diff mbox series

[v4] eal: fix unaligned loads/stores in rte_memcpy_generic

Message ID 20220116203312.6405-1-lucp.at.work@gmail.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers show
Series [v4] eal: fix unaligned loads/stores in rte_memcpy_generic | expand

Checks

Context Check Description
ci/iol-aarch64-compile-testing success Testing PASS
ci/iol-x86_64-compile-testing success Testing PASS
ci/iol-x86_64-unit-testing success Testing PASS
ci/iol-abi-testing success Testing PASS
ci/iol-aarch64-unit-testing success Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-broadcom-Performance success Performance Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/github-robot: build success github build: passed
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-broadcom-Functional success Functional Testing PASS
ci/intel-Testing success Testing PASS
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Luc Pelletier Jan. 16, 2022, 8:33 p.m. UTC
Calls to rte_memcpy_generic could result in unaligned loads/stores for
1 < n < 16. This is undefined behavior according to the C standard,
and it gets flagged by the clang undefined behavior sanitizer.

rte_memcpy_generic is called with unaligned src and dst addresses.
When 1 < n < 16, the code would cast both src and dst to a qword,
dword or word pointer, without verifying the alignment of src/dst. The
code was changed to use inline assembly for the load/store operations.
Unaligned load/store operations are permitted in x86/x64 assembly.

Fixes: d35cc1fe6a7a ("eal/x86: revert select optimized memcpy at run-time")
Cc: Xiaoyun Li <xiaoyun.li@intel.com>
Cc: stable@dpdk.org

Signed-off-by: Luc Pelletier <lucp.at.work@gmail.com>
---

Added volatile for asm statements. I had been testing with clang and it
was working but it seems gcc requires this, otherwise the asm is
completely discarded by the optimizer. The following note in the gcc
docs seems to explain why:

"If the C code that follows the asm makes no use of any of the output
operands, use volatile for the asm statement to prevent the optimizers
from discarding the asm statement as unneeded (see Volatile)."

 lib/eal/x86/include/rte_memcpy.h | 134 +++++++++++++++++--------------
 1 file changed, 72 insertions(+), 62 deletions(-)
diff mbox series

Patch

diff --git a/lib/eal/x86/include/rte_memcpy.h b/lib/eal/x86/include/rte_memcpy.h
index 1b6c6e585f..5af5c2ed61 100644
--- a/lib/eal/x86/include/rte_memcpy.h
+++ b/lib/eal/x86/include/rte_memcpy.h
@@ -45,6 +45,75 @@  extern "C" {
 static __rte_always_inline void *
 rte_memcpy(void *dst, const void *src, size_t n);
 
+#if defined(__i386__)
+	#define RTE_ACCUMULATOR_REGISTER_NAME "eax"
+#elif defined(__x86_64__)
+	#define RTE_ACCUMULATOR_REGISTER_NAME "rax"
+#endif
+
+/**
+ * Copy bytes from one location to another,
+ * locations should not overlap.
+ * Use with unaligned src/dst, and n <= 15.
+ */
+static __rte_always_inline void *
+rte_mov15_or_less_unaligned(void *dst, const void *src, size_t n)
+{
+	void *ret = dst;
+	if (n & 8) {
+		asm volatile(
+#if defined(__i386__)
+		"movl (%[src]), %%eax\n"
+		"movl %%eax, (%[dst])\n"
+		"add $4, %[src]\n"
+		"add $4, %[dst]\n"
+		"movl (%[src]), %%eax\n"
+		"movl %%eax, (%[dst])\n"
+		"add $4, %[src]\n"
+		"add $4, %[dst]\n"
+#elif defined(__x86_64__)
+		"movq (%[src]), %%rax\n"
+		"movq %%rax, (%[dst])\n"
+		"add $8, %[src]\n"
+		"add $8, %[dst]\n"
+#else
+		#error Unsupported architecture
+#endif
+		: [dst] "+r" (dst), [src] "+r" (src)
+		:
+		: RTE_ACCUMULATOR_REGISTER_NAME, "memory");
+	}
+	if (n & 4) {
+		asm volatile(
+		"movl (%[src]), %%eax\n"
+		"movl %%eax, (%[dst])\n"
+		"add $4, %[src]\n"
+		"add $4, %[dst]\n"
+		: [dst] "+r" (dst), [src] "+r" (src)
+		:
+		: RTE_ACCUMULATOR_REGISTER_NAME, "memory");
+	}
+	if (n & 2) {
+		asm volatile(
+		"movw (%[src]), %%ax\n"
+		"movw %%ax, (%[dst])\n"
+		"add $2, %[src]\n"
+		"add $2, %[dst]\n"
+		: [dst] "+r" (dst), [src] "+r" (src)
+		:
+		: RTE_ACCUMULATOR_REGISTER_NAME, "memory");
+	}
+	if (n & 1) {
+		asm volatile(
+		"movb (%[src]), %%al\n"
+		"movb %%al, (%[dst])\n"
+		: [dst] "+r" (dst), [src] "+r" (src)
+		:
+		: RTE_ACCUMULATOR_REGISTER_NAME, "memory");
+	}
+	return ret;
+}
+
 #if defined __AVX512F__ && defined RTE_MEMCPY_AVX512
 
