From: Andrew Jones <drjones@redhat.com>
To: Richard Henderson <richard.henderson@linaro.org>
Cc: qemu-arm@nongnu.org, qemu-devel@nongnu.org
Subject: Re: [PATCH v2 3/3] target/arm: Add sve-default-vector-length cpu property
Date: Mon, 26 Jul 2021 16:59:51 +0200 [thread overview]
Message-ID: <20210726145951.g3ywwhcfstmsi7vi@gator> (raw)
In-Reply-To: <20210723203344.968563-4-richard.henderson@linaro.org>
On Fri, Jul 23, 2021 at 10:33:44AM -1000, Richard Henderson wrote:
> Mirror the behavour of /proc/sys/abi/sve_default_vector_length
> under the real linux kernel. We have no way of passing along
> a real default across exec like the kernel can, but this is a
> decent way of adjusting the startup vector length of a process.
>
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/482
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
> docs/system/arm/cpu-features.rst | 11 ++++++
> target/arm/cpu.h | 5 +++
> target/arm/cpu.c | 14 ++++++--
> target/arm/cpu64.c | 60 ++++++++++++++++++++++++++++++++
> 4 files changed, 88 insertions(+), 2 deletions(-)
>
> diff --git a/docs/system/arm/cpu-features.rst b/docs/system/arm/cpu-features.rst
> index c455442eaf..4ff36cc83f 100644
> --- a/docs/system/arm/cpu-features.rst
> +++ b/docs/system/arm/cpu-features.rst
> @@ -376,3 +376,14 @@ verbose command lines. However, the recommended way to select vector
> lengths is to explicitly enable each desired length. Therefore only
> example's (1), (4), and (6) exhibit recommended uses of the properties.
>
> +SVE User-mode Default Vector Length Property
> +--------------------------------------------
> +
> +For qemu-aarch64, the cpu property `sve-default-vector-length=N` is
> +defined to mirror the Linux kernel parameter file
> +`/proc/sys/abi/sve_default_vector_length`. The default length, `N`,
> +is in units of bytes and must be between 16 and 8192.
Hmm. If a user inputs anything greater than 256, then won't it get
silently reduced to 256?
> +If not specified, the default vector length is 64.
> +
> +If the default length is larger than the maximum vector length enabled
> +with `sve<N>` properties, the actual vector length will be reduced.
Here it's pointed out that the default may be reduced, but it implies that
that only happens if an sve<N> property is also given. Won't users wonder
why they're only getting vectors that are 256 bytes large even when they
ask for more?
Thanks,
drew
> diff --git a/target/arm/cpu.h b/target/arm/cpu.h
> index be9a4dceae..9f0a5f84d5 100644
> --- a/target/arm/cpu.h
> +++ b/target/arm/cpu.h
> @@ -1006,6 +1006,11 @@ struct ARMCPU {
> /* Used to set the maximum vector length the cpu will support. */
> uint32_t sve_max_vq;
>
> +#ifdef CONFIG_USER_ONLY
> + /* Used to set the default vector length at process start. */
> + uint32_t sve_default_vq;
> +#endif
> +
> /*
> * In sve_vq_map each set bit is a supported vector length of
> * (bit-number + 1) * 16 bytes, i.e. each bit number + 1 is the vector
> diff --git a/target/arm/cpu.c b/target/arm/cpu.c
> index 752b15bb79..2866dd7658 100644
> --- a/target/arm/cpu.c
> +++ b/target/arm/cpu.c
> @@ -201,7 +201,8 @@ static void arm_cpu_reset(DeviceState *dev)
> env->cp15.cpacr_el1 = deposit64(env->cp15.cpacr_el1, 16, 2, 3);
> /* with reasonable vector length */
> if (cpu_isar_feature(aa64_sve, cpu)) {
> - env->vfp.zcr_el[1] = MIN(cpu->sve_max_vq - 1, 3);
> + env->vfp.zcr_el[1] =
> + aarch64_sve_zcr_get_valid_len(cpu, cpu->sve_default_vq - 1);
> }
> /*
> * Enable TBI0 but not TBI1.
> @@ -1051,7 +1052,16 @@ static void arm_cpu_initfn(Object *obj)
> QLIST_INIT(&cpu->pre_el_change_hooks);
> QLIST_INIT(&cpu->el_change_hooks);
>
> -#ifndef CONFIG_USER_ONLY
> +#ifdef CONFIG_USER_ONLY
> +# ifdef TARGET_AARCH64
> + /*
> + * The linux kernel defaults to 512-bit vectors, when sve is supported.
> + * See documentation for /proc/sys/abi/sve_default_vector_length, and
> + * our corresponding sve-default-vector-length cpu property.
