All of lore.kernel.org
 help / color / mirror / Atom feed
From: Dave.Martin@arm.com (Dave Martin)
To: linux-arm-kernel@lists.infradead.org
Subject: [RFC PATCH v2 35/41] arm64/sve: Add vector length inheritance control
Date: Wed, 22 Mar 2017 14:51:05 +0000	[thread overview]
Message-ID: <1490194274-30569-36-git-send-email-Dave.Martin@arm.com> (raw)
In-Reply-To: <1490194274-30569-1-git-send-email-Dave.Martin@arm.com>

Currently the vector length is inherited across both fork() and
exec().

Inheritance across fork() is desirable both for creating a copy of
a process (traditional fork) or creating a thread (where we want
all threads to share the same VL by default).

Inheritance across exec() is less desirable, because of the ABI
impact of large vector lengths on the size of the signal frame --
when running a new binary, there is no guarantee that the new
binary is compatible with these ABI changes.

This flag makes the vector length non-inherited by default.
Instead, the vector length is reset to a system default value,
unless the THREAD_VL_INHERIT flag has been set for the thread.

In order to permit clean launching of a new binary with a different
vector length, this patch also adds a prctl flag
PR_SVE_SET_VL_ONEXEC which causes the effect of the change to be
deferred until the calling thread's next exec.  This behaviour is
implemented by storing a shadow vector length in
thread_struct.sve_vl_onexec, which gets activated at exec.

Subsequent vl changes by the thread (if any) before the exec will
override such a pending change.

Without PR_SVE_SET_VL_ONEXEC, the effect of the change will be
immediate.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
---
 arch/arm64/include/asm/processor.h |  8 +++++
 arch/arm64/kernel/fpsimd.c         | 61 ++++++++++++++++++++++++++++++--------
 include/uapi/linux/prctl.h         |  6 ++++
 3 files changed, 63 insertions(+), 12 deletions(-)

diff --git a/arch/arm64/include/asm/processor.h b/arch/arm64/include/asm/processor.h
index 896e972..424fa5d 100644
--- a/arch/arm64/include/asm/processor.h
+++ b/arch/arm64/include/asm/processor.h
@@ -27,6 +27,7 @@
 
 #ifdef __KERNEL__
 
+#include <linux/prctl.h>
 #include <linux/string.h>
 
 #include <asm/alternative.h>
@@ -84,11 +85,18 @@ struct thread_struct {
 #endif
 	struct fpsimd_state	fpsimd_state;
 	u16			sve_vl;		/* SVE vector length */
+	u16			sve_vl_onexec;	/* SVE vl after next exec */
+	u16			sve_flags;	/* SVE related flags */
 	unsigned long		fault_address;	/* fault info */
 	unsigned long		fault_code;	/* ESR_EL1 value */
 	struct debug_info	debug;		/* debugging */
 };
 
+/* Flags for sve_flags (intentionally defined to match the prctl flags) */
+
+/* Inherit sve_vl and sve_flags across execve(): */
+#define THREAD_VL_INHERIT	PR_SVE_SET_VL_INHERIT
+
 #ifdef CONFIG_COMPAT
 #define task_user_tls(t)						\
 ({									\
diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
index ab00e9f..982b1d7 100644
--- a/arch/arm64/kernel/fpsimd.c
+++ b/arch/arm64/kernel/fpsimd.c
@@ -234,10 +234,11 @@ int sve_set_vector_length(struct task_struct *task,
 	 */
 	if (!(flags & PR_SVE_SET_VL_THREAD) && get_nr_threads(task) != 1)
 		return -EINVAL;
-
 	flags &= ~(unsigned long)PR_SVE_SET_VL_THREAD;
-	if (flags)
-		return -EINVAL; /* No other flags defined yet */
+
+	if (flags & ~(unsigned long)(PR_SVE_SET_VL_INHERIT |
+				     PR_SVE_SET_VL_ONEXEC))
+		return -EINVAL;
 
 	if (!sve_vl_valid(vl))
 		return -EINVAL;
@@ -247,6 +248,17 @@ int sve_set_vector_length(struct task_struct *task,
 		vl = sve_max_vl;
 	}
 
+	if (flags & (PR_SVE_SET_VL_ONEXEC |
+		     PR_SVE_SET_VL_INHERIT))
+		task->thread.sve_vl_onexec = vl;
+	else
+		/* Reset VL to system default on next exec: */
+		task->thread.sve_vl_onexec = 0;
+
+	/* Only actually set the VL if not deferred: */
+	if (flags & PR_SVE_SET_VL_ONEXEC)
+		goto out;
+
 	/*
 	 * To ensure the FPSIMD bits of the SVE vector registers are preserved,
 	 * write any live register state back to task_struct, and convert to a
@@ -272,9 +284,26 @@ int sve_set_vector_length(struct task_struct *task,
 
 	fpsimd_flush_task_state(task);
 
