Merge tag 'loongarch-fixes-6.2-1' of git://git.kernel.org/pub/scm/lin…

…ux/kernel/git/chenhuacai/linux-loongson

Pull LoongArch fixes from Huacai Chen:
 "Fix a missing elf_hwcap, fix some stack unwinder bugs and two trivial
  cleanups"

* tag 'loongarch-fixes-6.2-1' of git://git.kernel.org/pub/scm/linux/kernel/git/chenhuacai/linux-loongson:
  LoongArch: Add generic ex-handler unwind in prologue unwinder
  LoongArch: Strip guess unwinder out from prologue unwinder
  LoongArch: Use correct sp value to get graph addr in stack unwinders
  LoongArch: Get frame info in unwind_start() when regs is not available
  LoongArch: Adjust PC value when unwind next frame in unwinder
  LoongArch: Simplify larch_insn_gen_xxx implementation
  LoongArch: Use common function sign_extend64()
  LoongArch: Add HWCAP_LOONGARCH_CPUCFG to elf_hwcap

arch/loongarch/include/asm/ftrace.h

@@ -10,8 +10,6 @@
 #define FTRACE_REGS_PLT_IDX	1
 #define NR_FTRACE_PLTS		2
 
-#define GRAPH_FAKE_OFFSET (sizeof(struct pt_regs) - offsetof(struct pt_regs, regs[1]))
-
 #ifdef CONFIG_FUNCTION_TRACER
 
 #define MCOUNT_INSN_SIZE 4		/* sizeof mcount call */

arch/loongarch/include/asm/inst.h

@@ -377,14 +377,6 @@ static inline bool unsigned_imm_check(unsigned long val, unsigned int bit)
 	return val < (1UL << bit);
 }
 
-static inline unsigned long sign_extend(unsigned long val, unsigned int idx)
-{
-	if (!is_imm_negative(val, idx + 1))
-		return ((1UL << idx) - 1) & val;
-	else
-		return ~((1UL << idx) - 1) | val;
-}
-
 #define DEF_EMIT_REG0I26_FORMAT(NAME, OP)				\
 static inline void emit_##NAME(union loongarch_instruction *insn,	\
 			       int offset)				\
@@ -401,6 +393,7 @@ static inline void emit_##NAME(union loongarch_instruction *insn,	\
 }
 
 DEF_EMIT_REG0I26_FORMAT(b, b_op)
+DEF_EMIT_REG0I26_FORMAT(bl, bl_op)
 
 #define DEF_EMIT_REG1I20_FORMAT(NAME, OP)				\
 static inline void emit_##NAME(union loongarch_instruction *insn,	\

arch/loongarch/include/asm/unwind.h

@@ -8,7 +8,9 @@
 #define _ASM_UNWIND_H
 
 #include <linux/sched.h>
+#include <linux/ftrace.h>
 
+#include <asm/ptrace.h>
 #include <asm/stacktrace.h>
 
 enum unwinder_type {
@@ -20,11 +22,13 @@ struct unwind_state {
 	char type; /* UNWINDER_XXX */
 	struct stack_info stack_info;
 	struct task_struct *task;
-	bool first, error, is_ftrace;
+	bool first, error, reset;
 	int graph_idx;
 	unsigned long sp, pc, ra;
 };
 
+bool default_next_frame(struct unwind_state *state);
+
 void unwind_start(struct unwind_state *state,
 		  struct task_struct *task, struct pt_regs *regs);
 bool unwind_next_frame(struct unwind_state *state);
@@ -40,4 +44,39 @@ static inline bool unwind_error(struct unwind_state *state)
 	return state->error;
 }
 
+#define GRAPH_FAKE_OFFSET (sizeof(struct pt_regs) - offsetof(struct pt_regs, regs[1]))
+
+static inline unsigned long unwind_graph_addr(struct unwind_state *state,
+					unsigned long pc, unsigned long cfa)
+{
+	return ftrace_graph_ret_addr(state->task, &state->graph_idx,
+				     pc, (unsigned long *)(cfa - GRAPH_FAKE_OFFSET));
+}
+
+static __always_inline void __unwind_start(struct unwind_state *state,
+					struct task_struct *task, struct pt_regs *regs)
+{
+	memset(state, 0, sizeof(*state));
+	if (regs) {
+		state->sp = regs->regs[3];
+		state->pc = regs->csr_era;
+		state->ra = regs->regs[1];
+	} else if (task && task != current) {
+		state->sp = thread_saved_fp(task);
+		state->pc = thread_saved_ra(task);
+		state->ra = 0;
+	} else {
+		state->sp = (unsigned long)__builtin_frame_address(0);
+		state->pc = (unsigned long)__builtin_return_address(0);
+		state->ra = 0;
+	}
+	state->task = task;
+	get_stack_info(state->sp, state->task, &state->stack_info);
+	state->pc = unwind_graph_addr(state, state->pc, state->sp);
+}
+
+static __always_inline unsigned long __unwind_get_return_address(struct unwind_state *state)
+{
+	return unwind_done(state) ? 0 : state->pc;
+}
 #endif /* _ASM_UNWIND_H */

