s390/fpu: decrease stack usage for some cases

The kernel_fpu structure has a quite large size of 520 bytes. In order to
reduce stack footprint introduce several kernel fpu structures with
different and also smaller sizes. This way every kernel fpu user must use
the correct variant. A compile time check verifies that the correct variant
is used.

There are several users which use only 16 instead of all 32 vector
registers. For those users the new kernel_fpu_16 structure with a size of
only 266 bytes can be used.

Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

8 files changed, 96 insertions, 42 deletions

diff --git a/arch/s390/crypto/chacha-glue.c b/arch/s390/crypto/chacha-glue.c
index 97098add2079..f8b0c52e77a4 100644
--- a/arch/s390/crypto/chacha-glue.c
+++ b/arch/s390/crypto/chacha-glue.c
@@ -22,7 +22,7 @@ static void chacha20_crypt_s390(u32 *state, u8 *dst, const u8 *src,
 				unsigned int nbytes, const u32 *key,
 				u32 *counter)
 {
-	DECLARE_KERNEL_FPU_ONSTACK(vxstate);
+	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
 
 	kernel_fpu_begin(&vxstate, KERNEL_VXR);
 	chacha20_vx(dst, src, nbytes, key, counter);
diff --git a/arch/s390/crypto/crc32-vx.c b/arch/s390/crypto/crc32-vx.c
index dc2997f18e30..d9f1fdb66691 100644
--- a/arch/s390/crypto/crc32-vx.c
+++ b/arch/s390/crypto/crc32-vx.c
@@ -50,7 +50,7 @@ u32 crc32c_le_vgfm_16(u32 crc, unsigned char const *buf, size_t size);
 				unsigned char const *data, size_t datalen)  \
 	{								    \
 		unsigned long prealign, aligned, remaining;		    \
-		DECLARE_KERNEL_FPU_ONSTACK(vxstate);			    \
+		DECLARE_KERNEL_FPU_ONSTACK16(vxstate);			    \
 									    \
 		if (datalen < VX_MIN_LEN + VX_ALIGN_MASK)		    \
 			return ___crc32_sw(crc, data, datalen);		    \
diff --git a/arch/s390/include/asm/fpu-types.h b/arch/s390/include/asm/fpu-types.h
index 8e6927c23bdc..04c32b9fc849 100644
--- a/arch/s390/include/asm/fpu-types.h
+++ b/arch/s390/include/asm/fpu-types.h
@@ -16,14 +16,32 @@ struct fpu {
 	__vector128 vxrs[__NUM_VXRS] __aligned(8);
 };
 
-/* In-kernel FPU state structure */
+struct kernel_fpu_hdr {
+	int	mask;
+	u32	fpc;
+};
+
 struct kernel_fpu {
-	int	    mask;
-	u32	    fpc;
-	__vector128 vxrs[__NUM_VXRS] __aligned(8);
+	struct kernel_fpu_hdr hdr;
+	__vector128 vxrs[] __aligned(8);
 };
 
-#define DECLARE_KERNEL_FPU_ONSTACK(name)	\
-	struct kernel_fpu name __uninitialized
+#define KERNEL_FPU_STRUCT(vxr_size)				\
+struct kernel_fpu_##vxr_size {					\
+	struct kernel_fpu_hdr hdr;				\
+	__vector128 vxrs[vxr_size] __aligned(8);		\
+}
+
+KERNEL_FPU_STRUCT(16);
+KERNEL_FPU_STRUCT(32);
+
+#define DECLARE_KERNEL_FPU_ONSTACK(vxr_size, name)		\
+	struct kernel_fpu_##vxr_size name __uninitialized
+
+#define DECLARE_KERNEL_FPU_ONSTACK16(name)			\
+	DECLARE_KERNEL_FPU_ONSTACK(16, name)
+
+#define DECLARE_KERNEL_FPU_ONSTACK32(name)			\
+	DECLARE_KERNEL_FPU_ONSTACK(32, name)
 
 #endif /* _ASM_S390_FPU_TYPES_H */
diff --git a/arch/s390/include/asm/fpu.h b/arch/s390/include/asm/fpu.h
index e706af26c5d0..c1b3920092a1 100644
--- a/arch/s390/include/asm/fpu.h
+++ b/arch/s390/include/asm/fpu.h
@@ -162,28 +162,64 @@ static __always_inline void load_fp_regs_vx(__vector128 *vxrs)
 	__load_fp_regs(fprs, sizeof(__vector128) / sizeof(freg_t));
 }
 
