s390x: choose SIMD kernels at run-time based on OS and compiler support

Extend and simplify the run-time detection for dynamic architecture support for z
to check HW_CAP and only use SIMD features if advertised by the OS.
While at it, also honor the env variable LD_HWCAP_MASK and do not use
the CPU features masked there.

Note that we can only use the SIMD features on z13 or newer (i.e.,
Vector Facility or Vector-Enhancements Facilities) when the operating
system supports properly context-switching the vector registers. The OS
advertises that support as a bit in the HW_CAP value in the auxiliary
vector. While all recent Linux kernels have that support, we should
maintain compatibility with older versions that may still be in use.

Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>

driver/others/dynamic_zarch.c

@@ -13,6 +13,39 @@
 #define HAVE_Z14_SUPPORT
 #endif
 
+// Guard the use of getauxval() on glibc version >= 2.16
+#ifdef __GLIBC__
+#include <features.h>
+#if __GLIBC_PREREQ(2, 16)
+#include <sys/auxv.h>
+#define HAVE_GETAUXVAL 1
+
+static unsigned long get_hwcap(void)
+{
+	unsigned long hwcap = getauxval(AT_HWCAP);
+	char *maskenv;
+
+	// honor requests for not using specific CPU features in LD_HWCAP_MASK
+	maskenv = getenv("LD_HWCAP_MASK");
+	if (maskenv)
+		hwcap &= strtoul(maskenv, NULL, 0);
+
+	return hwcap;
+	// note that a missing auxval is interpreted as no capabilities
+	// available, which is safe.
+}
+
+#else // __GLIBC_PREREQ(2, 16)
+#warn "Cannot detect SIMD support in Z13 or newer architectures since glibc is older than 2.16"
+
+static unsigned long get_hwcap(void) {
+	// treat missing support for getauxval() as no capabilities available,
+	// which is safe.
+	return 0;
+}
+#endif // __GLIBC_PREREQ(2, 16)
+#endif // __GLIBC
+
 extern gotoblas_t gotoblas_ZARCH_GENERIC;
 #ifdef HAVE_Z13_SUPPORT
 extern gotoblas_t gotoblas_Z13;
@@ -44,39 +77,34 @@ char* gotoblas_corename(void) {
 	return corename[0];
 }
 
-// __builtin_cpu_is is not supported by zarch
+/**
+ * Detect the fitting set of kernels by retrieving the CPU features supported by
+ * OS from the auxiliary value AT_HWCAP and choosing the set of kernels
+ * ("coretype") that exploits most of the features and can be compiled with the
+ * available gcc version.
+ * Note that we cannot use vector registers on a z13 or newer unless supported
+ * by the OS kernel (which needs to handle them properly during context switch).
+ */
 static gotoblas_t* get_coretype(void) {
-	FILE* infile;
-	char buffer[512], * p;
-
-	p = (char*)NULL;
-	infile = fopen("/proc/sysinfo", "r");
-	while (fgets(buffer, sizeof(buffer), infile)) {
-		if (!strncmp("Type", buffer, 4)) {
-			p = strchr(buffer, ':') + 2;
-#if 0
-			fprintf(stderr, "%s\n", p);
-#endif
-			break;
-		}
-	}
-
-	fclose(infile);
 
-#ifdef HAVE_Z13_SUPPORT
-	if (strstr(p, "2964") || strstr(p, "2965")) return &gotoblas_Z13;
-#endif
+	unsigned long hwcap __attribute__((unused)) = get_hwcap();
 
-	// Z14 and Z15 systems
-	if (strstr(p, "3906") || strstr(p, "3907") || strstr(p, "8561") ||
-	    strstr(p, "8562"))
+	// z14 and z15 systems: exploit Vector Facility (SIMD) and
+	// Vector-Enhancements Facility 1 (float SIMD instructions), if present.
 #ifdef HAVE_Z14_SUPPORT
+	if ((hwcap & HWCAP_S390_VX) && (hwcap & HWCAP_S390_VXE))
 		return &gotoblas_Z14;
-#else
+#endif
+
+	// z13: Vector Facility (SIMD for double)
+#ifdef HAVE_Z13_SUPPORT
+	if (hwcap & HWCAP_S390_VX)
 		return &gotoblas_Z13;
 #endif
 
-	// unknown system or compiler too old? use generic code for z architecture
+	// fallback in case of missing compiler support, systems before z13, or
+	// when the OS does not advertise support for the Vector Facility (e.g.,
+	// missing support in the OS kernel)
 	return &gotoblas_ZARCH_GENERIC;
 }