ARM atomic ops rewrite

Implement ARMv6 atomic ops (ldrex/strex) using the same trick as PPC.

Signed-off-by: Paul Brook <paul@codesourcery.com>
This commit is contained in:
Paul Brook 2009-11-22 21:35:13 +00:00
parent eee4850422
commit 426f5abcaa
5 changed files with 251 additions and 176 deletions

View File

@ -524,6 +524,81 @@ do_kernel_trap(CPUARMState *env)
return 0;
}
static int do_strex(CPUARMState *env)
{
uint32_t val;
int size;
int rc = 1;
int segv = 0;
uint32_t addr;
start_exclusive();
addr = env->exclusive_addr;
if (addr != env->exclusive_test) {
goto fail;
}
size = env->exclusive_info & 0xf;
switch (size) {
case 0:
segv = get_user_u8(val, addr);
break;
case 1:
segv = get_user_u16(val, addr);
break;
case 2:
case 3:
segv = get_user_u32(val, addr);
break;
}
if (segv) {
env->cp15.c6_data = addr;
goto done;
}
if (val != env->exclusive_val) {
goto fail;
}
if (size == 3) {
segv = get_user_u32(val, addr + 4);
if (segv) {
env->cp15.c6_data = addr + 4;
goto done;
}
if (val != env->exclusive_high) {
goto fail;
}
}
val = env->regs[(env->exclusive_info >> 8) & 0xf];
switch (size) {
case 0:
segv = put_user_u8(val, addr);
break;
case 1:
segv = put_user_u16(val, addr);
break;
case 2:
case 3:
segv = put_user_u32(val, addr);
break;
}
if (segv) {
env->cp15.c6_data = addr;
goto done;
}
if (size == 3) {
val = env->regs[(env->exclusive_info >> 12) & 0xf];
segv = put_user_u32(val, addr);
if (segv) {
env->cp15.c6_data = addr + 4;
goto done;
}
}
rc = 0;
fail:
env->regs[(env->exclusive_info >> 4) & 0xf] = rc;
done:
end_exclusive();
return segv;
}
void cpu_loop(CPUARMState *env)
{
int trapnr;
@ -717,6 +792,11 @@ void cpu_loop(CPUARMState *env)
if (do_kernel_trap(env))
goto error;
break;
case EXCP_STREX:
if (do_strex(env)) {
addr = env->cp15.c6_data;
goto do_segv;
}
default:
error:
fprintf(stderr, "qemu: unhandled CPU exception 0x%x - aborting\n",

View File

@ -40,6 +40,7 @@
#define EXCP_BKPT 7
#define EXCP_EXCEPTION_EXIT 8 /* Return from v7M exception. */
#define EXCP_KERNEL_TRAP 9 /* Jumped to kernel code page. */
#define EXCP_STREX 10
#define ARMV7M_EXCP_RESET 1
#define ARMV7M_EXCP_NMI 2
@ -180,10 +181,12 @@ typedef struct CPUARMState {
float_status fp_status;
} vfp;
uint32_t exclusive_addr;
uint32_t exclusive_val;
uint32_t exclusive_high;
#if defined(CONFIG_USER_ONLY)
struct mmon_state *mmon_entry;
#else
uint32_t mmon_addr;
uint32_t exclusive_test;
uint32_t exclusive_info;
#endif
/* iwMMXt coprocessor state. */

