diff mbox series

[4/7] riscv: add floating point insn support to misaligned access emulation

Message ID 20230926150316.1129648-5-cleger@rivosinc.com (mailing list archive)
State Superseded
Headers show
Series Add support to handle misaligned accesses in S-mode | expand

Checks

Context Check Description
conchuod/cover_letter success Series has a cover letter
conchuod/tree_selection success Guessed tree name to be for-next at HEAD 0bb80ecc33a8
conchuod/fixes_present success Fixes tag not required for -next series
conchuod/maintainers_pattern success MAINTAINERS pattern errors before the patch: 5 and now 5
conchuod/verify_signedoff success Signed-off-by tag matches author and committer
conchuod/kdoc success Errors and warnings before: 0 this patch: 0
conchuod/build_rv64_clang_allmodconfig fail Failed to build the tree with this patch.
conchuod/module_param success Was 0 now: 0
conchuod/build_rv64_gcc_allmodconfig success Errors and warnings before: 9 this patch: 9
conchuod/build_rv32_defconfig fail Build failed
conchuod/dtb_warn_rv64 success Errors and warnings before: 29 this patch: 29
conchuod/header_inline success No static functions without inline keyword in header files
conchuod/checkpatch fail CHECK: Macro argument 'insn' may be better as '(insn)' to avoid precedence issues CHECK: No space is necessary after a cast ERROR: open brace '{' following function definitions go on the next line WARNING: externs should be avoided in .c files
conchuod/build_rv64_nommu_k210_defconfig success Build OK
conchuod/verify_fixes success No Fixes tag
conchuod/build_rv64_nommu_virt_defconfig success Build OK
conchuod/vmtest-for-next-PR fail PR summary
conchuod/patch-4-test-1 fail .github/scripts/patches/build_rv32_defconfig.sh
conchuod/patch-4-test-2 fail .github/scripts/patches/build_rv64_clang_allmodconfig.sh
conchuod/patch-4-test-3 success .github/scripts/patches/build_rv64_gcc_allmodconfig.sh
conchuod/patch-4-test-4 success .github/scripts/patches/build_rv64_nommu_k210_defconfig.sh
conchuod/patch-4-test-5 success .github/scripts/patches/build_rv64_nommu_virt_defconfig.sh
conchuod/patch-4-test-6 fail .github/scripts/patches/checkpatch.sh
conchuod/patch-4-test-7 success .github/scripts/patches/dtb_warn_rv64.sh
conchuod/patch-4-test-8 success .github/scripts/patches/header_inline.sh
conchuod/patch-4-test-9 success .github/scripts/patches/kdoc.sh
conchuod/patch-4-test-10 success .github/scripts/patches/maintainers_patterns.sh
conchuod/patch-4-test-11 success .github/scripts/patches/module_param.sh
conchuod/patch-4-test-12 success .github/scripts/patches/verify_fixes.sh
conchuod/patch-4-test-13 success .github/scripts/patches/verify_signedoff.sh

Commit Message

Clément Léger Sept. 26, 2023, 3:03 p.m. UTC
This support is partially based of openSBI misaligned emulation floating
point instruction support. It provides support for the existing
floating point instructions (both for 32/64 bits as well as compressed
ones). Since floating point registers are not part of the pt_regs
struct, we need to modify them directly using some assembly. We also
dirty the pt_regs status in case we modify them to be sure context
switch will save FP state. With this support, Linux is on par with
openSBI support.

