Skip to contentSkip to conversation

softmmu: fixing usage of cpu_st/ld* from helpers

PATCH In review Pavel Dovgalyuk 3 participants 3 messages Sep 15, 2014

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 softmmu: fixing usage of cpu_st/ld* from helpers

MMU helper functions are called from generated code and other helper functions. In both cases they try to get function's return address for using it while restoring virtual CPU state.

When MMU helper is called from some other helper function (like helper_maskmov_xmm) through cpu_st* function, the return address will point to that helper. That is why CPU state cannot be restored in the case of MMU fault.

This bug can occur when maskmov instruction is located in the middle of the translation block.

Execution sequence for this example:

TB start:
PC1: instr1
instr2
PC2: maskmov <page fault>
<page fault processing>
PC1: instr1
instr2
maskmov

At the start of TB execution guest PC points to instr1. When page fault occurs QEMU tries to restore guest PC (which should be equal to PC2). It reads host PC from the call stack and checks whether it points to TB or not. Bug in ldst helpers implementation provides incorrect host PC, which is not located within the TB. That's why QEMU cannot recover guest PC and it remains the same (PC1). After page fault processing QEMU restarts TB and executes instr1 and instr2 for the second time, because guest PC was not recovered.

This patch introduces several inline helpers to load return address which points to the right place. Correct return address allows correct restoring of the guest PC.

---
 include/exec/cpu_ldst_template.h |   31 +++++++++++++++++++++++++++----
 include/exec/exec-all.h          |   27 +++++++++++++++++++++++++++
 softmmu_template.h               |   18 ++++++++++++++++++
 3 files changed, 72 insertions(+), 4 deletions(-)
include/exec/cpu_ldst_template.h +27 -4
index 006093a..2658955 100644
@@ -61,6 +61,17 @@
 #define MMUSUFFIX _mmu
 #endif
 
+/* inline helper ld function */
+
+static inline DATA_TYPE
+glue(glue(helper_inline_ld, SUFFIX), MEMSUFFIX)(CPUArchState *env,
+                                                target_ulong addr,
+                                                int mmu_idx)
+{
+    return glue(glue(helper_call_ld, SUFFIX), MMUSUFFIX)(env, addr, mmu_idx,
+                                                         GETRA());
+}
+
 /* generic load/store macros */
 
 static inline RES_TYPE
@@ -76,7 +87,8 @@ glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr)
     mmu_idx = CPU_MMU_INDEX;
     if (unlikely(env->tlb_table[mmu_idx][page_index].ADDR_READ !=
                  (addr & (TARGET_PAGE_MASK | (DATA_SIZE - 1))))) {
-        res = glue(glue(helper_ld, SUFFIX), MMUSUFFIX)(env, addr, mmu_idx);
+        res = glue(glue(helper_inline_ld, SUFFIX), MEMSUFFIX)(env, addr,
+                                                              mmu_idx);
     } else {
         uintptr_t hostaddr = addr + env->tlb_table[mmu_idx][page_index].addend;
         res = glue(glue(ld, USUFFIX), _raw)(hostaddr);
@@ -97,8 +109,8 @@ glue(glue(cpu_lds, SUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr)
     mmu_idx = CPU_MMU_INDEX;
     if (unlikely(env->tlb_table[mmu_idx][page_index].ADDR_READ !=
                  (addr & (TARGET_PAGE_MASK | (DATA_SIZE - 1))))) {
-        res = (DATA_STYPE)glue(glue(helper_ld, SUFFIX),
-                               MMUSUFFIX)(env, addr, mmu_idx);
+        res = (DATA_STYPE)glue(glue(helper_inline_ld, SUFFIX),
+                               MEMSUFFIX)(env, addr, mmu_idx);
     } else {
         uintptr_t hostaddr = addr + env->tlb_table[mmu_idx][page_index].addend;
         res = glue(glue(lds, SUFFIX), _raw)(hostaddr);
@@ -109,6 +121,17 @@ glue(glue(cpu_lds, SUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr)
 
