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 <f.gruenbichler@proxmox.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #9346
Closes #9403
This commit is contained in:
Brian Behlendorf
2019-10-24 10:17:33 -07:00
committed by GitHub
parent b834b58ae6
commit 10fa254539
19 changed files with 276 additions and 294 deletions
+3 -29
View File
@@ -206,35 +206,9 @@ aes_mod_init(void)
{
int ret;
#if defined(_KERNEL)
/*
* Determine the fastest available implementation. The benchmarks
* are run in dedicated kernel threads to allow Linux 5.0+ kernels
* to use SIMD operations. If for some reason this isn't possible,
* fallback to the generic implementations. See the comment in
* linux/simd_x86.h for additional details. Additionally, this has
* the benefit of allowing them to be run in parallel.
*/
taskqid_t aes_id = taskq_dispatch(system_taskq, aes_impl_init,
NULL, TQ_SLEEP);
taskqid_t gcm_id = taskq_dispatch(system_taskq, gcm_impl_init,
NULL, TQ_SLEEP);
if (aes_id != TASKQID_INVALID) {
taskq_wait_id(system_taskq, aes_id);
} else {
aes_impl_init(NULL);
}
if (gcm_id != TASKQID_INVALID) {
taskq_wait_id(system_taskq, gcm_id);
} else {
gcm_impl_init(NULL);
}
#else
aes_impl_init(NULL);
gcm_impl_init(NULL);
#endif
/* Determine the fastest available implementation. */
aes_impl_init();
gcm_impl_init();
if ((ret = mod_install(&modlinkage)) != 0)
return (ret);