riscv: Unaligned load/store handling for M_MODE
authorDamien Le Moal <damien.lemoal@wdc.com>
Mon, 16 Mar 2020 00:47:36 +0000 (09:47 +0900)
committerPalmer Dabbelt <palmerdabbelt@google.com>
Fri, 3 Apr 2020 17:45:33 +0000 (10:45 -0700)
Add handlers for unaligned load and store traps that may be generated
by applications. Code heavily inspired from the OpenSBI project.
Handling of the unaligned access traps is suitable for applications
compiled with or without compressed instructions and is independent of
the kernel CONFIG_RISCV_ISA_C option value.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
arch/riscv/kernel/Makefile
arch/riscv/kernel/traps.c
arch/riscv/kernel/traps_misaligned.c [new file with mode: 0644]

index c121cc4..1bad93f 100644 (file)
@@ -30,7 +30,7 @@ obj-y += cacheinfo.o
 obj-y  += patch.o
 obj-$(CONFIG_MMU) += vdso.o vdso/
 
-obj-$(CONFIG_RISCV_M_MODE)     += clint.o
+obj-$(CONFIG_RISCV_M_MODE)     += clint.o traps_misaligned.o
 obj-$(CONFIG_FPU)              += fpu.o
 obj-$(CONFIG_SMP)              += smpboot.o
 obj-$(CONFIG_SMP)              += smp.o
index 23a57b9..29aef5c 100644 (file)
@@ -97,12 +97,33 @@ DO_ERROR_INFO(do_trap_insn_fault,
        SIGSEGV, SEGV_ACCERR, "instruction access fault");
 DO_ERROR_INFO(do_trap_insn_illegal,
        SIGILL, ILL_ILLOPC, "illegal instruction");
-DO_ERROR_INFO(do_trap_load_misaligned,
-       SIGBUS, BUS_ADRALN, "load address misaligned");
 DO_ERROR_INFO(do_trap_load_fault,
        SIGSEGV, SEGV_ACCERR, "load access fault");
+#ifndef CONFIG_RISCV_M_MODE
+DO_ERROR_INFO(do_trap_load_misaligned,
+       SIGBUS, BUS_ADRALN, "Oops - load address misaligned");
 DO_ERROR_INFO(do_trap_store_misaligned,
-       SIGBUS, BUS_ADRALN, "store (or AMO) address misaligned");
+       SIGBUS, BUS_ADRALN, "Oops - store (or AMO) address misaligned");
+#else
+int handle_misaligned_load(struct pt_regs *regs);
+int handle_misaligned_store(struct pt_regs *regs);
+
+asmlinkage void do_trap_load_misaligned(struct pt_regs *regs)
+{
+       if (!handle_misaligned_load(regs))
+               return;
+       do_trap_error(regs, SIGBUS, BUS_ADRALN, regs->epc,
+                     "Oops - load address misaligned");
+}
+
+asmlinkage void do_trap_store_misaligned(struct pt_regs *regs)
+{
+       if (!handle_misaligned_store(regs))
+               return;
+       do_trap_error(regs, SIGBUS, BUS_ADRALN, regs->epc,
+                     "Oops - store (or AMO) address misaligned");
+}
+#endif
 DO_ERROR_INFO(do_trap_store_fault,
        SIGSEGV, SEGV_ACCERR, "store (or AMO) access fault");
 DO_ERROR_INFO(do_trap_ecall_u,
diff --git a/arch/riscv/kernel/traps_misaligned.c b/arch/riscv/kernel/traps_misaligned.c
new file mode 100644 (file)
index 0000000..46c4daf
--- /dev/null
@@ -0,0 +1,370 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2020 Western Digital Corporation or its affiliates.
