rpms/valgrind
7
0
Fork 0
Code Issues Pull Requests Releases Activity
f1bdbbead9
valgrind/valgrind-3.18.1-arm64-atomic-align.patch
Mark Wielaard 54ca779d10 Add valgrind-3.18.1-arm64-atomic-align.patch
2021-11-17 22:54:13 +01:00

164 lines
6.7 KiB
Diff
Raw Blame History

commit 2be719921e700a9ac9b85f470ed87cb8adf8151b
Author: Julian Seward <jseward@acm.org>
Date: Sat Nov 13 09:27:01 2021 +0100
Bug 445415 - arm64 front end: alignment checks missing for atomic instructions.
For the arm64 front end, none of the atomic instructions have address
alignment checks included in their IR. They all should. The effect of
missing alignment checks in the IR is that, since this IR will in most cases
be translated back to atomic instructions in the back end, we will get
alignment traps (SIGBUS) on the host side and not on the guest side, which is
(very) incorrect behaviour of the simulation.
diff --git a/VEX/priv/guest_arm64_toIR.c b/VEX/priv/guest_arm64_toIR.c
index ee018c6a9..16a7e075f 100644
--- a/VEX/priv/guest_arm64_toIR.c
+++ b/VEX/priv/guest_arm64_toIR.c
@@ -4833,6 +4833,34 @@ static IRTemp gen_zwidening_load ( UInt szB, IRTemp addr )
}
+/* Generate a SIGBUS followed by a restart of the current instruction if
+ `effective_addr` is `align`-aligned. This is required behaviour for atomic
+ instructions. This assumes that guest_RIP_curr_instr is set correctly!
+
+ This is hardwired to generate SIGBUS because so far the only supported arm64
+ (arm64-linux) does that. Should we need to later extend it to generate some
+ other signal, use the same scheme as with gen_SIGNAL_if_not_XX_aligned in
+ guest_amd64_toIR.c. */
+static
+void gen_SIGBUS_if_not_XX_aligned ( IRTemp effective_addr, ULong align )
+{
+ if (align == 1) {
+ return;
+ }
+ vassert(align == 16 || align == 8 || align == 4 || align == 2);
+ stmt(
+ IRStmt_Exit(
+ binop(Iop_CmpNE64,
+ binop(Iop_And64,mkexpr(effective_addr),mkU64(align-1)),
+ mkU64(0)),
+ Ijk_SigBUS,
+ IRConst_U64(guest_PC_curr_instr),
+ OFFB_PC
+ )
+ );
+}
+
+
/* Generate a "standard 7" name, from bitQ and size. But also
allow ".1d" since that's occasionally useful. */
static
@@ -6670,7 +6698,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
IRTemp ea = newTemp(Ity_I64);
assign(ea, getIReg64orSP(nn));
- /* FIXME generate check that ea is szB-aligned */
+ gen_SIGBUS_if_not_XX_aligned(ea, szB);
if (isLD && ss == BITS5(1,1,1,1,1)) {
IRTemp res = newTemp(ty);
@@ -6803,7 +6831,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
IRTemp ea = newTemp(Ity_I64);
assign(ea, getIReg64orSP(nn));
- /* FIXME generate check that ea is 2*elemSzB-aligned */
+ gen_SIGBUS_if_not_XX_aligned(ea, fullSzB);
if (isLD && ss == BITS5(1,1,1,1,1)) {
if (abiinfo->guest__use_fallback_LLSC) {
@@ -7044,7 +7072,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
IRTemp ea = newTemp(Ity_I64);
assign(ea, getIReg64orSP(nn));
- /* FIXME generate check that ea is szB-aligned */
+ gen_SIGBUS_if_not_XX_aligned(ea, szB);
if (isLD) {
IRTemp res = newTemp(ty);
@@ -7159,6 +7187,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
IRTemp ea = newTemp(Ity_I64);
assign(ea, getIReg64orSP(nn));
+ gen_SIGBUS_if_not_XX_aligned(ea, szB);
// Insert barrier before loading for acquire and acquire-release variants:
// A and AL.