 #define ALIGNMENT_MASK 0x3F
@@ -171,8 +240,6 @@  rte_mov512blocks(uint8_t *dst, const uint8_t *src, size_t n)
 static __rte_always_inline void *
 rte_memcpy_generic(void *dst, const void *src, size_t n)
 {
-	uintptr_t dstu = (uintptr_t)dst;
-	uintptr_t srcu = (uintptr_t)src;
 	void *ret = dst;
 	size_t dstofss;
 	size_t bits;
@@ -181,24 +248,7 @@  rte_memcpy_generic(void *dst, const void *src, size_t n)
 	 * Copy less than 16 bytes
 	 */
 	if (n < 16) {
-		if (n & 0x01) {
-			*(uint8_t *)dstu = *(const uint8_t *)srcu;
-			srcu = (uintptr_t)((const uint8_t *)srcu + 1);
-			dstu = (uintptr_t)((uint8_t *)dstu + 1);
-		}
-		if (n & 0x02) {
-			*(uint16_t *)dstu = *(const uint16_t *)srcu;
-			srcu = (uintptr_t)((const uint16_t *)srcu + 1);
-			dstu = (uintptr_t)((uint16_t *)dstu + 1);
-		}
-		if (n & 0x04) {
-			*(uint32_t *)dstu = *(const uint32_t *)srcu;
-			srcu = (uintptr_t)((const uint32_t *)srcu + 1);
-			dstu = (uintptr_t)((uint32_t *)dstu + 1);
-		}
-		if (n & 0x08)
-			*(uint64_t *)dstu = *(const uint64_t *)srcu;
-		return ret;
+		return rte_mov15_or_less_unaligned(dst, src, n);
 	}
 
 	/**
@@ -379,8 +429,6 @@  rte_mov128blocks(uint8_t *dst, const uint8_t *src, size_t n)
 static __rte_always_inline void *
 rte_memcpy_generic(void *dst, const void *src, size_t n)
 {
-	uintptr_t dstu = (uintptr_t)dst;
-	uintptr_t srcu = (uintptr_t)src;
 	void *ret = dst;
 	size_t dstofss;
 	size_t bits;
@@ -389,25 +437,7 @@  rte_memcpy_generic(void *dst, const void *src, size_t n)
 	 * Copy less than 16 bytes
 	 */
 	if (n < 16) {
-		if (n & 0x01) {
-			*(uint8_t *)dstu = *(const uint8_t *)srcu;
-			srcu = (uintptr_t)((const uint8_t *)srcu + 1);
-			dstu = (uintptr_t)((uint8_t *)dstu + 1);
-		}
-		if (n & 0x02) {
-			*(uint16_t *)dstu = *(const uint16_t *)srcu;
-			srcu = (uintptr_t)((const uint16_t *)srcu + 1);
-			dstu = (uintptr_t)((uint16_t *)dstu + 1);
-		}
-		if (n & 0x04) {
-			*(uint32_t *)dstu = *(const uint32_t *)srcu;
-			srcu = (uintptr_t)((const uint32_t *)srcu + 1);
-			dstu = (uintptr_t)((uint32_t *)dstu + 1);
-		}
-		if (n & 0x08) {
-			*(uint64_t *)dstu = *(const uint64_t *)srcu;
-		}
-		return ret;
+		return rte_mov15_or_less_unaligned(dst, src, n);
 	}
 
 	/**
@@ -672,8 +702,6 @@  static __rte_always_inline void *
 rte_memcpy_generic(void *dst, const void *src, size_t n)
 {
 	__m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8;
-	uintptr_t dstu = (uintptr_t)dst;
-	uintptr_t srcu = (uintptr_t)src;
 	void *ret = dst;
 	size_t dstofss;
 	size_t srcofs;
@@ -682,25 +710,7 @@  rte_memcpy_generic(void *dst, const void *src, size_t n)
 	 * Copy less than 16 bytes
 	 */
 	if (n < 16) {
-		if (n & 0x01) {
-			*(uint8_t *)dstu = *(const uint8_t *)srcu;
-			srcu = (uintptr_t)((const uint8_t *)srcu + 1);
-			dstu = (uintptr_t)((uint8_t *)dstu + 1);
-		}
-		if (n & 0x02) {
-			*(uint16_t *)dstu = *(const uint16_t *)srcu;
-			srcu = (uintptr_t)((const uint16_t *)srcu + 1);
-			dstu = (uintptr_t)((uint16_t *)dstu + 1);
-		}
-		if (n & 0x04) {
-			*(uint32_t *)dstu = *(const uint32_t *)srcu;
-			srcu = (uintptr_t)((const uint32_t *)srcu + 1);
-			dstu = (uintptr_t)((uint32_t *)dstu + 1);
-		}
-		if (n & 0x08) {
-			*(uint64_t *)dstu = *(const uint64_t *)srcu;
-		}
-		return ret;
+		return rte_mov15_or_less_unaligned(dst, src, n);
 	}
 
 	/**