+ */
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/irq.h>
+
+#include <asm/processor.h>
+#include <asm/ptrace.h>
+#include <asm/csr.h>
+
+#define INSN_MATCH_LB                  0x3
+#define INSN_MASK_LB                   0x707f
+#define INSN_MATCH_LH                  0x1003
+#define INSN_MASK_LH                   0x707f
+#define INSN_MATCH_LW                  0x2003
+#define INSN_MASK_LW                   0x707f
+#define INSN_MATCH_LD                  0x3003
+#define INSN_MASK_LD                   0x707f
+#define INSN_MATCH_LBU                 0x4003
+#define INSN_MASK_LBU                  0x707f
+#define INSN_MATCH_LHU                 0x5003
+#define INSN_MASK_LHU                  0x707f
+#define INSN_MATCH_LWU                 0x6003
+#define INSN_MASK_LWU                  0x707f
+#define INSN_MATCH_SB                  0x23
+#define INSN_MASK_SB                   0x707f
+#define INSN_MATCH_SH                  0x1023
+#define INSN_MASK_SH                   0x707f
+#define INSN_MATCH_SW                  0x2023
+#define INSN_MASK_SW                   0x707f
+#define INSN_MATCH_SD                  0x3023
+#define INSN_MASK_SD                   0x707f
+
+#define INSN_MATCH_FLW                 0x2007
+#define INSN_MASK_FLW                  0x707f
+#define INSN_MATCH_FLD                 0x3007
+#define INSN_MASK_FLD                  0x707f
+#define INSN_MATCH_FLQ                 0x4007
+#define INSN_MASK_FLQ                  0x707f
+#define INSN_MATCH_FSW                 0x2027
+#define INSN_MASK_FSW                  0x707f
+#define INSN_MATCH_FSD                 0x3027
+#define INSN_MASK_FSD                  0x707f
+#define INSN_MATCH_FSQ                 0x4027
+#define INSN_MASK_FSQ                  0x707f
+
+#define INSN_MATCH_C_LD                        0x6000
+#define INSN_MASK_C_LD                 0xe003
+#define INSN_MATCH_C_SD                        0xe000
+#define INSN_MASK_C_SD                 0xe003
+#define INSN_MATCH_C_LW                        0x4000
+#define INSN_MASK_C_LW                 0xe003
+#define INSN_MATCH_C_SW                        0xc000
+#define INSN_MASK_C_SW                 0xe003
+#define INSN_MATCH_C_LDSP              0x6002
+#define INSN_MASK_C_LDSP               0xe003
+#define INSN_MATCH_C_SDSP              0xe002
+#define INSN_MASK_C_SDSP               0xe003
+#define INSN_MATCH_C_LWSP              0x4002
+#define INSN_MASK_C_LWSP               0xe003
+#define INSN_MATCH_C_SWSP              0xc002
+#define INSN_MASK_C_SWSP               0xe003
+
+#define INSN_MATCH_C_FLD               0x2000
+#define INSN_MASK_C_FLD                        0xe003
+#define INSN_MATCH_C_FLW               0x6000
+#define INSN_MASK_C_FLW                        0xe003
+#define INSN_MATCH_C_FSD               0xa000
+#define INSN_MASK_C_FSD                        0xe003
+#define INSN_MATCH_C_FSW               0xe000
+#define INSN_MASK_C_FSW                        0xe003
+#define INSN_MATCH_C_FLDSP             0x2002
+#define INSN_MASK_C_FLDSP              0xe003
+#define INSN_MATCH_C_FSDSP             0xa002
+#define INSN_MASK_C_FSDSP              0xe003
+#define INSN_MATCH_C_FLWSP             0x6002
+#define INSN_MASK_C_FLWSP              0xe003
+#define INSN_MATCH_C_FSWSP             0xe002
+#define INSN_MASK_C_FSWSP              0xe003
+
+#define INSN_LEN(insn)                 ((((insn) & 0x3) < 0x3) ? 2 : 4)
+
+#if defined(CONFIG_64BIT)
+#define LOG_REGBYTES                   3
+#define XLEN                           64
+#else
+#define LOG_REGBYTES                   2
+#define XLEN                           32