View File

@ -470,16 +470,6 @@ void do_interrupt (CPUState *env)
env->exception_index = -1;
}
/* Structure used to record exclusive memory locations. */
typedef struct mmon_state {
struct mmon_state *next;
CPUARMState *cpu_env;
uint32_t addr;
} mmon_state;
/* Chain of current locks. */
static mmon_state* mmon_head = NULL;
int cpu_arm_handle_mmu_fault (CPUState *env, target_ulong address, int rw,
int mmu_idx, int is_softmmu)
{
@ -493,62 +483,6 @@ int cpu_arm_handle_mmu_fault (CPUState *env, target_ulong address, int rw,
return 1;
}
static void allocate_mmon_state(CPUState *env)
{
env->mmon_entry = malloc(sizeof (mmon_state));
memset (env->mmon_entry, 0, sizeof (mmon_state));
env->mmon_entry->cpu_env = env;
mmon_head = env->mmon_entry;
}
/* Flush any monitor locks for the specified address. */
static void flush_mmon(uint32_t addr)
{
mmon_state *mon;
for (mon = mmon_head; mon; mon = mon->next)
{
if (mon->addr != addr)
continue;
mon->addr = 0;
break;
}
}
/* Mark an address for exclusive access. */
void HELPER(mark_exclusive)(CPUState *env, uint32_t addr)
{
if (!env->mmon_entry)
allocate_mmon_state(env);
/* Clear any previous locks. */
flush_mmon(addr);
env->mmon_entry->addr = addr;
}
/* Test if an exclusive address is still exclusive. Returns zero
if the address is still exclusive. */
uint32_t HELPER(test_exclusive)(CPUState *env, uint32_t addr)
{
int res;
if (!env->mmon_entry)
return 1;
if (env->mmon_entry->addr == addr)
res = 0;
else
res = 1;
flush_mmon(addr);
return res;
}
void HELPER(clrex)(CPUState *env)
{
if (!(env->mmon_entry && env->mmon_entry->addr))
return;
flush_mmon(env->mmon_entry->addr);
}
target_phys_addr_t cpu_get_phys_page_debug(CPUState *env, target_ulong addr)
{
return addr;
@ -1273,24 +1207,6 @@ target_phys_addr_t cpu_get_phys_page_debug(CPUState *env, target_ulong addr)
return phys_addr;
}
/* Not really implemented. Need to figure out a sane way of doing this.
Maybe add generic watchpoint support and use that. */
void HELPER(mark_exclusive)(CPUState *env, uint32_t addr)
{
env->mmon_addr = addr;
}
uint32_t HELPER(test_exclusive)(CPUState *env, uint32_t addr)
{
return (env->mmon_addr != addr);
}
void HELPER(clrex)(CPUState *env)
{
env->mmon_addr = -1;
}
void HELPER(set_cp)(CPUState *env, uint32_t insn, uint32_t val)
{
int cp_num = (insn >> 8) & 0xf;

View File

@ -68,10 +68,6 @@ DEF_HELPER_2(get_cp, i32, env, i32)
DEF_HELPER_2(get_r13_banked, i32, env, i32)
DEF_HELPER_3(set_r13_banked, void, env, i32, i32)
DEF_HELPER_2(mark_exclusive, void, env, i32)
DEF_HELPER_2(test_exclusive, i32, env, i32)
DEF_HELPER_1(clrex, void, env)
DEF_HELPER_1(get_user_reg, i32, i32)
DEF_HELPER_2(set_user_reg, void, i32, i32)