-static inline void kernel_fpu_begin(struct kernel_fpu *state, int flags)
+static inline void _kernel_fpu_begin(struct kernel_fpu *state, int flags)
 {
-	state->mask = READ_ONCE(current->thread.kfpu_flags);
+	state->hdr.mask = READ_ONCE(current->thread.kfpu_flags);
 	if (!test_thread_flag(TIF_FPU)) {
 		/* Save user space FPU state and register contents */
 		save_user_fpu_regs();
-	} else if (state->mask & flags) {
+	} else if (state->hdr.mask & flags) {
 		/* Save FPU/vector register in-use by the kernel */
 		__kernel_fpu_begin(state, flags);
 	}
 	__atomic_or(flags, &current->thread.kfpu_flags);
 }
 
-static inline void kernel_fpu_end(struct kernel_fpu *state, int flags)
+static inline void _kernel_fpu_end(struct kernel_fpu *state, int flags)
 {
-	WRITE_ONCE(current->thread.kfpu_flags, state->mask);
-	if (state->mask & flags) {
+	WRITE_ONCE(current->thread.kfpu_flags, state->hdr.mask);
+	if (state->hdr.mask & flags) {
 		/* Restore FPU/vector register in-use by the kernel */
 		__kernel_fpu_end(state, flags);
 	}
 }
 
+void __kernel_fpu_invalid_size(void);
+
+static __always_inline void kernel_fpu_check_size(int flags, unsigned int size)
+{
+	unsigned int cnt = 0;
+
+	if (flags & KERNEL_VXR_V0V7)
+		cnt += 8;
+	if (flags & KERNEL_VXR_V8V15)
+		cnt += 8;
+	if (flags & KERNEL_VXR_V16V23)
+		cnt += 8;
+	if (flags & KERNEL_VXR_V24V31)
+		cnt += 8;
+	if (cnt != size)
+		__kernel_fpu_invalid_size();
+}
+
+#define kernel_fpu_begin(state, flags)					\
+{									\
+	typeof(state) s = (state);					\
+	int _flags = (flags);						\
+									\
+	kernel_fpu_check_size(_flags, ARRAY_SIZE(s->vxrs));		\
+	_kernel_fpu_begin((struct kernel_fpu *)s, _flags);		\
+}
+
+#define kernel_fpu_end(state, flags)					\
+{									\
+	typeof(state) s = (state);					\
+	int _flags = (flags);						\
+									\
+	kernel_fpu_check_size(_flags, ARRAY_SIZE(s->vxrs));		\
+	_kernel_fpu_end((struct kernel_fpu *)s, _flags);		\
+}
+
 static inline void save_kernel_fpu_regs(struct thread_struct *thread)
 {
 	struct fpu *state = &thread->kfpu;
diff --git a/arch/s390/kernel/fpu.c b/arch/s390/kernel/fpu.c
index 733e188951b7..62e9befe7890 100644
--- a/arch/s390/kernel/fpu.c
+++ b/arch/s390/kernel/fpu.c
@@ -19,41 +19,41 @@ void __kernel_fpu_begin(struct kernel_fpu *state, int flags)
 	 * Limit the save to the FPU/vector registers already
 	 * in use by the previous context.
 	 */
-	flags &= state->mask;
+	flags &= state->hdr.mask;
 	if (flags & KERNEL_FPC)
-		fpu_stfpc(&state->fpc);
+		fpu_stfpc(&state->hdr.fpc);
 	if (!cpu_has_vx()) {
 		if (flags & KERNEL_VXR_LOW)
-			save_fp_regs_vx(state->vxrs);
+			save_fp_regs_vx(vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR;
 	if (mask == KERNEL_VXR) {
-		fpu_vstm(0, 15, &vxrs[0]);
-		fpu_vstm(16, 31, &vxrs[16]);
+		vxrs += fpu_vstm(0, 15, vxrs);
+		vxrs += fpu_vstm(16, 31, vxrs);
 		return;
 	}
 	if (mask == KERNEL_VXR_MID) {
-		fpu_vstm(8, 23, &vxrs[8]);
+		vxrs += fpu_vstm(8, 23, vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR_LOW;
 	if (mask) {
 		if (mask == KERNEL_VXR_LOW)
-			fpu_vstm(0, 15, &vxrs[0]);
+			vxrs += fpu_vstm(0, 15, vxrs);
 		else if (mask == KERNEL_VXR_V0V7)
-			fpu_vstm(0, 7, &vxrs[0]);
+			vxrs += fpu_vstm(0, 7, vxrs);
 		else
-			fpu_vstm(8, 15, &vxrs[8]);
+			vxrs += fpu_vstm(8, 15, vxrs);
 	}
