When we configure SVE, SSVE or ZA via ptrace we allow the user to configure
the vector length and specify any of the flags that are accepted when
configuring via prctl(). This includes the S[VM]E_SET_VL_ONEXEC flag which
defers the configuration of the VL until an exec(). We don't do anything to
limit the provision of register data as part of configuring the _ONEXEC VL
but as a function of the VL enumeration support we do this will be
interpreted using the vector length currently configured for the process.

This is all a bit surprising, and probably we should just not have allowed
register data to be specified with _ONEXEC, but it's our ABI so let's
add some explicit documentation in both the ABI documents and the source
calling out what happens.

The comments are also missing the fact that since SME does not have a
mandatory 128 bit VL it is possible for VL enumeration to result in the
configuration of a higher VL than was requested, cover that too.

Signed-off-by: Mark Brown <broonie@kernel.org>
---
 Documentation/arch/arm64/sme.rst |  4 ++++
 Documentation/arch/arm64/sve.rst |  4 ++++
 arch/arm64/kernel/ptrace.c       | 12 ++++++++++--
 3 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/Documentation/arch/arm64/sme.rst b/Documentation/arch/arm64/sme.rst
index be317d457417434c24811edce3ef2bad88f2cace..b2fa01f85cb5e33350c59f752aba9248e4fbd80e 100644
--- a/Documentation/arch/arm64/sme.rst
+++ b/Documentation/arch/arm64/sme.rst
@@ -346,6 +346,10 @@ The regset data starts with struct user_za_header, containing:
 
 * Writes to NT_ARM_ZT will set PSTATE.ZA to 1.
 
+* If any register data is provided along with SME_PT_VL_ONEXEC then the
+  registers data will be interpreted with the current vector length, not
+  the vector length configured for use on exec.
+
 
 8.  ELF coredump extensions
 ---------------------------
diff --git a/Documentation/arch/arm64/sve.rst b/Documentation/arch/arm64/sve.rst
index 8d8837fc39ec71414a72dfef9e834f0800106fdd..28152492c29cbc7a81165a367e87d4aa327690f4 100644
--- a/Documentation/arch/arm64/sve.rst
+++ b/Documentation/arch/arm64/sve.rst
@@ -402,6 +402,10 @@ The regset data starts with struct user_sve_header, containing:
   streaming mode and any SETREGSET of NT_ARM_SSVE will enter streaming mode
   if the target was not in streaming mode.
 
+* If any register data is provided along with SVE_PT_VL_ONEXEC then the
+  registers data will be interpreted with the current vector length, not
+  the vector length configured for use on exec.
+
 * The effect of writing a partial, incomplete payload is unspecified.
 
 
diff --git a/arch/arm64/kernel/ptrace.c b/arch/arm64/kernel/ptrace.c
index b756578aeaeea1d3250276734520e3eaae8a671d..f09ffd70c9166aeb306376a34316499010376044 100644
--- a/arch/arm64/kernel/ptrace.c
+++ b/arch/arm64/kernel/ptrace.c
@@ -898,7 +898,11 @@ static int sve_set_common(struct task_struct *target,
 	if (ret)
 		goto out;
 
-	/* Actual VL set may be less than the user asked for: */
+	/*
+	 * Actual VL set may be different from what the user asked
+	 * for, or we may have configured the _ONEXEC VL not the
+	 * current VL:
+	 */
 	vq = sve_vq_from_vl(task_get_vl(target, type));
 
 	/* Enter/exit streaming mode */
@@ -1125,7 +1129,11 @@ static int za_set(struct task_struct *target,
 	if (ret)
 		goto out;
 
-	/* Actual VL set may be less than the user asked for: */
+	/*
+	 * Actual VL set may be different from what the user asked
+	 * for, or we may have configured the _ONEXEC rather than
+	 * current VL:
+	 */
 	vq = sve_vq_from_vl(task_get_sme_vl(target));
 
 	/* Ensure there is some SVE storage for streaming mode */

---
base-commit: 8e929cb546ee42c9a61d24fae60605e9e3192354
change-id: 20241101-arm64-sve-ptrace-vl-set-db53eadbfc91

Best regards,
-- 
Mark Brown <broonie@kernel.org>

On Wed, 06 Nov 2024 17:41:32 +0000, Mark Brown wrote:
> When we configure SVE, SSVE or ZA via ptrace we allow the user to configure
> the vector length and specify any of the flags that are accepted when
> configuring via prctl(). This includes the S[VM]E_SET_VL_ONEXEC flag which
> defers the configuration of the VL until an exec(). We don't do anything to
> limit the provision of register data as part of configuring the _ONEXEC VL
> but as a function of the VL enumeration support we do this will be
> interpreted using the vector length currently configured for the process.
> 
> [...]

Applied to arm64 (for-next/misc), thanks!

[1/1] arm64/ptrace: Clarify documentation of VL configuration via ptrace
      https://git.kernel.org/arm64/c/c0139f6cbb1f

-- 
Catalin