4ef6ea38fc
arm64 has a similar wrapper. This permits defining <machine/fpu.h> as the standard header for fpu_kern_*. Reviewed by: kib Sponsored by: Netflix Differential Revision: https://reviews.freebsd.org/D26753
216 lines
6.8 KiB
Groff
216 lines
6.8 KiB
Groff
.\" Copyright (c) 2014
|
|
.\" Konstantin Belousov <kib@FreeBSD.org>. All rights reserved.
|
|
.\"
|
|
.\" Redistribution and use in source and binary forms, with or without
|
|
.\" modification, are permitted provided that the following conditions
|
|
.\" are met:
|
|
.\" 1. Redistributions of source code must retain the above copyright
|
|
.\" notice, this list of conditions and the following disclaimer.
|
|
.\" 2. Redistributions in binary form must reproduce the above copyright
|
|
.\" notice, this list of conditions and the following disclaimer in the
|
|
.\" documentation and/or other materials provided with the distribution.
|
|
.\"
|
|
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
|
|
.\" IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
|
|
.\" OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
|
|
.\" IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
.\" INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
|
|
.\" NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
.\" DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
.\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
.\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
|
.\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
.\"
|
|
.\" $FreeBSD$
|
|
.\"
|
|
.Dd October 13, 2020
|
|
.Dt FPU_KERN 9
|
|
.Os
|
|
.Sh NAME
|
|
.Nm fpu_kern
|
|
.Nd "facility to use the FPU in the kernel"
|
|
.Sh SYNOPSIS
|
|
.In machine/fpu.h
|
|
.Ft struct fpu_kern_ctx *
|
|
.Fn fpu_kern_alloc_ctx "u_int flags"
|
|
.Ft void
|
|
.Fn fpu_kern_free_ctx "struct fpu_kern_ctx *ctx"
|
|
.Ft void
|
|
.Fn fpu_kern_enter "struct thread *td" "struct fpu_kern_ctx *ctx" "u_int flags"
|
|
.Ft int
|
|
.Fn fpu_kern_leave "struct thread *td" "struct fpu_kern_ctx *ctx"
|
|
.Ft int
|
|
.Fn fpu_kern_thread "u_int flags"
|
|
.Ft int
|
|
.Fn is_fpu_kern_thread "u_int flags"
|
|
.Sh DESCRIPTION
|
|
The
|
|
.Nm
|
|
family of functions allows the use of FPU hardware in kernel code.
|
|
Modern FPUs are not limited to providing hardware implementation for
|
|
floating point arithmetic; they offer advanced accelerators for cryptography
|
|
and other computational-intensive algorithms.
|
|
These facilities share registers with the FPU hardware.
|
|
.Pp
|
|
Typical kernel code does not need access to the FPU.
|
|
Saving a large register file on each entry to the kernel would waste
|
|
time.
|
|
When kernel code uses the FPU, the current FPU state must be saved to
|
|
avoid corrupting the user-mode state, and vice versa.
|
|
.Pp
|
|
The management of the save and restore is automatic.
|
|
The processor catches accesses to the FPU registers
|
|
when the non-current context tries to access them.
|
|
Explicit calls are required for the allocation of the save area and
|
|
the notification of the start and end of the code using the FPU.
|
|
.Pp
|
|
The
|
|
.Fn fpu_kern_alloc_ctx
|
|
function allocates the memory used by
|
|
.Nm
|
|
to track the use of the FPU hardware state and the related software state.
|
|
The
|
|
.Fn fpu_kern_alloc_ctx
|
|
function requires the
|
|
.Fa flags
|
|
argument, which currently accepts the following flags:
|
|
.Bl -tag -width ".Dv FPU_KERN_NOWAIT" -offset indent
|
|
.It Dv FPU_KERN_NOWAIT
|
|
Do not wait for the available memory if the request could not be satisfied
|
|
without sleep.
|
|
.It 0
|
|
No special handling is required.
|
|
.El
|
|
.Pp
|
|
The function returns the allocated context area, or
|
|
.Va NULL
|
|
if the allocation failed.
|
|
.Pp
|
|
The
|
|
.Fn fpu_kern_free_ctx
|
|
function frees the context previously allocated by
|
|
.Fn fpu_kern_alloc_ctx .
|
|
.Pp
|
|
The
|
|
.Fn fpu_kern_enter
|
|
function designates the start of the region of kernel code where the
|
|
use of the FPU is allowed.
|
|
Its arguments are:
|
|
.Bl -tag -width ".Fa ctx" -offset indent
|
|
.It Fa td
|
|
Currently must be
|
|
.Va curthread .
