kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [kvm-unit-tests PATCH 1/2] x86: Do not assign values to unaligned pointer to 128 bits
@ 2021-05-11  1:50 Jacob Xu
  2021-05-11  1:50 ` [kvm-unit-tests PATCH 2/2] x86: remove use of compiler's memset from emulator.c Jacob Xu
  2021-06-10 16:14 ` [kvm-unit-tests PATCH 1/2] x86: Do not assign values to unaligned pointer to 128 bits Jacob Xu
  0 siblings, 2 replies; 5+ messages in thread
From: Jacob Xu @ 2021-05-11  1:50 UTC (permalink / raw)
  To: Paolo Bonzini, Jim Mattson, Sean Christopherson; +Cc: kvm, Jacob Xu

When compiled with clang, the following statement gets converted into a
movaps instructions.
mem->u[0] = 5; mem->u[1] = 6; mem->u[2] = 7; mem->u[3] = 8;

Since mem is an unaligned pointer to sse_union, we get a GP when
running. Let's avoid using a pointer to sse_union at all, since doing so
implies that the pointer is aligned to 128 bits.

Fixes: e5e76263b5 ("x86: add additional test cases for sse exceptions to
emulator.c")

Signed-off-by: Jacob Xu <jacobhxu@google.com>
---
 x86/emulator.c | 67 +++++++++++++++++++++++++-------------------------
 1 file changed, 33 insertions(+), 34 deletions(-)

diff --git a/x86/emulator.c b/x86/emulator.c
index 9705073..1d5c172 100644
--- a/x86/emulator.c
+++ b/x86/emulator.c
@@ -645,37 +645,34 @@ static void test_muldiv(long *mem)
 
 typedef unsigned __attribute__((vector_size(16))) sse128;
 
-typedef union {
-    sse128 sse;
-    unsigned u[4];
-} sse_union;
-
-static bool sseeq(sse_union *v1, sse_union *v2)
+static bool sseeq(uint32_t *v1, uint32_t *v2)
 {
     bool ok = true;
     int i;
 
     for (i = 0; i < 4; ++i) {
-	ok &= v1->u[i] == v2->u[i];
+	ok &= v1[i] == v2[i];
     }
 
     return ok;
 }
 
-static __attribute__((target("sse2"))) void test_sse(sse_union *mem)
+static __attribute__((target("sse2"))) void test_sse(uint32_t *mem)
 {
-	sse_union v;
+	sse128 vv;
+	uint32_t *v = (uint32_t *)&vv;
 
 	write_cr0(read_cr0() & ~6); /* EM, TS */
 	write_cr4(read_cr4() | 0x200); /* OSFXSR */
+	memset(&vv, 0, sizeof(vv));
 
 #define TEST_RW_SSE(insn) do { \
-		v.u[0] = 1; v.u[1] = 2; v.u[2] = 3; v.u[3] = 4; \
-		asm(insn " %1, %0" : "=m"(*mem) : "x"(v.sse)); \
-		report(sseeq(&v, mem), insn " (read)"); \
-		mem->u[0] = 5; mem->u[1] = 6; mem->u[2] = 7; mem->u[3] = 8; \
-		asm(insn " %1, %0" : "=x"(v.sse) : "m"(*mem)); \
-		report(sseeq(&v, mem), insn " (write)"); \
+		v[0] = 1; v[1] = 2; v[2] = 3; v[3] = 4; \
+		asm(insn " %1, %0" : "=m"(*mem) : "x"(vv) : "memory"); \
+		report(sseeq(v, mem), insn " (read)"); \
+		mem[0] = 5; mem[1] = 6; mem[2] = 7; mem[3] = 8; \
+		asm(insn " %1, %0" : "=x"(vv) : "m"(*mem) : "memory"); \
+		report(sseeq(v, mem), insn " (write)"); \
 } while (0)
 
 	TEST_RW_SSE("movdqu");
@@ -704,40 +701,41 @@ static void cross_movups_handler(struct ex_regs *regs)
 
 static __attribute__((target("sse2"))) void test_sse_exceptions(void *cross_mem)
 {
-	sse_union v;
-	sse_union *mem;
+	sse128 vv;
+	uint32_t *v = (uint32_t *)&vv;
+	uint32_t *mem;
 	uint8_t *bytes = cross_mem; // aligned on PAGE_SIZE*2
 	void *page2 = (void *)(&bytes[4096]);
 	struct pte_search search;
 	pteval_t orig_pte;
 