View File

@ -76,6 +76,13 @@ static TCGv_ptr cpu_env;
/* We reuse the same 64-bit temporaries for efficiency. */
static TCGv_i64 cpu_V0, cpu_V1, cpu_M0;
static TCGv_i32 cpu_R[16];
static TCGv_i32 cpu_exclusive_addr;
static TCGv_i32 cpu_exclusive_val;
static TCGv_i32 cpu_exclusive_high;
#ifdef CONFIG_USER_ONLY
static TCGv_i32 cpu_exclusive_test;
static TCGv_i32 cpu_exclusive_info;
#endif
/* FIXME: These should be removed. */
static TCGv cpu_F0s, cpu_F1s;
@ -99,6 +106,18 @@ void arm_translate_init(void)
offsetof(CPUState, regs[i]),
regnames[i]);
}
cpu_exclusive_addr = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUState, exclusive_addr), "exclusive_addr");
cpu_exclusive_val = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUState, exclusive_val), "exclusive_val");
cpu_exclusive_high = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUState, exclusive_high), "exclusive_high");
#ifdef CONFIG_USER_ONLY
cpu_exclusive_test = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUState, exclusive_test), "exclusive_test");
cpu_exclusive_info = tcg_global_mem_new_i32(TCG_AREG0,
offsetof(CPUState, exclusive_info), "exclusive_info");
#endif
#define GEN_HELPER 2
#include "helpers.h"
@ -5819,6 +5838,132 @@ static void gen_logicq_cc(TCGv_i64 val)
dead_tmp(tmp);
}
/* Load/Store exclusive instructions are implemented by remembering
the value/address loaded, and seeing if these are the same
when the store is performed. This should be is sufficient to implement
the architecturally mandated semantics, and avoids having to monitor
regular stores.
In system emulation mode only one CPU will be running at once, so
this sequence is effectively atomic. In user emulation mode we
throw an exception and handle the atomic operation elsewhere. */
static void gen_load_exclusive(DisasContext *s, int rt, int rt2,
TCGv addr, int size)
{
TCGv tmp;
switch (size) {
case 0:
tmp = gen_ld8u(addr, IS_USER(s));
break;
case 1:
tmp = gen_ld16u(addr, IS_USER(s));
break;
case 2:
case 3:
tmp = gen_ld32(addr, IS_USER(s));
break;
default:
abort();
}
tcg_gen_mov_i32(cpu_exclusive_val, tmp);
store_reg(s, rt, tmp);
if (size == 3) {
tcg_gen_addi_i32(addr, addr, 4);
tmp = gen_ld32(addr, IS_USER(s));
tcg_gen_mov_i32(cpu_exclusive_high, tmp);
store_reg(s, rt2, tmp);
}
tcg_gen_mov_i32(cpu_exclusive_addr, addr);
}
static void gen_clrex(DisasContext *s)
{
tcg_gen_movi_i32(cpu_exclusive_addr, -1);
}
#ifdef CONFIG_USER_ONLY
static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
TCGv addr, int size)
{
tcg_gen_mov_i32(cpu_exclusive_test, addr);
tcg_gen_movi_i32(cpu_exclusive_info,
size | (rd << 4) | (rt << 8) | (rt2 << 12));
gen_set_condexec(s);
gen_set_pc_im(s->pc - 4);
gen_exception(EXCP_STREX);
s->is_jmp = DISAS_JUMP;
}
#else
static void gen_store_exclusive(DisasContext *s, int rd, int rt, int rt2,
TCGv addr, int size)
{
TCGv tmp;
int done_label;
int fail_label;
/* if (env->exclusive_addr == addr && env->exclusive_val == [addr]) {