|
|
.It Fa ctx
|
|
The context save area previously allocated by
|
|
.Fn fpu_kern_alloc_ctx
|
|
and not currently in use by another call to
|
|
.Fn fpu_kern_enter .
|
|
.It Fa flags
|
|
This argument currently accepts the following flags:
|
|
.Bl -tag -width ".Dv FPU_KERN_NORMAL" -offset indent
|
|
.It Dv FPU_KERN_NORMAL
|
|
Indicates that the caller intends to access the full FPU state.
|
|
Must be specified currently.
|
|
.It Dv FPU_KERN_KTHR
|
|
Indicates that no saving of the current FPU state should be performed,
|
|
if the thread called
|
|
.Xr fpu_kern_thread 9
|
|
function.
|
|
This is intended to minimize code duplication in callers which
|
|
could be used from both kernel thread and syscall contexts.
|
|
The
|
|
.Fn fpu_kern_leave
|
|
function correctly handles such contexts.
|
|
.It Dv FPU_KERN_NOCTX
|
|
Avoid nesting save area.
|
|
If the flag is specified, the
|
|
.Fa ctx
|
|
must be passed as
|
|
.Va NULL .
|
|
The flag should only be used for really short code blocks
|
|
which can be executed in a critical section.
|
|
It avoids the need to allocate the FPU context by the cost
|
|
of increased system latency.
|
|
.El
|
|
.El
|
|
.Pp
|
|
The function does not sleep or block.
|
|
It could cause an FPU trap during execution, and on the first FPU access
|
|
after the function returns, as well as after each context switch.
|
|
On i386 and amd64 this will be the
|
|
.Nm Device Not Available
|
|
exception (see Intel Software Developer Manual for the reference).
|
|
.Pp
|
|
The
|
|
.Fn fpu_kern_leave
|
|
function ends the region started by
|
|
.Fn fpu_kern_enter .
|
|
It is erroneous to use the FPU in the kernel before
|
|
.Fn fpu_kern_enter
|
|
or after
|
|
.Fn fpu_kern_leave .
|
|
The function takes the
|
|
.Fa td
|
|
thread argument, which currently must be
|
|
.Va curthread ,
|
|
and the
|
|
.Fa ctx
|
|
context pointer, previously passed to
|
|
.Fn fpu_kern_enter .
|
|
After the function returns, the context may be freed or reused
|
|
by another invocation of
|
|
.Fn fpu_kern_enter .
|
|
The function always returns 0.
|
|
.Pp
|
|
The
|
|
.Fn fpu_kern_thread
|
|
function enables an optimization for threads which never leave to
|
|
the usermode.
|
|
The current thread will reuse the usermode save area for the kernel FPU state
|
|
instead of requiring an explicitly allocated context.
|
|
There are no flags defined for the function, and no error states
|
|
that the function returns.
|
|
Once this function has been called, neither
|
|
.Fn fpu_kern_enter
|
|
nor
|
|
.Fn fpu_kern_leave
|
|
is required to be called and the fpu is available for use in the calling thread.
|
|
.Pp
|
|
The
|
|
.Fn is_fpu_kern_thread
|
|
function returns the boolean indicating whether the current thread
|
|
entered the mode enabled by
|
|
.Fn fpu_kern_thread .
|
|
There is currently no flags defined for the function, the return
|
|
value is true if the current thread have the permanent FPU save area,
|
|
and false otherwise.
|
|
.Sh NOTES
|
|
The
|
|
.Nm
|
|
is currently implemented only for the i386, amd64, and arm64 architectures.
|
|
.Pp
|
|
There is no way to handle floating point exceptions raised from
|
|
kernel mode.
|
|
.Pp
|
|
The unused
|
|
.Fa flags
|
|
arguments
|
|
to the
|
|
.Nm
|
|
functions are to be extended to allow specification of the
|
|
set of the FPU hardware state used by the code region.
|
|
This would allow optimizations of saving and restoring the state.
|
|
.Sh AUTHORS
|
|
The
|
|
.Nm
|
|
facitily and this manual page were written by
|
|
.An Konstantin Belousov Aq Mt kib@FreeBSD.org .
|
|
The arm64 support was added by
|
|
.An Andrew Turner Aq Mt andrew@FreeBSD.org .
|
|
.Sh BUGS
|
|
.Fn fpu_kern_leave
|
|
should probably have type
|
|
.Ft void
|
|
(like
|
|
.Fn fpu_kern_enter ) .
|