arch/loongarch/kernel/Makefile

@@ -8,7 +8,7 @@ extra-y		:= vmlinux.lds
 obj-y		+= head.o cpu-probe.o cacheinfo.o env.o setup.o entry.o genex.o \
 		   traps.o irq.o idle.o process.o dma.o mem.o io.o reset.o switch.o \
 		   elf.o syscall.o signal.o time.o topology.o inst.o ptrace.o vdso.o \
-		   alternative.o unaligned.o
+		   alternative.o unaligned.o unwind.o
 
 obj-$(CONFIG_ACPI)		+= acpi.o
 obj-$(CONFIG_EFI) 		+= efi.o

arch/loongarch/kernel/alternative.c

@@ -74,7 +74,7 @@ static void __init_or_module recompute_jump(union loongarch_instruction *buf,
 	switch (src->reg0i26_format.opcode) {
 	case b_op:
 	case bl_op:
-		jump_addr = cur_pc + sign_extend((si_h << 16 | si_l) << 2, 27);
+		jump_addr = cur_pc + sign_extend64((si_h << 16 | si_l) << 2, 27);
 		if (in_alt_jump(jump_addr, start, end))
 			return;
 		offset = jump_addr - pc;
@@ -93,7 +93,7 @@ static void __init_or_module recompute_jump(union loongarch_instruction *buf,
 		fallthrough;
 	case beqz_op:
 	case bnez_op:
-		jump_addr = cur_pc + sign_extend((si_h << 16 | si_l) << 2, 22);
+		jump_addr = cur_pc + sign_extend64((si_h << 16 | si_l) << 2, 22);
 		if (in_alt_jump(jump_addr, start, end))
 			return;
 		offset = jump_addr - pc;
@@ -112,7 +112,7 @@ static void __init_or_module recompute_jump(union loongarch_instruction *buf,
 	case bge_op:
 	case bltu_op:
 	case bgeu_op:
-		jump_addr = cur_pc + sign_extend(si << 2, 17);
+		jump_addr = cur_pc + sign_extend64(si << 2, 17);
 		if (in_alt_jump(jump_addr, start, end))
 			return;
 		offset = jump_addr - pc;

arch/loongarch/kernel/cpu-probe.c

@@ -94,7 +94,7 @@ static void cpu_probe_common(struct cpuinfo_loongarch *c)
 	c->options = LOONGARCH_CPU_CPUCFG | LOONGARCH_CPU_CSR |
 		     LOONGARCH_CPU_TLB | LOONGARCH_CPU_VINT | LOONGARCH_CPU_WATCH;
 
-	elf_hwcap |= HWCAP_LOONGARCH_CRC32;
+	elf_hwcap = HWCAP_LOONGARCH_CPUCFG | HWCAP_LOONGARCH_CRC32;
 
 	config = read_cpucfg(LOONGARCH_CPUCFG1);
 	if (config & CPUCFG1_UAL) {

arch/loongarch/kernel/genex.S

@@ -67,14 +67,17 @@ SYM_FUNC_END(except_vec_cex)
 	.macro	BUILD_HANDLER exception handler prep
 	.align	5
 	SYM_FUNC_START(handle_\exception)
+	666:
 	BACKUP_T0T1
 	SAVE_ALL
 	build_prep_\prep
 	move	a0, sp
 	la.abs	t0, do_\handler
 	jirl	ra, t0, 0
+	668:
 	RESTORE_ALL_AND_RET
 	SYM_FUNC_END(handle_\exception)
+	SYM_DATA(unwind_hint_\exception, .word 668b - 666b)
 	.endm
 
 	BUILD_HANDLER ade ade badv

arch/loongarch/kernel/inst.c

@@ -58,47 +58,29 @@ u32 larch_insn_gen_nop(void)
 u32 larch_insn_gen_b(unsigned long pc, unsigned long dest)
 {
 	long offset = dest - pc;
-	unsigned int immediate_l, immediate_h;
 	union loongarch_instruction insn;
 
 	if ((offset & 3) || offset < -SZ_128M || offset >= SZ_128M) {
 		pr_warn("The generated b instruction is out of range.\n");
 		return INSN_BREAK;
 	}
 
-	offset >>= 2;
-
-	immediate_l = offset & 0xffff;
-	offset >>= 16;
-	immediate_h = offset & 0x3ff;
-
-	insn.reg0i26_format.opcode = b_op;
-	insn.reg0i26_format.immediate_l = immediate_l;
-	insn.reg0i26_format.immediate_h = immediate_h;
+	emit_b(&insn, offset >> 2);
 
 	return insn.word;
 }
 
 u32 larch_insn_gen_bl(unsigned long pc, unsigned long dest)
 {
 	long offset = dest - pc;
-	unsigned int immediate_l, immediate_h;
 	union loongarch_instruction insn;
 
 	if ((offset & 3) || offset < -SZ_128M || offset >= SZ_128M) {
 		pr_warn("The generated bl instruction is out of range.\n");
 		return INSN_BREAK;
 	}
 