+#endif
+#define REGBYTES                       (1 << LOG_REGBYTES)
+#define XLEN_MINUS_16                  ((XLEN) - 16)
+
+#define SH_RD                          7
+#define SH_RS1                         15
+#define SH_RS2                         20
+#define SH_RS2C                                2
+
+#define RV_X(x, s, n)                  (((x) >> (s)) & ((1 << (n)) - 1))
+#define RVC_LW_IMM(x)                  ((RV_X(x, 6, 1) << 2) | \
+                                        (RV_X(x, 10, 3) << 3) | \
+                                        (RV_X(x, 5, 1) << 6))
+#define RVC_LD_IMM(x)                  ((RV_X(x, 10, 3) << 3) | \
+                                        (RV_X(x, 5, 2) << 6))
+#define RVC_LWSP_IMM(x)                        ((RV_X(x, 4, 3) << 2) | \
+                                        (RV_X(x, 12, 1) << 5) | \
+                                        (RV_X(x, 2, 2) << 6))
+#define RVC_LDSP_IMM(x)                        ((RV_X(x, 5, 2) << 3) | \
+                                        (RV_X(x, 12, 1) << 5) | \
+                                        (RV_X(x, 2, 3) << 6))
+#define RVC_SWSP_IMM(x)                        ((RV_X(x, 9, 4) << 2) | \
+                                        (RV_X(x, 7, 2) << 6))
+#define RVC_SDSP_IMM(x)                        ((RV_X(x, 10, 3) << 3) | \
+                                        (RV_X(x, 7, 3) << 6))
+#define RVC_RS1S(insn)                 (8 + RV_X(insn, SH_RD, 3))
+#define RVC_RS2S(insn)                 (8 + RV_X(insn, SH_RS2C, 3))
+#define RVC_RS2(insn)                  RV_X(insn, SH_RS2C, 5)
+
+#define SHIFT_RIGHT(x, y)              \
+       ((y) < 0 ? ((x) << -(y)) : ((x) >> (y)))
+
+#define REG_MASK                       \
+       ((1 << (5 + LOG_REGBYTES)) - (1 << LOG_REGBYTES))
+
+#define REG_OFFSET(insn, pos)          \
+       (SHIFT_RIGHT((insn), (pos) - LOG_REGBYTES) & REG_MASK)
+
+#define REG_PTR(insn, pos, regs)       \
+       (ulong *)((ulong)(regs) + REG_OFFSET(insn, pos))
+
+#define GET_RM(insn)                   (((insn) >> 12) & 7)
+
+#define GET_RS1(insn, regs)            (*REG_PTR(insn, SH_RS1, regs))
+#define GET_RS2(insn, regs)            (*REG_PTR(insn, SH_RS2, regs))
+#define GET_RS1S(insn, regs)           (*REG_PTR(RVC_RS1S(insn), 0, regs))
+#define GET_RS2S(insn, regs)           (*REG_PTR(RVC_RS2S(insn), 0, regs))
+#define GET_RS2C(insn, regs)           (*REG_PTR(insn, SH_RS2C, regs))
+#define GET_SP(regs)                   (*REG_PTR(2, 0, regs))
+#define SET_RD(insn, regs, val)                (*REG_PTR(insn, SH_RD, regs) = (val))
+#define IMM_I(insn)                    ((s32)(insn) >> 20)
+#define IMM_S(insn)                    (((s32)(insn) >> 25 << 5) | \
+                                        (s32)(((insn) >> 7) & 0x1f))
+#define MASK_FUNCT3                    0x7000
+
+#define GET_PRECISION(insn) (((insn) >> 25) & 3)
+#define GET_RM(insn) (((insn) >> 12) & 7)
+#define PRECISION_S 0
+#define PRECISION_D 1
+
+#define STR(x) XSTR(x)
+#define XSTR(x) #x
+
+#define DECLARE_UNPRIVILEGED_LOAD_FUNCTION(type, insn)                 \
+static inline type load_##type(const type *addr)                       \
+{                                                                      \
+       type val;                                                       \
+       asm (#insn " %0, %1"                                            \
+       : "=&r" (val) : "m" (*addr));                                   \
+       return val;                                                     \
+}
+
+#define DECLARE_UNPRIVILEGED_STORE_FUNCTION(type, insn)                        \