[addr] = {Rt};
{Rd} = 0;
} else {
{Rd} = 1;
} */
fail_label = gen_new_label();
done_label = gen_new_label();
tcg_gen_brcond_i32(TCG_COND_NE, addr, cpu_exclusive_addr, fail_label);
switch (size) {
case 0:
tmp = gen_ld8u(addr, IS_USER(s));
break;
case 1:
tmp = gen_ld16u(addr, IS_USER(s));
break;
case 2:
case 3:
tmp = gen_ld32(addr, IS_USER(s));
break;
default:
abort();
}
tcg_gen_brcond_i32(TCG_COND_NE, tmp, cpu_exclusive_val, fail_label);
dead_tmp(tmp);
if (size == 3) {
TCGv tmp2 = new_tmp();
tcg_gen_addi_i32(tmp2, addr, 4);
tmp = gen_ld32(addr, IS_USER(s));
dead_tmp(tmp2);
tcg_gen_brcond_i32(TCG_COND_NE, tmp, cpu_exclusive_high, fail_label);
dead_tmp(tmp);
}
tmp = load_reg(s, rt);
switch (size) {
case 0:
gen_st8(tmp, addr, IS_USER(s));
break;
case 1:
gen_st16(tmp, addr, IS_USER(s));
break;
case 2:
case 3:
gen_st32(tmp, addr, IS_USER(s));
break;
default:
abort();
}
if (size == 3) {
tcg_gen_addi_i32(addr, addr, 4);
tmp = load_reg(s, rt2);
gen_st32(tmp, addr, IS_USER(s));
}
tcg_gen_movi_i32(cpu_R[rd], 0);
tcg_gen_br(done_label);
gen_set_label(fail_label);
tcg_gen_movi_i32(cpu_R[rd], 1);
gen_set_label(done_label);
tcg_gen_movi_i32(cpu_exclusive_addr, -1);
}
#endif
static void disas_arm_insn(CPUState * env, DisasContext *s)
{
unsigned int cond, insn, val, op1, i, shift, rm, rs, rn, rd, sh;
@ -5869,7 +6014,7 @@ static void disas_arm_insn(CPUState * env, DisasContext *s)
switch ((insn >> 4) & 0xf) {
case 1: /* clrex */
ARCH(6K);
gen_helper_clrex(cpu_env);
gen_clrex(s);
return;
case 4: /* dsb */
case 5: /* dmb */
@ -6454,57 +6599,40 @@ static void disas_arm_insn(CPUState * env, DisasContext *s)
addr = tcg_temp_local_new_i32();
load_reg_var(s, addr, rn);
if (insn & (1 << 20)) {
gen_helper_mark_exclusive(cpu_env, addr);
switch (op1) {
case 0: /* ldrex */
tmp = gen_ld32(addr, IS_USER(s));
gen_load_exclusive(s, rd, 15, addr, 2);
break;
case 1: /* ldrexd */
tmp = gen_ld32(addr, IS_USER(s));
store_reg(s, rd, tmp);
tcg_gen_addi_i32(addr, addr, 4);
tmp = gen_ld32(addr, IS_USER(s));
rd++;
gen_load_exclusive(s, rd, rd + 1, addr, 3);
break;
case 2: /* ldrexb */
tmp = gen_ld8u(addr, IS_USER(s));
gen_load_exclusive(s, rd, 15, addr, 0);
break;
case 3: /* ldrexh */
tmp = gen_ld16u(addr, IS_USER(s));
gen_load_exclusive(s, rd, 15, addr, 1);
break;
default:
abort();
}
store_reg(s, rd, tmp);
} else {
int label = gen_new_label();
rm = insn & 0xf;
tmp2 = tcg_temp_local_new_i32();
gen_helper_test_exclusive(tmp2, cpu_env, addr);
tcg_gen_brcondi_i32(TCG_COND_NE, tmp2, 0, label);
tmp = load_reg(s,rm);
switch (op1) {
case 0: /* strex */
gen_st32(tmp, addr, IS_USER(s));
gen_store_exclusive(s, rd, rm, 15, addr, 2);
break;
case 1: /* strexd */
gen_st32(tmp, addr, IS_USER(s));
tcg_gen_addi_i32(addr, addr, 4);
tmp = load_reg(s, rm + 1);
gen_st32(tmp, addr, IS_USER(s));
gen_store_exclusive(s, rd, rm, rm + 1, addr, 2);
break;
case 2: /* strexb */
gen_st8(tmp, addr, IS_USER(s));