-	offset >>= 2;
-
-	immediate_l = offset & 0xffff;
-	offset >>= 16;
-	immediate_h = offset & 0x3ff;
-
-	insn.reg0i26_format.opcode = bl_op;
-	insn.reg0i26_format.immediate_l = immediate_l;
-	insn.reg0i26_format.immediate_h = immediate_h;
+	emit_bl(&insn, offset >> 2);
 
 	return insn.word;
 }
@@ -107,10 +89,7 @@ u32 larch_insn_gen_or(enum loongarch_gpr rd, enum loongarch_gpr rj, enum loongar
 {
 	union loongarch_instruction insn;
 
-	insn.reg3_format.opcode = or_op;
-	insn.reg3_format.rd = rd;
-	insn.reg3_format.rj = rj;
-	insn.reg3_format.rk = rk;
+	emit_or(&insn, rd, rj, rk);
 
 	return insn.word;
 }
@@ -124,9 +103,7 @@ u32 larch_insn_gen_lu12iw(enum loongarch_gpr rd, int imm)
 {
 	union loongarch_instruction insn;
 
-	insn.reg1i20_format.opcode = lu12iw_op;
-	insn.reg1i20_format.rd = rd;
-	insn.reg1i20_format.immediate = imm;
+	emit_lu12iw(&insn, rd, imm);
 
 	return insn.word;
 }
@@ -135,9 +112,7 @@ u32 larch_insn_gen_lu32id(enum loongarch_gpr rd, int imm)
 {
 	union loongarch_instruction insn;
 
-	insn.reg1i20_format.opcode = lu32id_op;
-	insn.reg1i20_format.rd = rd;
-	insn.reg1i20_format.immediate = imm;
+	emit_lu32id(&insn, rd, imm);
 
 	return insn.word;
 }
@@ -146,10 +121,7 @@ u32 larch_insn_gen_lu52id(enum loongarch_gpr rd, enum loongarch_gpr rj, int imm)
 {
 	union loongarch_instruction insn;
 
-	insn.reg2i12_format.opcode = lu52id_op;
-	insn.reg2i12_format.rd = rd;
-	insn.reg2i12_format.rj = rj;
-	insn.reg2i12_format.immediate = imm;
+	emit_lu52id(&insn, rd, rj, imm);
 
 	return insn.word;
 }
@@ -158,10 +130,7 @@ u32 larch_insn_gen_jirl(enum loongarch_gpr rd, enum loongarch_gpr rj, unsigned l
 {
 	union loongarch_instruction insn;
 
-	insn.reg2i16_format.opcode = jirl_op;
-	insn.reg2i16_format.rd = rd;
-	insn.reg2i16_format.rj = rj;
-	insn.reg2i16_format.immediate = (dest - pc) >> 2;
+	emit_jirl(&insn, rj, rd, (dest - pc) >> 2);
 
 	return insn.word;
 }

arch/loongarch/kernel/process.c

@@ -191,20 +191,14 @@ int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
 
 unsigned long __get_wchan(struct task_struct *task)
 {
-	unsigned long pc;
+	unsigned long pc = 0;
 	struct unwind_state state;
 
 	if (!try_get_task_stack(task))
 		return 0;
 
-	unwind_start(&state, task, NULL);
-	state.sp = thread_saved_fp(task);
-	get_stack_info(state.sp, state.task, &state.stack_info);
-	state.pc = thread_saved_ra(task);
-#ifdef CONFIG_UNWINDER_PROLOGUE
-	state.type = UNWINDER_PROLOGUE;
-#endif
-	for (; !unwind_done(&state); unwind_next_frame(&state)) {
+	for (unwind_start(&state, task, NULL);
+	     !unwind_done(&state); unwind_next_frame(&state)) {
 		pc = unwind_get_return_address(&state);
 		if (!pc)
 			break;

arch/loongarch/kernel/traps.c

@@ -72,9 +72,6 @@ static void show_backtrace(struct task_struct *task, const struct pt_regs *regs,
 	if (!task)
 		task = current;
 
-	if (user_mode(regs))
-		state.type = UNWINDER_GUESS;
-
 	printk("%sCall Trace:", loglvl);
 	for (unwind_start(&state, task, pregs);
 	      !unwind_done(&state); unwind_next_frame(&state)) {

arch/loongarch/kernel/unwind.c

@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2022-2023 Loongson Technology Corporation Limited
+ */
+#include <linux/kernel.h>
+#include <linux/ftrace.h>
+
+#include <asm/unwind.h>
+
+bool default_next_frame(struct unwind_state *state)
+{
+	struct stack_info *info = &state->stack_info;
+	unsigned long addr;
+
+	if (unwind_done(state))
+		return false;
+
+	do {
+		for (state->sp += sizeof(unsigned long);
+		     state->sp < info->end; state->sp += sizeof(unsigned long)) {
+			addr = *(unsigned long *)(state->sp);
+			state->pc = unwind_graph_addr(state, addr, state->sp + 8);
+			if (__kernel_text_address(state->pc))
+				return true;
+		}
+
+		state->sp = info->next_sp;
+
+	} while (!get_stack_info(state->sp, state->task, info));
+
+	return false;
+}