summaryrefslogtreecommitdiffstats
path: root/arch/s390/kernel/fpu.c
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-06 01:02:30 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-06 01:02:30 +0000
commit76cb841cb886eef6b3bee341a2266c76578724ad (patch)
treef5892e5ba6cc11949952a6ce4ecbe6d516d6ce58 /arch/s390/kernel/fpu.c
parentInitial commit. (diff)
downloadlinux-76cb841cb886eef6b3bee341a2266c76578724ad.tar.xz
linux-76cb841cb886eef6b3bee341a2266c76578724ad.zip
Adding upstream version 4.19.249.upstream/4.19.249upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'arch/s390/kernel/fpu.c')
-rw-r--r--arch/s390/kernel/fpu.c177
1 files changed, 177 insertions, 0 deletions
diff --git a/arch/s390/kernel/fpu.c b/arch/s390/kernel/fpu.c
new file mode 100644
index 000000000..594464f21
--- /dev/null
+++ b/arch/s390/kernel/fpu.c
@@ -0,0 +1,177 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * In-kernel vector facility support functions
+ *
+ * Copyright IBM Corp. 2015
+ * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
+ */
+#include <linux/kernel.h>
+#include <linux/cpu.h>
+#include <linux/sched.h>
+#include <asm/fpu/types.h>
+#include <asm/fpu/api.h>
+
+asm(".include \"asm/vx-insn.h\"\n");
+
+void __kernel_fpu_begin(struct kernel_fpu *state, u32 flags)
+{
+ /*
+ * Limit the save to the FPU/vector registers already
+ * in use by the previous context
+ */
+ flags &= state->mask;
+
+ if (flags & KERNEL_FPC)
+ /* Save floating point control */
+ asm volatile("stfpc %0" : "=m" (state->fpc));
+
+ if (!MACHINE_HAS_VX) {
+ if (flags & KERNEL_VXR_V0V7) {
+ /* Save floating-point registers */
+ asm volatile("std 0,%0" : "=Q" (state->fprs[0]));
+ asm volatile("std 1,%0" : "=Q" (state->fprs[1]));
+ asm volatile("std 2,%0" : "=Q" (state->fprs[2]));
+ asm volatile("std 3,%0" : "=Q" (state->fprs[3]));
+ asm volatile("std 4,%0" : "=Q" (state->fprs[4]));
+ asm volatile("std 5,%0" : "=Q" (state->fprs[5]));
+ asm volatile("std 6,%0" : "=Q" (state->fprs[6]));
+ asm volatile("std 7,%0" : "=Q" (state->fprs[7]));
+ asm volatile("std 8,%0" : "=Q" (state->fprs[8]));
+ asm volatile("std 9,%0" : "=Q" (state->fprs[9]));
+ asm volatile("std 10,%0" : "=Q" (state->fprs[10]));
+ asm volatile("std 11,%0" : "=Q" (state->fprs[11]));
+ asm volatile("std 12,%0" : "=Q" (state->fprs[12]));
+ asm volatile("std 13,%0" : "=Q" (state->fprs[13]));
+ asm volatile("std 14,%0" : "=Q" (state->fprs[14]));
+ asm volatile("std 15,%0" : "=Q" (state->fprs[15]));
+ }
+ return;
+ }
+
+ /* Test and save vector registers */
+ asm volatile (
+ /*
+ * Test if any vector register must be saved and, if so,
+ * test if all register can be saved.
+ */
+ " la 1,%[vxrs]\n" /* load save area */
+ " tmll %[m],30\n" /* KERNEL_VXR */
+ " jz 7f\n" /* no work -> done */
+ " jo 5f\n" /* -> save V0..V31 */
+ /*
+ * Test for special case KERNEL_FPU_MID only. In this
+ * case a vstm V8..V23 is the best instruction
+ */
+ " chi %[m],12\n" /* KERNEL_VXR_MID */
+ " jne 0f\n" /* -> save V8..V23 */
+ " VSTM 8,23,128,1\n" /* vstm %v8,%v23,128(%r1) */
+ " j 7f\n"
+ /* Test and save the first half of 16 vector registers */
+ "0: tmll %[m],6\n" /* KERNEL_VXR_LOW */
+ " jz 3f\n" /* -> KERNEL_VXR_HIGH */
+ " jo 2f\n" /* 11 -> save V0..V15 */
+ " brc 2,1f\n" /* 10 -> save V8..V15 */
+ " VSTM 0,7,0,1\n" /* vstm %v0,%v7,0(%r1) */
+ " j 3f\n"
+ "1: VSTM 8,15,128,1\n" /* vstm %v8,%v15,128(%r1) */
+ " j 3f\n"
+ "2: VSTM 0,15,0,1\n" /* vstm %v0,%v15,0(%r1) */