+static inline void store_##type(type *addr, type val)                  \
+{                                                                      \
+       asm volatile (#insn " %0, %1\n"                                 \
+       : : "r" (val), "m" (*addr));                                    \
+}
+
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(u8, lbu)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(u16, lhu)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(s8, lb)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(s16, lh)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(s32, lw)
+DECLARE_UNPRIVILEGED_STORE_FUNCTION(u8, sb)
+DECLARE_UNPRIVILEGED_STORE_FUNCTION(u16, sh)
+DECLARE_UNPRIVILEGED_STORE_FUNCTION(u32, sw)
+#if defined(CONFIG_64BIT)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(u32, lwu)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(u64, ld)
+DECLARE_UNPRIVILEGED_STORE_FUNCTION(u64, sd)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(ulong, ld)
+#else
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(u32, lw)
+DECLARE_UNPRIVILEGED_LOAD_FUNCTION(ulong, lw)
+
+static inline u64 load_u64(const u64 *addr)
+{
+       return load_u32((u32 *)addr)
+               + ((u64)load_u32((u32 *)addr + 1) << 32);
+}
+
+static inline void store_u64(u64 *addr, u64 val)
+{
+       store_u32((u32 *)addr, val);
+       store_u32((u32 *)addr + 1, val >> 32);
+}
+#endif
+
+static inline ulong get_insn(ulong mepc)
+{
+       register ulong __mepc asm ("a2") = mepc;
+       ulong val, rvc_mask = 3, tmp;
+
+       asm ("and %[tmp], %[addr], 2\n"
+               "bnez %[tmp], 1f\n"
+#if defined(CONFIG_64BIT)
+               STR(LWU) " %[insn], (%[addr])\n"
+#else
+               STR(LW) " %[insn], (%[addr])\n"
+#endif
+               "and %[tmp], %[insn], %[rvc_mask]\n"
+               "beq %[tmp], %[rvc_mask], 2f\n"
+               "sll %[insn], %[insn], %[xlen_minus_16]\n"
+               "srl %[insn], %[insn], %[xlen_minus_16]\n"
+               "j 2f\n"
+               "1:\n"
+               "lhu %[insn], (%[addr])\n"
+               "and %[tmp], %[insn], %[rvc_mask]\n"
+               "bne %[tmp], %[rvc_mask], 2f\n"
+               "lhu %[tmp], 2(%[addr])\n"
+               "sll %[tmp], %[tmp], 16\n"
+               "add %[insn], %[insn], %[tmp]\n"
+               "2:"
+       : [insn] "=&r" (val), [tmp] "=&r" (tmp)
+       : [addr] "r" (__mepc), [rvc_mask] "r" (rvc_mask),
+         [xlen_minus_16] "i" (XLEN_MINUS_16));
+
+       return val;
+}
+
+union reg_data {
+       u8 data_bytes[8];
+       ulong data_ulong;
+       u64 data_u64;
+};
+
+int handle_misaligned_load(struct pt_regs *regs)
+{
+       union reg_data val;
+       unsigned long epc = regs->epc;
+       unsigned long insn = get_insn(epc);
+       unsigned long addr = csr_read(mtval);
+       int i, fp = 0, shift = 0, len = 0;
+
+       regs->epc = 0;
+
+       if ((insn & INSN_MASK_LW) == INSN_MATCH_LW) {
+               len = 4;
+               shift = 8 * (sizeof(unsigned long) - len);
+#if defined(CONFIG_64BIT)
+       } else if ((insn & INSN_MASK_LD) == INSN_MATCH_LD) {
+               len = 8;
+               shift = 8 * (sizeof(unsigned long) - len);
+       } else if ((insn & INSN_MASK_LWU) == INSN_MATCH_LWU) {
+               len = 4;
+#endif
+       } else if ((insn & INSN_MASK_FLD) == INSN_MATCH_FLD) {
+               fp = 1;
+               len = 8;
+       } else if ((insn & INSN_MASK_FLW) == INSN_MATCH_FLW) {
+               fp = 1;
+               len = 4;
+       } else if ((insn & INSN_MASK_LH) == INSN_MATCH_LH) {