@@ -7266,6 +7295,10 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
IRType ty = integerIRTypeOfSize(szB);
Bool is64 = szB == 8;
+ IRTemp ea = newTemp(Ity_I64);
+ assign(ea, getIReg64orSP(nn));
+ gen_SIGBUS_if_not_XX_aligned(ea, szB);
+
IRExpr *exp = narrowFrom64(ty, getIReg64orZR(ss));
IRExpr *new = narrowFrom64(ty, getIReg64orZR(tt));
@@ -7275,7 +7308,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
// Store the result back if LHS remains unchanged in memory.
IRTemp old = newTemp(ty);
stmt( IRStmt_CAS(mkIRCAS(/*oldHi*/IRTemp_INVALID, old,
- Iend_LE, getIReg64orSP(nn),
+ Iend_LE, mkexpr(ea),
/*expdHi*/NULL, exp,
/*dataHi*/NULL, new)) );
@@ -7307,6 +7340,10 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
if ((ss & 0x1) || (tt & 0x1)) {
/* undefined; fall through */
} else {
+ IRTemp ea = newTemp(Ity_I64);
+ assign(ea, getIReg64orSP(nn));
+ gen_SIGBUS_if_not_XX_aligned(ea, is64 ? 16 : 8);
+
IRExpr *expLo = getIRegOrZR(is64, ss);
IRExpr *expHi = getIRegOrZR(is64, ss + 1);
IRExpr *newLo = getIRegOrZR(is64, tt);
@@ -7318,7 +7355,7 @@ Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn,
stmt(IRStmt_MBE(Imbe_Fence));
stmt( IRStmt_CAS(mkIRCAS(oldHi, oldLo,
- Iend_LE, getIReg64orSP(nn),
+ Iend_LE, mkexpr(ea),
expHi, expLo,
newHi, newLo)) );
diff --git a/VEX/priv/host_arm64_defs.c b/VEX/priv/host_arm64_defs.c
index b65e27db4..39c6aaa46 100644
--- a/VEX/priv/host_arm64_defs.c
+++ b/VEX/priv/host_arm64_defs.c
@@ -4033,6 +4033,7 @@ Int emit_ARM64Instr ( /*MB_MOD*/Bool* is_profInc,
case Ijk_FlushDCache: trcval = VEX_TRC_JMP_FLUSHDCACHE; break;
case Ijk_NoRedir: trcval = VEX_TRC_JMP_NOREDIR; break;
case Ijk_SigTRAP: trcval = VEX_TRC_JMP_SIGTRAP; break;
+ case Ijk_SigBUS: trcval = VEX_TRC_JMP_SIGBUS; break;
//case Ijk_SigSEGV: trcval = VEX_TRC_JMP_SIGSEGV; break;
case Ijk_Boring: trcval = VEX_TRC_JMP_BORING; break;
/* We don't expect to see the following being assisted. */
diff --git a/VEX/priv/host_arm64_isel.c b/VEX/priv/host_arm64_isel.c
index 094e7e74b..82cb2d78c 100644
--- a/VEX/priv/host_arm64_isel.c
+++ b/VEX/priv/host_arm64_isel.c
@@ -4483,6 +4483,7 @@ static void iselStmt ( ISelEnv* env, IRStmt* stmt )
case Ijk_InvalICache:
case Ijk_FlushDCache:
case Ijk_SigTRAP:
+ case Ijk_SigBUS:
case Ijk_Yield: {
HReg r = iselIntExpr_R(env, IRExpr_Const(stmt->Ist.Exit.dst));
addInstr(env, ARM64Instr_XAssisted(r, amPC, cc,
@@ -4576,8 +4577,8 @@ static void iselNext ( ISelEnv* env,
case Ijk_InvalICache:
case Ijk_FlushDCache:
case Ijk_SigTRAP:
- case Ijk_Yield:
- {
+ case Ijk_SigBUS:
+ case Ijk_Yield: {
HReg r = iselIntExpr_R(env, next);
ARM64AMode* amPC = mk_baseblock_64bit_access_amode(offsIP);
addInstr(env, ARM64Instr_XAssisted(r, amPC, ARM64cc_AL, jk));
Reference in New Issue View Git Blame Copy Permalink