> + */
> + cpu->sve_default_vq = 4;
> +# endif
> +#else
> /* Our inbound IRQ and FIQ lines */
> if (kvm_enabled()) {
> /* VIRQ and VFIQ are unused with KVM but we add them to maintain
> diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c
> index c7a1626bec..c690318a9b 100644
> --- a/target/arm/cpu64.c
> +++ b/target/arm/cpu64.c
> @@ -559,6 +559,59 @@ static void cpu_arm_set_sve(Object *obj, bool value, Error **errp)
> cpu->isar.id_aa64pfr0 = t;
> }
>
> +#ifdef CONFIG_USER_ONLY
> +/* Mirror linux /proc/sys/abi/sve_default_vector_length. */
> +static void cpu_arm_set_sve_default_vec_len(Object *obj, Visitor *v,
> + const char *name, void *opaque,
> + Error **errp)
> +{
> + ARMCPU *cpu = ARM_CPU(obj);
> + int32_t default_len, default_vq, remainder;
> +
> + if (!visit_type_int32(v, name, &default_len, errp)) {
> + return;
> + }
> +
> + /* Undocumented, but the kernel allows -1 to indicate "maximum". */
> + if (default_len == -1) {
> + cpu->sve_default_vq = ARM_MAX_VQ;
> + return;
> + }
> +
> + default_vq = default_len / 16;
> + remainder = default_len % 16;
> +
> + /*
> + * Note that the 512 max comes from include/uapi/asm/sve_context.h
> + * and is the maximum architectural width of ZCR_ELx.LEN.
> + */
> + if (remainder || default_vq < 1 || default_vq > 512) {
> + error_setg(errp, "cannot set sve-default-vector-length");
> + if (remainder) {
> + error_append_hint(errp, "Vector length not a multiple of 16\n");
> + } else if (default_vq < 1) {
> + error_append_hint(errp, "Vector length smaller than 16\n");
> + } else {
> + error_append_hint(errp, "Vector length larger than %d\n",
> + 512 * 16);
> + }
> + return;
> + }
> +
> + cpu->sve_default_vq = default_vq;
> +}
> +
> +static void cpu_arm_get_sve_default_vec_len(Object *obj, Visitor *v,
> + const char *name, void *opaque,
> + Error **errp)
> +{
> + ARMCPU *cpu = ARM_CPU(obj);
> + int32_t value = cpu->sve_default_vq * 16;
> +
> + visit_type_int32(v, name, &value, errp);
> +}
> +#endif
> +
> void aarch64_add_sve_properties(Object *obj)
> {
> uint32_t vq;
> @@ -571,6 +624,13 @@ void aarch64_add_sve_properties(Object *obj)
> object_property_add(obj, name, "bool", cpu_arm_get_sve_vq,
> cpu_arm_set_sve_vq, NULL, NULL);
> }
> +
> +#ifdef CONFIG_USER_ONLY
> + /* Mirror linux /proc/sys/abi/sve_default_vector_length. */
> + object_property_add(obj, "sve-default-vector-length", "int32",
> + cpu_arm_get_sve_default_vec_len,
> + cpu_arm_set_sve_default_vec_len, NULL, NULL);
> +#endif
> }
>
> void arm_cpu_pauth_finalize(ARMCPU *cpu, Error **errp)
> --
> 2.25.1
>
>
next prev parent reply other threads:[~2021-07-26 15:01 UTC|newest]
Thread overview: 14+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-07-23 20:33 [PATCH v2 0/3] target/arm: Add sve-default-vector-length cpu property Richard Henderson
2021-07-23 20:33 ` [PATCH v2 1/3] target/arm: Correctly bound length in sve_zcr_get_valid_len Richard Henderson
2021-07-26 10:57 ` Peter Maydell
2021-07-23 20:33 ` [PATCH v2 2/3] target/arm: Export aarch64_sve_zcr_get_valid_len Richard Henderson
2021-07-26 10:57 ` Peter Maydell
2021-07-23 20:33 ` [PATCH v2 3/3] target/arm: Add sve-default-vector-length cpu property Richard Henderson
2021-07-26 11:07 ` Peter Maydell
2021-07-26 14:59 ` Andrew Jones [this message]
2021-07-26 18:33 ` Richard Henderson
2021-07-26 18:40 ` Andrew Jones
2021-07-26 19:31 ` Peter Maydell
2021-07-26 12:42 ` [PATCH v2 0/3] " Peter Maydell
2021-07-26 15:00 ` Andrew Jones
2021-07-26 15:41 ` Peter Maydell
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210726145951.g3ywwhcfstmsi7vi@gator \
--to=drjones@redhat.com \
--cc=qemu-arm@nongnu.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).