Fix race condition on access to env->interrupt_request
RFC PATCH 1/1 Fix race condition on access to env->interrupt_request
env->interrupt_request is accessed as the bit level from both main code and signal handler, making a race condition possible even on CISC CPU. This causes freeze of QEMU under high load when running the dyntick clock.
The patch below move the bit corresponding to CPU_INTERRUPT_EXIT in a separate variable, declared as volatile sig_atomic_t, so it should be work even on RISC CPU.
We may want to move the cpu_interrupt(env, CPU_INTERRUPT_EXIT) case in its own function and get rid of CPU_INTERRUPT_EXIT. That can be done later, I wanted to keep the patch short for easier review.
cpu-defs.h +2 -0
@@ -27,6 +27,7 @@#include "config.h" #include <setjmp.h> #include <inttypes.h> +#include <signal.h> #include "osdep.h" #include "sys-queue.h"
@@ -170,6 +171,7 @@ typedef struct CPUWatchpoint {memory was accessed */ \uint32_t halted;/* Nonzero if the CPU is in suspend state */ \uint32_t interrupt_request; \ +volatile sig_atomic_t exit_request; \/* The meaning of the MMU modes is defined in the target code. */ \CPUTLBEntry tlb_table[NB_MMU_MODES ][CPU_TLB_SIZE ]; \target_phys_addr_t iotlb[NB_MMU_MODES ][CPU_TLB_SIZE ]; \
cpu-exec.c +8 -8
@@ -311,7 +311,7 @@ int cpu_exec(CPUState *env1) env->exception_index =-1 ; }#ifdef USE_KQEMU -if (kqemu_is_ok (env) && env->interrupt_request ==0 ) { +if (kqemu_is_ok (env) && env->interrupt_request ==0 && env->exit_request ==0 ) {int ret; env->eflags = env->eflags |helper_cc_compute_all (CC_OP ) | (DF &DF_MASK ); ret =kqemu_cpu_exec (env);
@@ -326,7 +326,7 @@ int cpu_exec(CPUState *env1) }else if (ret ==2 ) {/* softmmu execution needed */ }else { -if (env->interrupt_request !=0 ) { +if (env->interrupt_request !=0 || env->exit_request !=0 ) {/* hardware interrupt will be executed just after */ }else {/* otherwise, we restart */
@@ -525,11 +525,11 @@ int cpu_exec(CPUState *env1)the program flow was changed */ next_tb =0 ; } -if (interrupt_request &CPU_INTERRUPT_EXIT ) { - env->interrupt_request &= ~CPU_INTERRUPT_EXIT ; - env->exception_index =EXCP_INTERRUPT ; -cpu_loop_exit (); - } + } +if (unlikely (env->exit_request)) { + env->exit_request =0 ; + env->exception_index =EXCP_INTERRUPT ; +cpu_loop_exit (); }#ifdef DEBUG_EXEC if (qemu_loglevel_mask (CPU_LOG_TB_CPU )) {
@@ -599,7 +599,7 @@ int cpu_exec(CPUState *env1)TB, but before it is linked into a potentially infinite loop and becomes env->current_tb. Avoid starting execution if there is a pending interrupt. */ -if (unlikely (env->interrupt_request &CPU_INTERRUPT_EXIT )) +if (unlikely (env->exit_request)) env->current_tb =NULL ;while (env->current_tb) {
exec.c +6 -5
@@ -1501,9 +1501,12 @@ void cpu_interrupt(CPUState *env, int mask)#endif int old_mask; +if (mask &CPU_INTERRUPT_EXIT ) { + env->exit_request =1 ; + mask &= ~CPU_INTERRUPT_EXIT ; + } + old_mask = env->interrupt_request; -/* FIXME: This is probably not threadsafe. A different thread could -be in the middle of a read-modify-write operation. */ env->interrupt_request |= mask;#if defined(USE_NPTL )/* FIXME: TB unchaining isn't SMP safe. For now just ignore the
@@ -1514,10 +1517,8 @@ void cpu_interrupt(CPUState *env, int mask)if (use_icount) { env->icount_decr.u16.high =0xffff ;#ifndef CONFIG_USER_ONLY -/* CPU_INTERRUPT_EXIT isn't a real interrupt. It just means -an async event happened and we need to process it. */ if (!can_do_io (env) - && (mask & ~(old_mask |CPU_INTERRUPT_EXIT )) !=0 ) { + && (mask & ~old_mask) !=0 ) {cpu_abort (env,"Raised interrupt while not in I/O function" ); }#endif
kvm-all.c +3 -3
@@ -445,7 +445,7 @@ int kvm_cpu_exec(CPUState *env) do { kvm_arch_pre_run(env, run); - if ((env->interrupt_request & CPU_INTERRUPT_EXIT)) { + if (env->exit_request) { dprintf("interrupt exit requested\n"); ret = 0; break;
@@ -512,8 +512,8 @@ int kvm_cpu_exec(CPUState *env) } }while (ret >0 ); -if ((env->interrupt_request &CPU_INTERRUPT_EXIT )) { - env->interrupt_request &= ~CPU_INTERRUPT_EXIT ; +if (env->exit_request) { + env->exit_request =0 ; env->exception_index =EXCP_INTERRUPT ; }
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
Quoted 9 lines · Aurelien Jarno
env->interrupt_requestis accessed as the bit level from both main code and signal handler, making a race condition possible even on CISC CPU. This causes freeze of QEMU under high load when running the dyntick clock.The patch below move the bit corresponding to
CPU_INTERRUPT_EXITin a separate variable, declared as volatilesig_atomic_t, so it should be work even on RISC CPU.
Quoted 6 lines
We may want to move the
cpu_interrupt(env, CPU_INTERRUPT_EXIT)case in its own function and get rid ofCPU_INTERRUPT_EXIT. That can be done later, I wanted to keep the patch short for easier review.Signed-off-by Aurelien Jarno <[email protected]>
Looks good.
Signature · 1 line
mailto:[email protected]