 	mask = flags & KERNEL_VXR_HIGH;
 	if (mask) {
 		if (mask == KERNEL_VXR_HIGH)
-			fpu_vstm(16, 31, &vxrs[16]);
+			vxrs += fpu_vstm(16, 31, vxrs);
 		else if (mask == KERNEL_VXR_V16V23)
-			fpu_vstm(16, 23, &vxrs[16]);
+			vxrs += fpu_vstm(16, 23, vxrs);
 		else
-			fpu_vstm(24, 31, &vxrs[24]);
+			vxrs += fpu_vstm(24, 31, vxrs);
 	}
 }
 EXPORT_SYMBOL(__kernel_fpu_begin);
@@ -68,41 +68,41 @@ void __kernel_fpu_end(struct kernel_fpu *state, int flags)
 	 * previous context that have been overwritten by the
 	 * current context.
 	 */
-	flags &= state->mask;
+	flags &= state->hdr.mask;
 	if (flags & KERNEL_FPC)
-		fpu_lfpc(&state->fpc);
+		fpu_lfpc(&state->hdr.fpc);
 	if (!cpu_has_vx()) {
 		if (flags & KERNEL_VXR_LOW)
-			load_fp_regs_vx(state->vxrs);
+			load_fp_regs_vx(vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR;
 	if (mask == KERNEL_VXR) {
-		fpu_vlm(0, 15, &vxrs[0]);
-		fpu_vlm(16, 31, &vxrs[16]);
+		vxrs += fpu_vlm(0, 15, vxrs);
+		vxrs += fpu_vlm(16, 31, vxrs);
 		return;
 	}
 	if (mask == KERNEL_VXR_MID) {
-		fpu_vlm(8, 23, &vxrs[8]);
+		vxrs += fpu_vlm(8, 23, vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR_LOW;
 	if (mask) {
 		if (mask == KERNEL_VXR_LOW)
-			fpu_vlm(0, 15, &vxrs[0]);
+			vxrs += fpu_vlm(0, 15, vxrs);
 		else if (mask == KERNEL_VXR_V0V7)
-			fpu_vlm(0, 7, &vxrs[0]);
+			vxrs += fpu_vlm(0, 7, vxrs);
 		else
-			fpu_vlm(8, 15, &vxrs[8]);
+			vxrs += fpu_vlm(8, 15, vxrs);
 	}
 	mask = flags & KERNEL_VXR_HIGH;
 	if (mask) {
 		if (mask == KERNEL_VXR_HIGH)
-			fpu_vlm(16, 31, &vxrs[16]);
+			vxrs += fpu_vlm(16, 31, vxrs);
 		else if (mask == KERNEL_VXR_V16V23)
-			fpu_vlm(16, 23, &vxrs[16]);
+			vxrs += fpu_vlm(16, 23, vxrs);
 		else
-			fpu_vlm(24, 31, &vxrs[24]);
+			vxrs += fpu_vlm(24, 31, vxrs);
 	}
 }
 EXPORT_SYMBOL(__kernel_fpu_end);
diff --git a/arch/s390/kernel/sysinfo.c b/arch/s390/kernel/sysinfo.c
index 061d45cf0261..4cd6428bfab2 100644
--- a/arch/s390/kernel/sysinfo.c
+++ b/arch/s390/kernel/sysinfo.c
@@ -426,7 +426,7 @@ subsys_initcall(create_proc_service_level);
  */
 void s390_adjust_jiffies(void)
 {
-	DECLARE_KERNEL_FPU_ONSTACK(fpu);
+	DECLARE_KERNEL_FPU_ONSTACK16(fpu);
 	struct sysinfo_1_2_2 *info;
 	unsigned long capability;
 
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 8467945344b5..8c222b0dfbf2 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -5026,7 +5026,7 @@ static void store_regs(struct kvm_vcpu *vcpu)
 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 {
 	struct kvm_run *kvm_run = vcpu->run;
-	DECLARE_KERNEL_FPU_ONSTACK(fpu);
+	DECLARE_KERNEL_FPU_ONSTACK32(fpu);
 	int rc;
 
 	/*
diff --git a/lib/raid6/s390vx.uc b/lib/raid6/s390vx.uc
index bc2f4fbe5a82..92c05b7596bc 100644
--- a/lib/raid6/s390vx.uc
+++ b/lib/raid6/s390vx.uc
@@ -80,7 +80,7 @@ static inline void COPY_VEC(int x, int y)
 
 static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
 {
-	DECLARE_KERNEL_FPU_ONSTACK(vxstate);
+	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
 	u8 **dptr, *p, *q;
 	int d, z, z0;
 
@@ -113,7 +113,7 @@ static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
 static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
 					size_t bytes, void **ptrs)
 {
-	DECLARE_KERNEL_FPU_ONSTACK(vxstate);
+	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
 	u8 **dptr, *p, *q;
 	int d, z, z0;