+out:
+	/* The THREAD_VL_* flag encodings match the relevant PR_* flags: */
+	task->thread.sve_flags = flags & PR_SVE_SET_VL_INHERIT;
+
 	return 0;
 }
 
+/*
+ * Encode the current vector length and flags for return.
+ * This is only required for prctl(): ptrace has separate fields
+ */
+static int sve_prctl_status(struct task_struct const *task)
+{
+	int ret = task->thread.sve_vl;
+
+	ret |= task->thread.sve_flags << 16;
+
+	return ret;
+}
+
 /* PR_SVE_SET_VL */
 int sve_set_task_vl(struct task_struct *task,
 		    unsigned long vector_length, unsigned long flags)
@@ -293,7 +322,7 @@ int sve_set_task_vl(struct task_struct *task,
 	if (ret)
 		return ret;
 
-	return task->thread.sve_vl;
+	return sve_prctl_status(task);
 }
 
 /* PR_SVE_GET_VL */
@@ -302,7 +331,7 @@ int sve_get_task_vl(struct task_struct *task)
 	if (!(elf_hwcap & HWCAP_SVE))
 		return -EINVAL;
 
-	return task->thread.sve_vl;
+	return sve_prctl_status(task);
 }
 
 #else /* ! CONFIG_ARM64_SVE */
@@ -459,17 +488,25 @@ void fpsimd_flush_thread(void)
 	if (IS_ENABLED(CONFIG_ARM64_SVE) && (elf_hwcap & HWCAP_SVE)) {
 		clear_sve_regs(current);
 
+		current->thread.sve_vl = current->thread.sve_vl_onexec ?
+			current->thread.sve_vl_onexec : sve_max_vl;
+
 		/*
 		 * User tasks must have a valid vector length set, but tasks
-		 * forked early (e.g., init) may not have one yet.
-		 * By now, we will know what the hardware supports, so set the
-		 * task vector length if it doesn't have one:
+		 * forked early (e.g., init) may not initially have one.
+		 * By now, we will know what the hardware supports, so
+		 * sve_max_vl should be valid, and thus the above
+		 * assignment should ensure a valid VL for the task.
+		 * If not, something went badly wrong.
 		 */
-		if (!current->thread.sve_vl) {
-			BUG_ON(!sve_vl_valid(sve_max_vl));
+		BUG_ON(!sve_vl_valid(current->thread.sve_vl));
 
-			current->thread.sve_vl = sve_max_vl;
-		}
+		/*
+		 * If the task is not set to inherit, ensure that the vector
+		 * length will be reset by a subsequent exec:
+		 */
+		if (!(current->thread.sve_flags & THREAD_VL_INHERIT))
+			current->thread.sve_vl_onexec = 0;
 	}
 
 	set_thread_flag(TIF_FOREIGN_FPSTATE);
diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
index c55530b..d56c447 100644
--- a/include/uapi/linux/prctl.h
+++ b/include/uapi/linux/prctl.h
@@ -200,6 +200,12 @@ struct prctl_mm_map {
 /* arm64 Scalable Vector Extension controls */
 #define PR_SVE_SET_VL			48	/* set task vector length */
 # define PR_SVE_SET_VL_THREAD		(1 << 1) /* set just this thread */
+# define PR_SVE_SET_VL_INHERIT		(1 << 2) /* inherit across exec */
+# define PR_SVE_SET_VL_ONEXEC		(1 << 3) /* defer effect until exec */
 #define PR_SVE_GET_VL			49	/* get task vector length */
+/* Decode helpers for the return value from PR_SVE_GET_VL: */
+# define PR_SVE_GET_VL_LEN(ret)		((ret) & 0x3fff) /* vector length */
+# define PR_SVE_GET_VL_INHERIT		(PR_SVE_SET_VL_INHERIT << 16)
+/* For conveinence, PR_SVE_SET_VL returns the result in the same encoding */
 
 #endif /* _LINUX_PRCTL_H */
-- 
2.1.4

  parent reply	other threads:[~2017-03-22 14:51 UTC|newest]