Signed-off-by: Clément Léger <cleger@rivosinc.com>
---
 arch/riscv/kernel/fpu.S              | 117 +++++++++++++++++++++
 arch/riscv/kernel/traps_misaligned.c | 152 ++++++++++++++++++++++++++-
 2 files changed, 265 insertions(+), 4 deletions(-)
diff mbox series

Patch

diff --git a/arch/riscv/kernel/fpu.S b/arch/riscv/kernel/fpu.S
index dd2205473de7..2785badb247c 100644
--- a/arch/riscv/kernel/fpu.S
+++ b/arch/riscv/kernel/fpu.S
@@ -104,3 +104,120 @@  ENTRY(__fstate_restore)
 	csrc CSR_STATUS, t1
 	ret
 ENDPROC(__fstate_restore)
+
+#define get_f32(which) fmv.x.s a0, which; j 2f
+#define put_f32(which) fmv.s.x which, a1; j 2f
+#if __riscv_xlen == 64
+# define get_f64(which) fmv.x.d a0, which; j 2f
+# define put_f64(which) fmv.d.x which, a1; j 2f
+#else
+# define get_f64(which) fsd which, 0(a1); j 2f
+# define put_f64(which) fld which, 0(a1); j 2f
+#endif
+
+.macro fp_access_prologue
+	/*
+	 * Compute jump offset to store the correct FP register since we don't
+	 * have indirect FP register access
+	 */
+	sll t0, a0, 3
+	la t2, 1f
+	add t0, t0, t2
+	li t1, SR_FS
+	csrs CSR_STATUS, t1
+	jr t0
+1:
+.endm
+
+.macro fp_access_epilogue
+2:
+	csrc CSR_STATUS, t1
+	ret
+.endm
+
+#define fp_access_body(__access_func) \
+	__access_func(f0); \
+	__access_func(f1); \
+	__access_func(f2); \
+	__access_func(f3); \
+	__access_func(f4); \
+	__access_func(f5); \
+	__access_func(f6); \
+	__access_func(f7); \
+	__access_func(f8); \
+	__access_func(f9); \
+	__access_func(f10); \
+	__access_func(f11); \
+	__access_func(f12); \
+	__access_func(f13); \
+	__access_func(f14); \
+	__access_func(f15); \
+	__access_func(f16); \
+	__access_func(f17); \
+	__access_func(f18); \
+	__access_func(f19); \
+	__access_func(f20); \
+	__access_func(f21); \
+	__access_func(f22); \
+	__access_func(f23); \
+	__access_func(f24); \
+	__access_func(f25); \
+	__access_func(f26); \
+	__access_func(f27); \
+	__access_func(f28); \
+	__access_func(f29); \
+	__access_func(f30); \
+	__access_func(f31)
+
+
+/*
+ * Disable compressed instructions set to keep a constant offset between FP
+ * load/store/move instructions
+ */
+.option norvc
+/*
+ * put_f32_reg - Set a FP register from a register containing the value
+ * a0 = FP register index to be set
+ * a1 = value to be loaded in the FP register
+ */
+SYM_FUNC_START(put_f32_reg)
+	fp_access_prologue
+	fp_access_body(put_f32)
+	fp_access_epilogue
+SYM_FUNC_END(put_f32_reg)
+
+/*
+ * get_f32_reg - Get a FP register value and return it
+ * a0 = FP register index to be retrieved
+ */
+SYM_FUNC_START(get_f32_reg)
+	fp_access_prologue
+	fp_access_body(get_f32)
+	fp_access_epilogue
+SYM_FUNC_END(put_f32_reg)
+
+/*
+ * put_f64_reg - Set a 64 bits FP register from a value or a pointer.
+ * a0 = FP register index to be set
+ * a1 = value/pointer to be loaded in the FP register (when xlen == 32 bits, we
+ * load the value to a pointer).
+ */
+SYM_FUNC_START(put_f64_reg)
+	fp_access_prologue
+	fp_access_body(put_f64)
+	fp_access_epilogue
+SYM_FUNC_END(put_f64_reg)
+
+/*
+ * put_f64_reg - Get a 64 bits FP register value and returned it or store it to
+ *	 	 a pointer.
+ * a0 = FP register index to be retrieved
+ * a1 = If xlen == 32, pointer which should be loaded with the FP register value
+ *	or unused if xlen == 64. In which case the FP register value is returned
+ *	through a0
+ */
+SYM_FUNC_START(get_f64_reg)
+	fp_access_prologue
+	fp_access_body(get_f64)
+	fp_access_epilogue
+SYM_FUNC_END(get_f64_reg)
diff --git a/arch/riscv/kernel/traps_misaligned.c b/arch/riscv/kernel/traps_misaligned.c
index 804f6c5e0e44..041fd2dbd955 100644
--- a/arch/riscv/kernel/traps_misaligned.c
+++ b/arch/riscv/kernel/traps_misaligned.c
@@ -153,6 +153,115 @@ 
 #define PRECISION_S 0
 #define PRECISION_D 1
 