 #ifndef SOFTMMU_CODE_ACCESS
 
+/* inline helper st function */
+
+static inline void
+glue(glue(helper_inline_st, SUFFIX), MEMSUFFIX)(CPUArchState *env,
+                                                target_ulong addr,
+                                                DATA_TYPE val, int mmu_idx)
+{
+    glue(glue(helper_call_st, SUFFIX), MMUSUFFIX)(env, addr, val, mmu_idx,
+                                                  GETRA());
+}
+
 /* generic store macro */
 
 static inline void
@@ -124,7 +147,7 @@ glue(glue(cpu_st, SUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr,
     mmu_idx = CPU_MMU_INDEX;
     if (unlikely(env->tlb_table[mmu_idx][page_index].addr_write !=
                  (addr & (TARGET_PAGE_MASK | (DATA_SIZE - 1))))) {
-        glue(glue(helper_st, SUFFIX), MMUSUFFIX)(env, addr, v, mmu_idx);
+        glue(glue(helper_inline_st, SUFFIX), MEMSUFFIX)(env, addr, v, mmu_idx);
     } else {
         uintptr_t hostaddr = addr + env->tlb_table[mmu_idx][page_index].addend;
         glue(glue(st, SUFFIX), _raw)(hostaddr, v);
include/exec/exec-all.h +27 -0
index 5e5d86e..528928f 100644
@@ -344,6 +344,33 @@ bool io_mem_write(struct MemoryRegion *mr, hwaddr addr,
 void tlb_fill(CPUState *cpu, target_ulong addr, int is_write, int mmu_idx,
               uintptr_t retaddr);
 
+uint8_t helper_call_ldb_cmmu(CPUArchState *env, target_ulong addr,
+                             int mmu_idx, uintptr_t retaddr);
+uint16_t helper_call_ldw_cmmu(CPUArchState *env, target_ulong addr,
+                              int mmu_idx, uintptr_t retaddr);
+uint32_t helper_call_ldl_cmmu(CPUArchState *env, target_ulong addr,
+                              int mmu_idx, uintptr_t retaddr);
+uint64_t helper_call_ldq_cmmu(CPUArchState *env, target_ulong addr,
+                              int mmu_idx, uintptr_t retaddr);
+
+uint8_t helper_call_ldb_mmu(CPUArchState *env, target_ulong addr,
+                            int mmu_idx, uintptr_t retaddr);
+uint16_t helper_call_ldw_mmu(CPUArchState *env, target_ulong addr,
+                             int mmu_idx, uintptr_t retaddr);
+uint32_t helper_call_ldl_mmu(CPUArchState *env, target_ulong addr,
+                             int mmu_idx, uintptr_t retaddr);
+uint64_t helper_call_ldq_mmu(CPUArchState *env, target_ulong addr,
+                             int mmu_idx, uintptr_t retaddr);
+
+void helper_call_stb_mmu(CPUArchState *env, target_ulong addr,
+                         uint8_t val, int mmu_idx, uintptr_t retaddr);
+void helper_call_stw_mmu(CPUArchState *env, target_ulong addr,
+                         uint16_t val, int mmu_idx, uintptr_t retaddr);
+void helper_call_stl_mmu(CPUArchState *env, target_ulong addr,
+                         uint32_t val, int mmu_idx, uintptr_t retaddr);
+void helper_call_stq_mmu(CPUArchState *env, target_ulong addr,
+                         uint64_t val, int mmu_idx, uintptr_t retaddr);
+
 #endif
 
 #if defined(CONFIG_USER_ONLY)
softmmu_template.h +18 -0
index 5a07f99..1053cf3 100644
@@ -311,6 +311,15 @@ glue(glue(helper_ld, SUFFIX), MMUSUFFIX)(CPUArchState *env, target_ulong addr,
     return helper_te_ld_name (env, addr, mmu_idx, GETRA());
 }
 
+DATA_TYPE
+glue(glue(helper_call_ld, SUFFIX), MMUSUFFIX)(CPUArchState *env,
+                                              target_ulong addr,
+                                              int mmu_idx,
+                                              uintptr_t retaddr)
+{
+    return helper_te_ld_name(env, addr, mmu_idx, retaddr);
+}
+
 #ifndef SOFTMMU_CODE_ACCESS
 
 /* Provide signed versions of the load routines as well.  We can of course
@@ -505,6 +514,15 @@ glue(glue(helper_st, SUFFIX), MMUSUFFIX)(CPUArchState *env, target_ulong addr,
     helper_te_st_name(env, addr, val, mmu_idx, GETRA());
 }
 
