Skip to contentSkip to conversation

Add missing trace points in emulator path

PATCH Applied Hou Wenlong 2 participants 4 messages Sep 2, 2022 – Sep 20, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

v2 cover Add missing trace points in emulator path

Some existed trace points are missing in emulator path, e.g., RDMSR/WRMSR emulation and CR read/write emulation. However, if add those trace points in emulator common interfaces in arch/x86/kvm/x86.c, other instruction emulation may use those interfaces too and cause too much trace records. But add those trace points in em_* functions in arch/x86/kvm/emulate.c seems to be ugly. Luckily, RDMSR/WRMSR emulation uses a sepreate interface, so add trace points for RDMSR/WRMSR in emulator path is acceptable like normal path.

Changed from v1:
- As Sean suggested, use X86EMUL_PROPAGATE_FAULT instead of X86EMUL_UNHANDLEABLE for error path.
- As Sean suggested, move "r < 0" handling into the set helper, and add "r < 0" check in the get helper.

v1: https://lore.kernel.org/kvm/[email protected]

Hou Wenlong (2):
KVM: x86: Return emulator error if RDMSR/WRMSR emulation failed
KVM: x86: Add missing trace points for RDMSR/WRMSR in emulator path

 arch/x86/kvm/emulate.c | 20 ++++++++------------
 arch/x86/kvm/x86.c     | 32 ++++++++++++++++++++++----------
 2 files changed, 30 insertions(+), 22 deletions(-)

--
2.31.1

#1 lore raw

PATCH v2 1/2 KVM: x86: Return emulator error if RDMSR/WRMSR emulation failed

The return value of emulator_{get|set}_mst_with_filter() is confused, since msr access error and emulator error are mixed. Although, KVM_MSR_RET_* doesn't conflict with X86EMUL_IO_NEEDED at present, it is better to convert msr access error to emulator error if error value is needed. So move "r < 0" handling for wrmsr emulation into the set helper function, then only X86EMUL_* is returned in the helper functions. Also add "r < 0" check in the get helper function, although KVM doesn't return -errno today, but assuming that will always hold true is unnecessarily risking.

---
 arch/x86/kvm/emulate.c | 20 ++++++++------------
 arch/x86/kvm/x86.c     | 26 ++++++++++++++++----------
 2 files changed, 24 insertions(+), 22 deletions(-)
arch/x86/kvm/emulate.c +8 -12
index f092c54d1a2f..d16f7d6a81bb 100644
@@ -3645,13 +3645,10 @@ static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
 
-	if (r == X86EMUL_IO_NEEDED)
-		return r;
-
-	if (r > 0)
+	if (r == X86EMUL_PROPAGATE_FAULT)
 		return emulate_gp(ctxt, 0);
 
-	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
+	return r;
 }
 
 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
@@ -3662,15 +3659,14 @@ static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
 
 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
 
-	if (r == X86EMUL_IO_NEEDED)
-		return r;
-
-	if (r)
+	if (r == X86EMUL_PROPAGATE_FAULT)
 		return emulate_gp(ctxt, 0);
 
-	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
-	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
-	return X86EMUL_CONTINUE;
+	if (r == X86EMUL_CONTINUE) {
+		*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
+		*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
+	}
+	return r;
 }
 
 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
arch/x86/kvm/x86.c +16 -10
index d7374d768296..2596f5736743 100644
@@ -7920,14 +7920,17 @@ static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
 	int r;
 
 	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
+	if (r < 0)
+		return X86EMUL_UNHANDLEABLE;
 
-	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
-				    complete_emulated_rdmsr, r)) {
-		/* Bounce to user space */
-		return X86EMUL_IO_NEEDED;
+	if (r) {
+		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
+				       complete_emulated_rdmsr, r))
+			return X86EMUL_IO_NEEDED;
+		return X86EMUL_PROPAGATE_FAULT;
 	}
 
-	return r;
+	return X86EMUL_CONTINUE;
 }
 
 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
@@ -7937,14 +7940,17 @@ static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
 	int r;
 
 	r = kvm_set_msr_with_filter(vcpu, msr_index, data);
+	if (r < 0)
+		return X86EMUL_UNHANDLEABLE;
 
-	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
-				    complete_emulated_msr_access, r)) {
-		/* Bounce to user space */
-		return X86EMUL_IO_NEEDED;
+	if (r) {
+		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
+				       complete_emulated_msr_access, r))
+			return X86EMUL_IO_NEEDED;
+		return X86EMUL_PROPAGATE_FAULT;
 	}
 
-	return r;
+	return X86EMUL_CONTINUE;
 }
 
 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
Signature · 1 line
2.31.1
#2 lore raw

PATCH v2 2/2 KVM: x86: Add missing trace points for RDMSR/WRMSR in emulator path

In reply to Hou Wenlong

Since the RDMSR/WRMSR emulation uses a sepearte emualtor interface, the trace points for RDMSR/WRMSR can be added in emulator path like normal path.

---
 arch/x86/kvm/x86.c | 6 ++++++
 1 file changed, 6 insertions(+)
arch/x86/kvm/x86.c +6 -0
index 2596f5736743..1a224e575b0b 100644
@@ -7927,9 +7927,12 @@ static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
 		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
 				       complete_emulated_rdmsr, r))
 			return X86EMUL_IO_NEEDED;
+
+		trace_kvm_msr_read_ex(msr_index);
 		return X86EMUL_PROPAGATE_FAULT;
 	}
 
+	trace_kvm_msr_read(msr_index, *pdata);
 	return X86EMUL_CONTINUE;
 }
 
@@ -7947,9 +7950,12 @@ static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
 		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
 				       complete_emulated_msr_access, r))
 			return X86EMUL_IO_NEEDED;
+
+		trace_kvm_msr_write_ex(msr_index, data);
 		return X86EMUL_PROPAGATE_FAULT;
 	}
 
+	trace_kvm_msr_write(msr_index, data);
 	return X86EMUL_CONTINUE;
 }
 
Signature · 1 line
2.31.1
maintainer Applied
#3 lore raw
In reply to Hou Wenlong
Quoted 16 lines · Hou Wenlong

Some existed trace points are missing in emulator path, e.g., RDMSR/WRMSR emulation and CR read/write emulation. However, if add those trace points in emulator common interfaces in arch/x86/kvm/x86.c, other instruction emulation may use those interfaces too and cause too much trace records. But add those trace points in em_* functions in arch/x86/kvm/emulate.c seems to be ugly. Luckily, RDMSR/WRMSR emulation uses a sepreate interface, so add trace points for RDMSR/WRMSR in emulator path is acceptable like normal path.

Changed from v1:
- As Sean suggested, use X86EMUL_PROPAGATE_FAULT instead of X86EMUL_UNHANDLEABLE for error path.
- As Sean suggested, move "r < 0" handling into the set helper, and add "r < 0" check in the get helper.

Quoted 5 lines

v1: https://lore.kernel.org/kvm/[email protected]

Hou Wenlong (2):
KVM: x86: Return emulator error if RDMSR/WRMSR emulation failed
KVM: x86: Add missing trace points for RDMSR/WRMSR in emulator path

Pushed to branch for_paolo/6.1 at:

https://github.com/sean-jc/linux.git

Unless you hear otherwise, it will make its way to kvm/queue "soon".