+#ifdef CONFIG_FPU
+
+#define FP_GET_RD(insn)		(insn >> 7 & 0x1F)
+
+extern void put_f32_reg(unsigned long fp_reg, unsigned long value);
+
+static int set_f32_rd(unsigned long insn, struct pt_regs *regs,
+		      unsigned long val)
+{
+	unsigned long fp_reg = FP_GET_RD(insn);
+
+	put_f32_reg(fp_reg, val);
+	regs->status |= SR_FS_DIRTY;
+
+	return 0;
+}
+
+extern void put_f64_reg(unsigned long fp_reg, unsigned long value);
+
+static int set_f64_rd(unsigned long insn, struct pt_regs *regs, u64 val)
+{
+	unsigned long fp_reg = FP_GET_RD(insn);
+	unsigned long value;
+
+#if __riscv_xlen == 32
+	value = (unsigned long) &val;
+#else
+	value = val;
+#endif
+	put_f64_reg(fp_reg, value);
+	regs->status |= SR_FS_DIRTY;
+
+	return 0;
+}
+
+#if __riscv_xlen == 32
+extern void get_f64_reg(unsigned long fp_reg, u64 *value);
+
+static u64 get_f64_rs(unsigned long insn, u8 fp_reg_offset,
+		      struct pt_regs *regs)
+{
+	unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
+	u64 val;
+
+	get_f64_reg(fp_reg, &val);
+	regs->status |= SR_FS_DIRTY;
+
+	return val;
+}
+#else
+
+extern unsigned long get_f64_reg(unsigned long fp_reg);
+
+static unsigned long get_f64_rs(unsigned long insn, u8 fp_reg_offset,
+				struct pt_regs *regs)
+{
+	unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
+	unsigned long val;
+
+	val = get_f64_reg(fp_reg);
+	regs->status |= SR_FS_DIRTY;
+
+	return val;
+}
+
+#endif
+
+extern unsigned long get_f32_reg(unsigned long fp_reg);
+
+static unsigned long get_f32_rs(unsigned long insn, u8 fp_reg_offset,
+				struct pt_regs *regs)
+{
+	unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
+	unsigned long val;
+
+	val = get_f32_reg(fp_reg);
+	regs->status |= SR_FS_DIRTY;
+
+	return val;
+}
+
+#else /* CONFIG_FPU */
+static void set_f32_rd(unsigned long insn, struct pt_regs *regs,
+		       unsigned long val) {}
+
+static void set_f64_rd(unsigned long insn, struct pt_regs *regs, u64 val) {}
+
+static unsigned long get_f64_rs(unsigned long insn, u8 fp_reg_offset,
+				struct pt_regs *regs)
+{
+	return 0;
+}
+
+static unsigned long get_f32_rs(unsigned long insn, u8 fp_reg_offset,
+				struct pt_regs *regs)
+{
+	return 0;
+}
+
+#endif
+
+#define GET_F64_RS2(insn, regs) (get_f64_rs(insn, 20, regs))
+#define GET_F64_RS2C(insn, regs) (get_f64_rs(insn, 2, regs))
+#define GET_F64_RS2S(insn, regs) (get_f64_rs(RVC_RS2S(insn), 0, regs))
+
+#define GET_F32_RS2(insn, regs) (get_f32_rs(insn, 20, regs))
+#define GET_F32_RS2C(insn, regs) (get_f32_rs(insn, 2, regs))
+#define GET_F32_RS2S(insn, regs) (get_f32_rs(RVC_RS2S(insn), 0, regs))
+
 #ifdef CONFIG_RISCV_M_MODE
 static inline int load_u8(struct pt_regs *regs, const u8 *addr, u8 *r_val)
 {
@@ -362,15 +471,21 @@  int handle_misaligned_load(struct pt_regs *regs)
 		return -1;
 	}
 