gen_store_exclusive(s, rd, rm, 15, addr, 0);
break;
case 3: /* strexh */
gen_st16(tmp, addr, IS_USER(s));
gen_store_exclusive(s, rd, rm, 15, addr, 1);
break;
default:
abort();
}
gen_set_label(label);
tcg_gen_mov_i32(cpu_R[rd], tmp2);
tcg_temp_free(tmp2);
}
tcg_temp_free(addr);
} else {
@ -7259,20 +7387,11 @@ static int disas_thumb2_insn(CPUState *env, DisasContext *s, uint16_t insn_hw1)
/* Load/store exclusive word. */
addr = tcg_temp_local_new();
load_reg_var(s, addr, rn);
tcg_gen_addi_i32(addr, addr, (insn & 0xff) << 2);
if (insn & (1 << 20)) {
gen_helper_mark_exclusive(cpu_env, addr);
tmp = gen_ld32(addr, IS_USER(s));
store_reg(s, rd, tmp);
gen_load_exclusive(s, rs, 15, addr, 2);
} else {
int label = gen_new_label();
tmp2 = tcg_temp_local_new();
gen_helper_test_exclusive(tmp2, cpu_env, addr);
tcg_gen_brcondi_i32(TCG_COND_NE, tmp2, 0, label);
tmp = load_reg(s, rs);
gen_st32(tmp, addr, IS_USER(s));
gen_set_label(label);
tcg_gen_mov_i32(cpu_R[rd], tmp2);
tcg_temp_free(tmp2);
gen_store_exclusive(s, rd, rs, 15, addr, 2);
}
tcg_temp_free(addr);
} else if ((insn & (1 << 6)) == 0) {
@ -7300,56 +7419,17 @@ static int disas_thumb2_insn(CPUState *env, DisasContext *s, uint16_t insn_hw1)
store_reg(s, 15, tmp);
} else {
/* Load/store exclusive byte/halfword/doubleword. */
/* ??? These are not really atomic. However we know
we never have multiple CPUs running in parallel,
so it is good enough. */
ARCH(7);
op = (insn >> 4) & 0x3;
if (op == 2) {
goto illegal_op;
}
addr = tcg_temp_local_new();
load_reg_var(s, addr, rn);
if (insn & (1 << 20)) {
gen_helper_mark_exclusive(cpu_env, addr);
switch (op) {
case 0:
tmp = gen_ld8u(addr, IS_USER(s));
break;
case 1:
tmp = gen_ld16u(addr, IS_USER(s));
break;
case 3:
tmp = gen_ld32(addr, IS_USER(s));
tcg_gen_addi_i32(addr, addr, 4);
tmp2 = gen_ld32(addr, IS_USER(s));
store_reg(s, rd, tmp2);
break;
default:
goto illegal_op;
}
store_reg(s, rs, tmp);
gen_load_exclusive(s, rs, rd, addr, op);
} else {
int label = gen_new_label();
tmp2 = tcg_temp_local_new();
gen_helper_test_exclusive(tmp2, cpu_env, addr);
tcg_gen_brcondi_i32(TCG_COND_NE, tmp2, 0, label);
tmp = load_reg(s, rs);
switch (op) {
case 0:
gen_st8(tmp, addr, IS_USER(s));
break;
case 1:
gen_st16(tmp, addr, IS_USER(s));
break;
case 3:
gen_st32(tmp, addr, IS_USER(s));
tcg_gen_addi_i32(addr, addr, 4);
tmp = load_reg(s, rd);
gen_st32(tmp, addr, IS_USER(s));
break;
default:
goto illegal_op;
}
gen_set_label(label);
tcg_gen_mov_i32(cpu_R[rm], tmp2);
tcg_temp_free(tmp2);
gen_store_exclusive(s, rm, rs, rd, addr, op);
}
tcg_temp_free(addr);
}
@ -7845,16 +7925,16 @@ static int disas_thumb2_insn(CPUState *env, DisasContext *s, uint16_t insn_hw1)
}
break;
case 3: /* Special control operations. */
ARCH(7);
op = (insn >> 4) & 0xf;
switch (op) {
case 2: /* clrex */
gen_helper_clrex(cpu_env);
gen_clrex(s);
break;
case 4: /* dsb */
case 5: /* dmb */
case 6: /* isb */
/* These execute as NOPs. */
ARCH(7);
break;
default:
goto illegal_op;