Thread overview: 75+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-03-22 14:50 [RFC PATCH v2 00/41] Scalable Vector Extension (SVE) core support Dave Martin
2017-03-22 14:50 ` Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 01/41] arm64: signal: Refactor sigcontext parsing in rt_sigreturn Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 02/41] arm64: signal: factor frame layout and population into separate passes Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 03/41] arm64: signal: factor out signal frame record allocation Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 04/41] arm64: signal: Allocate extra sigcontext space as needed Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 05/41] arm64: signal: Parse extra_context during sigreturn Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 06/41] arm64: efi: Add missing Kconfig dependency on KERNEL_MODE_NEON Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 07/41] arm64/sve: Allow kernel-mode NEON to be disabled in Kconfig Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 08/41] arm64/sve: Low-level save/restore code Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 09/41] arm64/sve: Boot-time feature detection and reporting Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 10/41] arm64/sve: Boot-time feature enablement Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 11/41] arm64/sve: Expand task_struct for Scalable Vector Extension state Dave Martin
2017-03-22 16:20   ` Mark Rutland
2017-03-23 10:49     ` Dave Martin
2017-03-23 11:26       ` Mark Rutland
2017-03-22 14:50 ` [RFC PATCH v2 12/41] arm64/sve: Save/restore SVE state on context switch paths Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 13/41] arm64/sve: [BROKEN] Basic support for KERNEL_MODE_NEON Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 14/41] Revert "arm64/sve: Allow kernel-mode NEON to be disabled in Kconfig" Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 15/41] arm64/sve: Restore working FPSIMD save/restore around signals Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 16/41] arm64/sve: signal: Add SVE state record to sigcontext Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 17/41] arm64/sve: signal: Dump Scalable Vector Extension registers to user stack Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 18/41] arm64/sve: signal: Restore FPSIMD/SVE state in rt_sigreturn Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 19/41] arm64/sve: Avoid corruption when replacing the SVE state Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 20/41] arm64/sve: traps: Add descriptive string for SVE exceptions Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 21/41] arm64/sve: Enable SVE on demand for userspace Dave Martin
2017-03-22 16:48   ` Mark Rutland
2017-03-23 11:24     ` Dave Martin
2017-03-23 11:30       ` Suzuki K Poulose
2017-03-23 11:52         ` Mark Rutland
2017-03-23 12:07           ` Dave Martin
2017-03-23 13:40             ` Mark Rutland
2017-03-23 13:45               ` Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 22/41] arm64/sve: Implement FPSIMD-only context for tasks not using SVE Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 23/41] arm64/sve: Move ZEN handling to the common task_fpsimd_load() path Dave Martin
2017-03-22 16:55   ` Mark Rutland
2017-03-23 11:52     ` Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 24/41] arm64/sve: Discard SVE state on system call Dave Martin
2017-03-22 17:03   ` Mark Rutland
2017-03-23 11:59     ` Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 25/41] arm64/sve: Avoid preempt_disable() during sigreturn Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 26/41] arm64/sve: Avoid stale user register state after SVE access exception Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 27/41] arm64/sve: ptrace support Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 28/41] arm64: KVM: Treat SVE use by guests as undefined instruction execution Dave Martin
2017-03-22 17:06   ` Mark Rutland
2017-03-23 12:10     ` Dave Martin
2017-03-22 14:50 ` [RFC PATCH v2 29/41] prctl: Add skeleton for PR_SVE_{SET,GET}_VL controls Dave Martin
2017-03-22 14:50   ` [RFC PATCH v2 29/41] prctl: Add skeleton for PR_SVE_{SET, GET}_VL controls Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 30/41] arm64/sve: Track vector length for each task Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 31/41] arm64/sve: Set CPU vector length to match current task Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 32/41] arm64/sve: Factor out clearing of tasks' SVE regs Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 33/41] arm64/sve: Wire up vector length control prctl() calls Dave Martin
2017-03-22 14:51   ` Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 34/41] arm64/sve: Disallow VL setting for individual threads by default Dave Martin
2017-03-22 14:51 ` Dave Martin [this message]
2017-03-22 14:51 ` [RFC PATCH v2 36/41] arm64/sve: ptrace: Wire up vector length control and reporting Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 37/41] arm64/sve: Enable default vector length control via procfs Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 38/41] arm64/sve: Detect SVE via the cpufeature framework Dave Martin
2017-03-23 14:11   ` Suzuki K Poulose
2017-03-23 14:37     ` Dave Martin
2017-03-23 14:43       ` Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 39/41] arm64/sve: Migrate to cpucap based detection for runtime SVE code Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 40/41] arm64/sve: Allocate task SVE context storage dynamically Dave Martin
2017-03-22 14:51 ` [RFC PATCH v2 41/41] arm64/sve: Documentation: Add overview of the SVE userspace ABI Dave Martin
2017-03-22 14:51   ` Dave Martin
2017-03-31 15:28 ` [RFC PATCH v2 00/41] Scalable Vector Extension (SVE) core support Ard Biesheuvel
2017-03-31 15:28   ` Ard Biesheuvel
2017-04-03  9:45   ` Dave Martin
2017-04-03  9:45     ` Dave Martin
2017-04-03 10:01     ` Ard Biesheuvel
2017-04-03 10:01       ` Ard Biesheuvel
2017-04-03 10:51       ` Dave Martin
2017-04-03 10:51         ` Dave Martin
2017-04-03 10:55         ` Ard Biesheuvel
2017-04-03 10:55           ` Ard Biesheuvel

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=1490194274-30569-36-git-send-email-Dave.Martin@arm.com \
    --to=dave.martin@arm.com \
    --cc=linux-arm-kernel@lists.infradead.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.