+ /* Test and save the second half of 16 vector registers */
+ "3: tmll %[m],24\n" /* KERNEL_VXR_HIGH */
+ " jz 7f\n"
+ " jo 6f\n" /* 11 -> save V16..V31 */
+ " brc 2,4f\n" /* 10 -> save V24..V31 */
+ " VSTM 16,23,256,1\n" /* vstm %v16,%v23,256(%r1) */
+ " j 7f\n"
+ "4: VSTM 24,31,384,1\n" /* vstm %v24,%v31,384(%r1) */
+ " j 7f\n"
+ "5: VSTM 0,15,0,1\n" /* vstm %v0,%v15,0(%r1) */
+ "6: VSTM 16,31,256,1\n" /* vstm %v16,%v31,256(%r1) */
+ "7:"
+ : [vxrs] "=Q" (*(struct vx_array *) &state->vxrs)
+ : [m] "d" (flags)
+ : "1", "cc");
+}
+EXPORT_SYMBOL(__kernel_fpu_begin);
+
+void __kernel_fpu_end(struct kernel_fpu *state, u32 flags)
+{
+ /*
+ * Limit the restore to the FPU/vector registers of the
+ * previous context that have been overwritte by the
+ * current context
+ */
+ flags &= state->mask;
+
+ if (flags & KERNEL_FPC)
+ /* Restore floating-point controls */
+ asm volatile("lfpc %0" : : "Q" (state->fpc));
+
+ if (!MACHINE_HAS_VX) {
+ if (flags & KERNEL_VXR_V0V7) {
+ /* Restore floating-point registers */
+ asm volatile("ld 0,%0" : : "Q" (state->fprs[0]));
+ asm volatile("ld 1,%0" : : "Q" (state->fprs[1]));
+ asm volatile("ld 2,%0" : : "Q" (state->fprs[2]));
+ asm volatile("ld 3,%0" : : "Q" (state->fprs[3]));
+ asm volatile("ld 4,%0" : : "Q" (state->fprs[4]));
+ asm volatile("ld 5,%0" : : "Q" (state->fprs[5]));
+ asm volatile("ld 6,%0" : : "Q" (state->fprs[6]));
+ asm volatile("ld 7,%0" : : "Q" (state->fprs[7]));
+ asm volatile("ld 8,%0" : : "Q" (state->fprs[8]));
+ asm volatile("ld 9,%0" : : "Q" (state->fprs[9]));
+ asm volatile("ld 10,%0" : : "Q" (state->fprs[10]));
+ asm volatile("ld 11,%0" : : "Q" (state->fprs[11]));
+ asm volatile("ld 12,%0" : : "Q" (state->fprs[12]));
+ asm volatile("ld 13,%0" : : "Q" (state->fprs[13]));
+ asm volatile("ld 14,%0" : : "Q" (state->fprs[14]));
+ asm volatile("ld 15,%0" : : "Q" (state->fprs[15]));
+ }
+ return;
+ }
+
+ /* Test and restore (load) vector registers */
+ asm volatile (
+ /*
+ * Test if any vector register must be loaded and, if so,
+ * test if all registers can be loaded at once.
+ */
+ " la 1,%[vxrs]\n" /* load restore area */
+ " tmll %[m],30\n" /* KERNEL_VXR */
+ " jz 7f\n" /* no work -> done */
+ " jo 5f\n" /* -> restore V0..V31 */
+ /*
+ * Test for special case KERNEL_FPU_MID only. In this
+ * case a vlm V8..V23 is the best instruction
+ */
+ " chi %[m],12\n" /* KERNEL_VXR_MID */
+ " jne 0f\n" /* -> restore V8..V23 */
+ " VLM 8,23,128,1\n" /* vlm %v8,%v23,128(%r1) */
+ " j 7f\n"
+ /* Test and restore the first half of 16 vector registers */
+ "0: tmll %[m],6\n" /* KERNEL_VXR_LOW */
+ " jz 3f\n" /* -> KERNEL_VXR_HIGH */
+ " jo 2f\n" /* 11 -> restore V0..V15 */
+ " brc 2,1f\n" /* 10 -> restore V8..V15 */
+ " VLM 0,7,0,1\n" /* vlm %v0,%v7,0(%r1) */
+ " j 3f\n"
+ "1: VLM 8,15,128,1\n" /* vlm %v8,%v15,128(%r1) */
+ " j 3f\n"
+ "2: VLM 0,15,0,1\n" /* vlm %v0,%v15,0(%r1) */
+ /* Test and restore the second half of 16 vector registers */
+ "3: tmll %[m],24\n" /* KERNEL_VXR_HIGH */
+ " jz 7f\n"
+ " jo 6f\n" /* 11 -> restore V16..V31 */
+ " brc 2,4f\n" /* 10 -> restore V24..V31 */
+ " VLM 16,23,256,1\n" /* vlm %v16,%v23,256(%r1) */
+ " j 7f\n"
+ "4: VLM 24,31,384,1\n" /* vlm %v24,%v31,384(%r1) */
+ " j 7f\n"
+ "5: VLM 0,15,0,1\n" /* vlm %v0,%v15,0(%r1) */
+ "6: VLM 16,31,256,1\n" /* vlm %v16,%v31,256(%r1) */
+ "7:"
+ : [vxrs] "=Q" (*(struct vx_array *) &state->vxrs)
+ : [m] "d" (flags)
+ : "1", "cc");
+}
+EXPORT_SYMBOL(__kernel_fpu_end);