Linux 4.14, 4.19, 5.0+ compat: SIMD save/restore

Contrary to initial testing we cannot rely on these kernels to
invalidate the per-cpu FPU state and restore the FPU registers.
Nor can we guarantee that the kernel won't modify the FPU state
which we saved in the task struck.

Therefore, the kfpu_begin() and kfpu_end() functions have been
updated to save and restore the FPU state using our own dedicated
per-cpu FPU state variables.

This has the additional advantage of allowing us to use the FPU
again in user threads.  So we remove the code which was added to
use task queues to ensure some functions ran in kernel threads.

Reviewed-by: Fabian Grünbichler <[email protected]>
Reviewed-by: Tony Hutter <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Issue #9346
Closes #9403 

2 files changed, 18 insertions, 16 deletions

diff --git a/module/zcommon/zfs_fletcher.c b/module/zcommon/zfs_fletcher.c
index 3b4052c8a..1280ace31 100644
--- a/module/zcommon/zfs_fletcher.c
+++ b/module/zcommon/zfs_fletcher.c
@@ -726,7 +726,7 @@ fletcher_4_benchmark_impl(boolean_t native, char *data, uint64_t data_size)
  * Initialize and benchmark all supported implementations.
  */
 static void
-fletcher_4_benchmark(void *arg)
+fletcher_4_benchmark(void)
 {
 	fletcher_4_ops_t *curr_impl;
 	int i, c;
@@ -769,20 +769,10 @@ fletcher_4_benchmark(void *arg)
 void
 fletcher_4_init(void)
 {
-#if defined(_KERNEL)
-	/*
-	 * For 5.0 and latter Linux kernels the fletcher 4 benchmarks are
-	 * run in a kernel threads.  This is needed to take advantage of the
-	 * SIMD functionality, see linux/simd_x86.h for details.
-	 */
-	taskqid_t id = taskq_dispatch(system_taskq, fletcher_4_benchmark,
-	    NULL, TQ_SLEEP);
-	if (id != TASKQID_INVALID) {
-		taskq_wait_id(system_taskq, id);
-	} else {
-		fletcher_4_benchmark(NULL);
-	}
+	/* Determine the fastest available implementation. */
+	fletcher_4_benchmark();
 
+#if defined(_KERNEL)
 	/* Install kstats for all implementations */
 	fletcher_4_kstat = kstat_create("zfs", 0, "fletcher_4_bench", "misc",
 	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VIRTUAL);
@@ -795,8 +785,6 @@ fletcher_4_init(void)
 		    fletcher_4_kstat_addr);
 		kstat_install(fletcher_4_kstat);
 	}
-#else
-	fletcher_4_benchmark(NULL);
 #endif
 
 	/* Finish initialization */
diff --git a/module/zcommon/zfs_prop.c b/module/zcommon/zfs_prop.c
index c42f046da..10b521065 100644
--- a/module/zcommon/zfs_prop.c
+++ b/module/zcommon/zfs_prop.c
@@ -865,10 +865,23 @@ zfs_prop_align_right(zfs_prop_t prop)
 #endif
 
 #if defined(_KERNEL)
+
+#include <sys/simd.h>
+
+#if defined(HAVE_KERNEL_FPU_INTERNAL)
+union fpregs_state **zfs_kfpu_fpregs;
+EXPORT_SYMBOL(zfs_kfpu_fpregs);
+#endif /* HAVE_KERNEL_FPU_INTERNAL */
+
 static int __init
 zcommon_init(void)
 {
+	int error = kfpu_init();
+	if (error)
+		return (error);
+
 	fletcher_4_init();
+
 	return (0);
 }
 
@@ -876,6 +889,7 @@ static void __exit
 zcommon_fini(void)
 {
 	fletcher_4_fini();
+	kfpu_fini();
 }
 
 module_init(zcommon_init);