+void
+glue(glue(helper_call_st, SUFFIX), MMUSUFFIX)(CPUArchState *env,
+                                              target_ulong addr,
+                                              DATA_TYPE val, int mmu_idx,
+                                              uintptr_t retaddr)
+{
+    helper_te_st_name(env, addr, val, mmu_idx, retaddr);
+}
+
 #endif /* !defined(SOFTMMU_CODE_ACCESS) */
 
 #undef READ_ACCESS_TYPE
#1 lore raw

CCing RTH who was the last person to do something with this area of the code I think. I thought the correct answer to this problem was "don't use the cpu_st* functions but use something else you can pass GETRA() to" ?

Quoted 31 lines · Pavel Dovgalyuk

MMU helper functions are called from generated code and other helper functions. In both cases they try to get function's return address for using it while restoring virtual CPU state.

When MMU helper is called from some other helper function (like helper_maskmov_xmm) through cpu_st* function, the return address will point to that helper. That is why CPU state cannot be restored in the case of MMU fault.

This bug can occur when maskmov instruction is located in the middle of the translation block.

Execution sequence for this example:

TB start:
PC1: instr1
instr2
PC2: maskmov <page fault>
<page fault processing>
PC1: instr1
instr2
maskmov

At the start of TB execution guest PC points to instr1. When page fault occurs QEMU tries to restore guest PC (which should be equal to PC2). It reads host PC from the call stack and checks whether it points to TB or not. Bug in ldst helpers implementation provides incorrect host PC, which is not located within the TB. That's why QEMU cannot recover guest PC and it remains the same (PC1). After page fault processing QEMU restarts TB and executes instr1 and instr2 for the second time, because guest PC was not recovered.

Quoted 3 lines

This patch introduces several inline helpers to load return address which points to the right place. Correct return address allows correct restoring of the guest PC.

thanks
-- PMM

#2 lore raw
In reply to Pavel Dovgalyuk

include/exec/cpu_ldst_template.h line 73

Quoted 10 lines · Pavel Dovgalyuk
include/exec/cpu_ldst_template.h +10 -0
+/* inline helper ld function */
+
+static inline DATA_TYPE
+glue(glue(helper_inline_ld, SUFFIX), MEMSUFFIX)(CPUArchState *env,
+                                                target_ulong addr,
+                                                int mmu_idx)
+{
+    return glue(glue(helper_call_ld, SUFFIX), MMUSUFFIX)(env, addr, mmu_idx,
+                                                         GETRA());
+}

You'd have to mark this always_inline to make absolutely sure that the caller's GETRA value is used. That said...

include/exec/cpu_ldst_template.h glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr) · line 94

Quoted 10 lines
include/exec/cpu_ldst_template.h +2 -1
@@ -76,7 +87,8 @@ glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(CPUArchState *env, target_ulong ptr)
     mmu_idx = CPU_MMU_INDEX;
     if (unlikely(env->tlb_table[mmu_idx][page_index].ADDR_READ !=
                  (addr & (TARGET_PAGE_MASK | (DATA_SIZE - 1))))) {
-        res = glue(glue(helper_ld, SUFFIX), MMUSUFFIX)(env, addr, mmu_idx);
+        res = glue(glue(helper_inline_ld, SUFFIX), MEMSUFFIX)(env, addr,
+                                                              mmu_idx);
     } else {
         uintptr_t hostaddr = addr + env->tlb_table[mmu_idx][page_index].addend;
         res = glue(glue(ld, USUFFIX), _raw)(hostaddr);

... this is also the wrong context.

The only GETRA value that helps you at all is the one from the *top level* helper -- the one that's directly called from TCG code. So, in the case of maskmov, helper_maskmov_xmm. Anything else and you aren't getting the call site address from the TCG code, and so can't be used to detect the PC of the MMU fault.

I guess there are only two real possibilities:

(1) Have the cpu_ldst_template helpers all be marked always_inline so that they could use GETRA. I'm not too fond of this because we'd still get the wrong results if these are not used from top-level helpers.

(2) Add helpers that accept the GETRA value from the top-level helper. And not hidden within a macro or always_inline function. This helps us see what portions of the code have been audited for the new interface. This will involve quite a bit more code churn, but shouldn't been too difficult for any single function.

r~