+               len = 2;
+               shift = 8 * (sizeof(unsigned long) - len);
+       } else if ((insn & INSN_MASK_LHU) == INSN_MATCH_LHU) {
+               len = 2;
+#if defined(CONFIG_64BIT)
+       } else if ((insn & INSN_MASK_C_LD) == INSN_MATCH_C_LD) {
+               len = 8;
+               shift = 8 * (sizeof(unsigned long) - len);
+               insn = RVC_RS2S(insn) << SH_RD;
+       } else if ((insn & INSN_MASK_C_LDSP) == INSN_MATCH_C_LDSP &&
+                  ((insn >> SH_RD) & 0x1f)) {
+               len = 8;
+               shift = 8 * (sizeof(unsigned long) - len);
+#endif
+       } else if ((insn & INSN_MASK_C_LW) == INSN_MATCH_C_LW) {
+               len = 4;
+               shift = 8 * (sizeof(unsigned long) - len);
+               insn = RVC_RS2S(insn) << SH_RD;
+       } else if ((insn & INSN_MASK_C_LWSP) == INSN_MATCH_C_LWSP &&
+                  ((insn >> SH_RD) & 0x1f)) {
+               len = 4;
+               shift = 8 * (sizeof(unsigned long) - len);
+       } else if ((insn & INSN_MASK_C_FLD) == INSN_MATCH_C_FLD) {
+               fp = 1;
+               len = 8;
+               insn = RVC_RS2S(insn) << SH_RD;
+       } else if ((insn & INSN_MASK_C_FLDSP) == INSN_MATCH_C_FLDSP) {
+               fp = 1;
+               len = 8;
+#if defined(CONFIG_32BIT)
+       } else if ((insn & INSN_MASK_C_FLW) == INSN_MATCH_C_FLW) {
+               fp = 1;
+               len = 4;
+               insn = RVC_RS2S(insn) << SH_RD;
+       } else if ((insn & INSN_MASK_C_FLWSP) == INSN_MATCH_C_FLWSP) {
+               fp = 1;
+               len = 4;
+#endif
+       } else {
+               regs->epc = epc;
+               return -1;
+       }
+
+       val.data_u64 = 0;
+       for (i = 0; i < len; i++)
+               val.data_bytes[i] = load_u8((void *)(addr + i));
+
+       if (fp)
+               return -1;
+       SET_RD(insn, regs, val.data_ulong << shift >> shift);
+
+       regs->epc = epc + INSN_LEN(insn);
+
+       return 0;
+}
+
+int handle_misaligned_store(struct pt_regs *regs)
+{
+       union reg_data val;
+       unsigned long epc = regs->epc;
+       unsigned long insn = get_insn(epc);
+       unsigned long addr = csr_read(mtval);
+       int i, len = 0;
+
+       regs->epc = 0;
+
+       val.data_ulong = GET_RS2(insn, regs);
+
+       if ((insn & INSN_MASK_SW) == INSN_MATCH_SW) {
+               len = 4;
+#if defined(CONFIG_64BIT)
+       } else if ((insn & INSN_MASK_SD) == INSN_MATCH_SD) {
+               len = 8;
+#endif
+       } else if ((insn & INSN_MASK_SH) == INSN_MATCH_SH) {
+               len = 2;
+#if defined(CONFIG_64BIT)
+       } else if ((insn & INSN_MASK_C_SD) == INSN_MATCH_C_SD) {
+               len = 8;
+               val.data_ulong = GET_RS2S(insn, regs);
+       } else if ((insn & INSN_MASK_C_SDSP) == INSN_MATCH_C_SDSP &&
+                  ((insn >> SH_RD) & 0x1f)) {
+               len = 8;
+               val.data_ulong = GET_RS2C(insn, regs);
+#endif
+       } else if ((insn & INSN_MASK_C_SW) == INSN_MATCH_C_SW) {
+               len = 4;
+               val.data_ulong = GET_RS2S(insn, regs);
+       } else if ((insn & INSN_MASK_C_SWSP) == INSN_MATCH_C_SWSP &&
+                  ((insn >> SH_RD) & 0x1f)) {
+               len = 4;
+               val.data_ulong = GET_RS2C(insn, regs);
+       } else {
+               regs->epc = epc;
+               return -1;
+       }
+
+       for (i = 0; i < len; i++)
+               store_u8((void *)(addr + i), val.data_bytes[i]);
+
+       regs->epc = epc + INSN_LEN(insn);
+
+       return 0;
+}