Skip to contentSkip to conversation

accel/tcg: Fix undefined jump with x86_cpu_tlb_fill

PATCH In review Daniel Hoffman 2 participants 2 messages Dec 1, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 accel/tcg: Fix undefined jump with x86_cpu_tlb_fill

---
 target/i386/tcg/sysemu/excp_helper.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
target/i386/tcg/sysemu/excp_helper.c +2 -1
index 405a5d414a..6a93e96eb7 100644
@@ -595,6 +595,7 @@ bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size,
     CPUX86State *env = cs->env_ptr;
     TranslateResult out;
     TranslateFault err;
+    bool use_stage2 = env->hflags2 & HF2_NPT_MASK;
 
     if (get_physical_address(env, addr, access_type, mmu_idx, &out, &err)) {
         /*
@@ -615,7 +616,7 @@ bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size,
         return false;
     }
 
-    if (err.stage2 != S2_NONE) {
+    if (use_stage2 && err.stage2 != S2_NONE) {
         raise_stage2(env, &err, retaddr);
     }
 
Signature · 1 line
2.34.1
#1 lore raw
Quoted 18 lines · Daniel Hoffman
---
  target/i386/tcg/sysemu/excp_helper.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
target/i386/tcg/sysemu/excp_helper.c +1 -0
index 405a5d414a..6a93e96eb7 100644
@@ -595,6 +595,7 @@ bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size,
      CPUX86State *env = cs->env_ptr;
      TranslateResult out;
      TranslateFault err;
+    bool use_stage2 = env->hflags2 & HF2_NPT_MASK;
 
      if (get_physical_address(env, addr, access_type, mmu_idx, &out, &err)) {
          /*

target/i386/tcg/sysemu/excp_helper.c bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size, · line 622

Quoted 8 lines
target/i386/tcg/sysemu/excp_helper.c +1 -1
@@ -615,7 +616,7 @@ bool x86_cpu_tlb_fill(CPUState *cs, vaddr addr, int size,
          return false;
      }
 
-    if (err.stage2 != S2_NONE) {
+    if (use_stage2 && err.stage2 != S2_NONE) {
          raise_stage2(env, &err, retaddr);
      }
 

The patch description is missing, which might have included some indication of how the problem might have been triggered.

From inspection I can guess that it comes from the incomplete initialization of TranslateFault in get_physical_address, after the check for a canonical address form. In any case this is not an ideal fix.

r~