 	// setup memory for unaligned access
-	mem = (sse_union *)(&bytes[8]);
+	mem = (uint32_t *)(&bytes[8]);
 
 	// test unaligned access for movups, movupd and movaps
-	v.u[0] = 1; v.u[1] = 2; v.u[2] = 3; v.u[3] = 4;
-	mem->u[0] = 5; mem->u[1] = 6; mem->u[2] = 7; mem->u[3] = 8;
-	asm("movups %1, %0" : "=m"(*mem) : "x"(v.sse));
-	report(sseeq(&v, mem), "movups unaligned");
-
-	v.u[0] = 1; v.u[1] = 2; v.u[2] = 3; v.u[3] = 4;
-	mem->u[0] = 5; mem->u[1] = 6; mem->u[2] = 7; mem->u[3] = 8;
-	asm("movupd %1, %0" : "=m"(*mem) : "x"(v.sse));
-	report(sseeq(&v, mem), "movupd unaligned");
+	v[0] = 1; v[1] = 2; v[2] = 3; v[3] = 4;
+	mem[0] = 5; mem[1] = 6; mem[2] = 8; mem[3] = 9;
+	asm("movups %1, %0" : "=m"(*mem) : "x"(vv) : "memory");
+	report(sseeq(v, mem), "movups unaligned");
+
+	v[0] = 1; v[1] = 2; v[2] = 3; v[3] = 4;
+	mem[0] = 5; mem[1] = 6; mem[2] = 7; mem[3] = 8;
+	asm("movupd %1, %0" : "=m"(*mem) : "x"(vv) : "memory");
+	report(sseeq(v, mem), "movupd unaligned");
 	exceptions = 0;
 	handle_exception(GP_VECTOR, unaligned_movaps_handler);
 	asm("movaps %1, %0\n\t unaligned_movaps_cont:"
-			: "=m"(*mem) : "x"(v.sse));
+			: "=m"(*mem) : "x"(vv));
 	handle_exception(GP_VECTOR, 0);
 	report(exceptions == 1, "unaligned movaps exception");
 
 	// setup memory for cross page access
-	mem = (sse_union *)(&bytes[4096-8]);
-	v.u[0] = 1; v.u[1] = 2; v.u[2] = 3; v.u[3] = 4;
-	mem->u[0] = 5; mem->u[1] = 6; mem->u[2] = 7; mem->u[3] = 8;
+	mem = (uint32_t *)(&bytes[4096-8]);
+	v[0] = 1; v[1] = 2; v[2] = 3; v[3] = 4;
+	mem[0] = 5; mem[1] = 6; mem[2] = 7; mem[3] = 8;
 
-	asm("movups %1, %0" : "=m"(*mem) : "x"(v.sse));
-	report(sseeq(&v, mem), "movups unaligned crosspage");
+	asm("movups %1, %0" : "=m"(*mem) : "x"(vv) : "memory");
+	report(sseeq(v, mem), "movups unaligned crosspage");
 
 	// invalidate second page
 	search = find_pte_level(current_page_table(), page2, 1);
@@ -747,7 +745,8 @@ static __attribute__((target("sse2"))) void test_sse_exceptions(void *cross_mem)
 
 	exceptions = 0;
 	handle_exception(PF_VECTOR, cross_movups_handler);
-	asm("movups %1, %0\n\t cross_movups_cont:" : "=m"(*mem) : "x"(v.sse));
+	asm("movups %1, %0\n\t cross_movups_cont:" : "=m"(*mem) : "x"(vv) :
+			"memory");
 	handle_exception(PF_VECTOR, 0);
 	report(exceptions == 1, "movups crosspage exception");
 
-- 
2.31.1.607.g51e8a6a459-goog


^ permalink raw reply related	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2021-06-10 16:56 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-05-11  1:50 [kvm-unit-tests PATCH 1/2] x86: Do not assign values to unaligned pointer to 128 bits Jacob Xu
2021-05-11  1:50 ` [kvm-unit-tests PATCH 2/2] x86: remove use of compiler's memset from emulator.c Jacob Xu
2021-05-11 21:13   ` Sean Christopherson
2021-06-10 16:14 ` [kvm-unit-tests PATCH 1/2] x86: Do not assign values to unaligned pointer to 128 bits Jacob Xu
2021-06-10 16:55   ` Paolo Bonzini

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).