @@
> +/* Multiple versions of memset. AARCH64 version.
> + Copyright (C) 2017 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <http://www.gnu.org/licenses/>. */
> +
> +/* Define multiple versions only for the definition in libc. */
> +
> +#if IS_IN (libc)
> +/* Redefine memset so that the compiler won't complain about the type
> + mismatch with the IFUNC selector in strong_alias, below. */
> +# undef memset
> +# define memset __redirect_memset
> +# include <string.h>
> +# include <init-arch.h>
> +
> +unsigned __aarch64_zva_size;
> +
> +extern __typeof (__redirect_memset) __libc_memset;
> +
> +extern __typeof (__redirect_memset) __memset_nozva attribute_hidden;
> +extern __typeof (__redirect_memset) __memset_zva_64 attribute_hidden;
> +extern __typeof (__redirect_memset) __memset_zva_128 attribute_hidden;
> +extern __typeof (__redirect_memset) __memset_zva_default attribute_hidden;
> +
> +libc_ifunc (__libc_memset, (zva_size == 0 ? __memset_nozva
> + : (zva_size == 64 ? __memset_zva_64
> + : (zva_size == 128 ? __memset_zva_128
> + : __memset_zva_default))));
> +
> +# undef memset
> +strong_alias (__libc_memset, memset);
> +#else
> +#include <string/memset.c>
> +#endif
You don't need use the default version for the loader, you can use the
generic sysdeps/aarch64/memset.S by creating a rtld-memset.S on
multiarch and defining the required macros.
> diff --git a/sysdeps/aarch64/multiarch/memset_generic.S b/sysdeps/aarch64/multiarch/memset_generic.S
> new file mode 100644
> index 0000000..56f1e02
> --- /dev/null
> +++ b/sysdeps/aarch64/multiarch/memset_generic.S
> @@ -0,0 +1,27 @@
> +/* Memset for aarch64, default version for internal use.
> + Copyright (C) 2017 Free Software Foundation, Inc.
> +
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library. If not, see
> + <http://www.gnu.org/licenses/>. */
> +
> +#define MEMSET __memset_generic
> +#define INTERNAL_MEMSET
> +#define MEMSET_L(label) L(label)
> +#ifdef SHARED
> + .globl __GI_memset; __GI_memset = __memset_generic
> +#endif
I would add a comment stating it is essentially doing libc_hidden_def(memset)
and redirecting the internal implementation to __memset_generic.
> +
> +#include <sysdeps/aarch64/memset.S>
> diff --git a/sysdeps/aarch64/multiarch/memset_nozva.S b/sysdeps/aarch64/multiarch/memset_nozva.S
> new file mode 100644
> index 0000000..98045ac
> --- /dev/null
> +++ b/sysdeps/aarch64/multiarch/memset_nozva.S
> @@ -0,0 +1,22 @@
> +/* Memset for aarch64, ZVA disabled.
> + Copyright (C) 2017 Free Software Foundation, Inc.
> +
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library. If not, see
> + <http://www.gnu.org/licenses/>. */
> +
> +#define MEMSET __memset_nozva
> +#define MEMSET_L(label) L(label)
> +#include <sysdeps/aarch64/memset.S>
Although not strictly required, I think it should avoid build these
for !IS_IN(libc) as for memset_zva.S. Same applied for memset_generic.S.
> diff --git a/sysdeps/aarch64/multiarch/memset_zva.S b/sysdeps/aarch64/multiarch/memset_zva.S
> new file mode 100644
> index 0000000..5d02b89
> --- /dev/null
> +++ b/sysdeps/aarch64/multiarch/memset_zva.S
> @@ -0,0 +1,41 @@
> +/* Memset for aarch64, ZVA enabled.
> + Copyright (C) 2017 Free Software Foundation, Inc.
> +
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library. If not, see
> + <http://www.gnu.org/licenses/>. */
> +
> +#if IS_IN (libc)
> +# define MEMSET __memset_zva_64
> +# define MEMSET_ZVA 64
> +# define MEMSET_L(label) L(label ## _zva64)
> +# include <sysdeps/aarch64/memset.S>
> +
> +# undef MEMSET
> +# undef MEMSET_ZVA
> +# undef MEMSET_L
> +# define MEMSET __memset_zva_128
> +# define MEMSET_ZVA 128
> +# define MEMSET_L(label) L(label ## _zva128)
> +# include <sysdeps/aarch64/memset.S>
> +
> +# undef MEMSET
> +# undef MEMSET_ZVA
> +# undef MEMSET_L
> +# define MEMSET __memset_zva_default
> +# define MEMSET_ZVA 1
> +# define MEMSET_L(label) L(label ## _zvadef)
> +# include <sysdeps/aarch64/memset.S>
> +#endif
> diff --git a/sysdeps/unix/sysv/linux/aarch64/cpu-features.c b/sysdeps/unix/sysv/linux/aarch64/cpu-features.c
> index e769eeb..092ee81 100644
> --- a/sysdeps/unix/sysv/linux/aarch64/cpu-features.c
> +++ b/sysdeps/unix/sysv/linux/aarch64/cpu-features.c
> @@ -20,6 +20,9 @@
> #include <sys/auxv.h>
> #include <elf/dl-hwcaps.h>
>
> +#define DCZID_DZP_MASK (1 << 4)
> +#define DCZID_BS_MASK (0xf)
> +
> #if HAVE_TUNABLES
> struct cpu_list
> {
> @@ -72,4 +75,11 @@ init_cpu_features (struct cpu_features *cpu_features)
> }
>
> cpu_features->midr_el1 = midr;
> +
> + /* Check if ZVA is enabled. */
> + unsigned dczid;
> + asm volatile ("mrs %0, dczid_el0" : "=r"(dczid));
> +
> + if ((dczid & DCZID_DZP_MASK) == 0)
> + cpu_features->zva_size = 4 << (dczid & DCZID_BS_MASK);
> }
> diff --git a/sysdeps/unix/sysv/linux/aarch64/cpu-features.h b/sysdeps/unix/sysv/linux/aarch64/cpu-features.h
> index 73cb53d..f2b6afd 100644
> --- a/sysdeps/unix/sysv/linux/aarch64/cpu-features.h
> +++ b/sysdeps/unix/sysv/linux/aarch64/cpu-features.h
> @@ -47,6 +47,7 @@
> struct cpu_features
> {
> uint64_t midr_el1;
> + unsigned zva_size;
> };
>
> #endif /* _CPU_FEATURES_AARCH64_H */
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>