+	if (!IS_ENABLED(CONFIG_FPU) && fp)
+		return -EOPNOTSUPP;
+
 	val.data_u64 = 0;
 	for (i = 0; i < len; i++) {
 		if (load_u8(regs, (void *)(addr + i), &val.data_bytes[i]))
 			return -1;
 	}
 
-	if (fp)
-		return -1;
-	SET_RD(insn, regs, val.data_ulong << shift >> shift);
+	if (!fp)
+		SET_RD(insn, regs, val.data_ulong << shift >> shift);
+	else if (len == 8)
+		set_f64_rd(insn, regs, val.data_u64);
+	else
+		set_f32_rd(insn, regs, val.data_ulong);
 
 	regs->epc = epc + INSN_LEN(insn);
 
@@ -383,7 +498,7 @@  int handle_misaligned_store(struct pt_regs *regs)
 	unsigned long epc = regs->epc;
 	unsigned long insn;
 	unsigned long addr = regs->badaddr;
-	int i, len = 0;
+	int i, len = 0, fp = 0;
 
 	perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
 
@@ -400,6 +515,14 @@  int handle_misaligned_store(struct pt_regs *regs)
 	} else if ((insn & INSN_MASK_SD) == INSN_MATCH_SD) {
 		len = 8;
 #endif
+	} else if ((insn & INSN_MASK_FSD) == INSN_MATCH_FSD) {
+		fp = 1;
+		len = 8;
+		val.data_u64 = GET_F64_RS2(insn, regs);
+	} else if ((insn & INSN_MASK_FSW) == INSN_MATCH_FSW) {
+		fp = 1;
+		len = 4;
+		val.data_ulong = GET_F32_RS2(insn, regs);
 	} else if ((insn & INSN_MASK_SH) == INSN_MATCH_SH) {
 		len = 2;
 #if defined(CONFIG_64BIT)
@@ -418,11 +541,32 @@  int handle_misaligned_store(struct pt_regs *regs)
 		   ((insn >> SH_RD) & 0x1f)) {
 		len = 4;
 		val.data_ulong = GET_RS2C(insn, regs);
+	} else if ((insn & INSN_MASK_C_FSD) == INSN_MATCH_C_FSD) {
+		fp = 1;
+		len = 8;
+		val.data_u64 = GET_F64_RS2S(insn, regs);
+	} else if ((insn & INSN_MASK_C_FSDSP) == INSN_MATCH_C_FSDSP) {
+		fp = 1;
+		len = 8;
+		val.data_u64 = GET_F64_RS2C(insn, regs);
+#if !defined(CONFIG_64BIT)
+	} else if ((insn & INSN_MASK_C_FSW) == INSN_MATCH_C_FSW) {
+		fp = 1;
+		len = 4;
+		val.data_ulong = GET_F32_RS2S(insn, regs);
+	} else if ((insn & INSN_MASK_C_FSWSP) == INSN_MATCH_C_FSWSP) {
+		fp = 1;
+		len = 4;
+		val.data_ulong = GET_F32_RS2C(insn, regs);
+#endif
 	} else {
 		regs->epc = epc;
 		return -1;
 	}
 
+	if (!IS_ENABLED(CONFIG_FPU) && fp)
+		return -EOPNOTSUPP;
+
 	for (i = 0; i < len; i++) {
 		if (store_u8(regs, (void *)(addr + i), val.data_bytes[i]))
 			return -1;