Received-SPF: pass (google.com: domain of linux-kernel+bounces-172545-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: "Chang S. Bae" <chang.seok.bae@intel.com>
To: linux-kernel@vger.kernel.org
Cc: x86@kernel.org,
	platform-driver-x86@vger.kernel.org,
	tglx@linutronix.de,
	mingo@redhat.com,
	bp@alien8.de,
	dave.hansen@linux.intel.com,
	hdegoede@redhat.com,
	ilpo.jarvinen@linux.intel.com,
	tony.luck@intel.com,
	ashok.raj@intel.com,
	jithu.joseph@intel.com,
	chang.seok.bae@intel.com
Subject: [PATCH v2 1/2] x86/fpu: Extend kernel_fpu_begin_mask() to initialize AMX state
Date: Tue,  7 May 2024 16:53:43 -0700
Message-Id: <20240507235344.249103-2-chang.seok.bae@intel.com>
In-Reply-To: <20240507235344.249103-1-chang.seok.bae@intel.com>
References: <20240430212508.105117-1-chang.seok.bae@intel.com>
 <20240507235344.249103-1-chang.seok.bae@intel.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

The In-Field Scan (IFS) test [1] is a destructive process, overwriting
the existing state to test the logic on the fly. As part of this test
process, the architectural state is saved before the test begins and
then restored upon completion.

However, due to resource constraints in storage, AMX state is excluded
from the scope of state recovery. Consequently, AMX state must be in its
initialized state for the IFS test to run.

When AMX workloads are running, an active user AMX state remains even
after a context switch, optimizing to reduce the state reload cost. In
such cases, the test cannot proceed if it is scheduled.

System administrators may attempt to mitigate this issue, by arranging
AMX workloads not to run on CPUs selected for the tests. However, this
approach is disruptive for managing large-scaled systems, diminishing the
benefit of the live testing.

The kernel can help by properly initializing the state before the test.
This initialization impacts the performance to some degree. But, this
approach is considerably cheaper than adding hardware resources and
simpler than a userspace approach.

While fpu_idle_fpregs() can initialize the AMX state, its usage should be
limited to specialized cases, primarily before entering the sleep state.
The restore_fpregs_from_fpstate() function offers a suitable mechanism
for initializing fpstate in general, which remains within the core code.

Extend kernel_fpu_begin_mask() to allow the IFS driver to initialize AMX
state through restore_fpregs_from_fpstate().

[1]: https://docs.kernel.org/arch/x86/ifs.html
Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>
---
V1 -> V2: Revise the changelog (Dave Hansen and Ashok Raj).

The recently published IFS documentation [2] elaborates its purpose and
the requirements of the context restoration after the scan test.

Additionally, the necessity for AMX initialization is emphasized in the
Intel Software Development Manual as of March 2024, in Section 18.2 of
Vol.1.

Side note: restore_fpregs_from_fpstate() also sets the x87 state to a
fixed value. However, this only applies to AMD CPUs with the FXSAVE_LEAK
quirk.

[2] IFS Technical Paper: Finding Faulty Components in a Live Fleet
    Environment
    https://www.intel.com/content/www/us/en/content-details/822279/finding-faulty-components-in-a-live-fleet-environment.html
---
 arch/x86/include/asm/fpu/api.h | 1 +
 arch/x86/kernel/fpu/core.c     | 3 +++
 2 files changed, 4 insertions(+)

diff --git a/arch/x86/include/asm/fpu/api.h b/arch/x86/include/asm/fpu/api.h
index a2be3aefff9f..67887fc45c24 100644
--- a/arch/x86/include/asm/fpu/api.h
+++ b/arch/x86/include/asm/fpu/api.h
@@ -25,6 +25,7 @@
 /* Kernel FPU states to initialize in kernel_fpu_begin_mask() */
 #define KFPU_387	_BITUL(0)	/* 387 state will be initialized */
 #define KFPU_MXCSR	_BITUL(1)	/* MXCSR will be initialized */
+#define KFPU_AMX	_BITUL(2)	/* AMX will be initialized */
 
 extern void kernel_fpu_begin_mask(unsigned int kfpu_mask);
 extern void kernel_fpu_end(void);
diff --git a/arch/x86/kernel/fpu/core.c b/arch/x86/kernel/fpu/core.c
index 1209c7aebb21..04cc6f14ca42 100644
--- a/arch/x86/kernel/fpu/core.c
+++ b/arch/x86/kernel/fpu/core.c
@@ -440,6 +440,9 @@ void kernel_fpu_begin_mask(unsigned int kfpu_mask)
 
 	if (unlikely(kfpu_mask & KFPU_387) && boot_cpu_has(X86_FEATURE_FPU))
 		asm volatile ("fninit");
+
+	if (unlikely(kfpu_mask & KFPU_AMX) && boot_cpu_has(X86_FEATURE_AMX_TILE))
+		restore_fpregs_from_fpstate(&init_fpstate, XFEATURE_MASK_XTILE);
 }
 EXPORT_SYMBOL_GPL(kernel_fpu_begin_mask);
 
-- 
2.34.1