From: Andy Lutomirski <luto@amacapital.net>
To: Rik van Riel <riel@redhat.com>
Cc: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
X86 ML <x86@kernel.org>, Thomas Gleixner <tglx@linutronix.de>,
Paolo Bonzini <pbonzini@redhat.com>,
Ingo Molnar <mingo@redhat.com>,
Andrew Lutomirski <luto@kernel.org>,
"H. Peter Anvin" <hpa@zytor.com>,
Dave Hansen <dave.hansen@linux.intel.com>,
Borislav Petkov <bp@suse.de>
Subject: Re: [PATCH RFC 2/5] x86,fpu: delay FPU register loading until switch to userspace
Date: Sat, 1 Oct 2016 16:44:19 -0700 [thread overview]
Message-ID: <CALCETrWnhA4Gd5AxDEJz7d-VQNDN7O8ayyna7fz3=HNtNnARxQ@mail.gmail.com> (raw)
In-Reply-To: <1475353895-22175-3-git-send-email-riel@redhat.com>
On Sat, Oct 1, 2016 at 1:31 PM, <riel@redhat.com> wrote:
> From: Rik van Riel <riel@redhat.com>
>
> Delay the loading of FPU registers until a process switches back to
> userspace. This allows us to skip FPU saving & restoring for kernel
> threads, the idle task, and tasks that are spinning in kernel space.
>
> It also allows us to not repeatedly save & restore the userspace FPU
> context on repeated invocations of kernel_fpu_start & kernel_fpu_end.
>
> Not overwriting the FPU state of a task unless we need to also allows
> us to be be lazier about restoring it, in a future patch.
>
> Signed-off-by: Rik van Riel <riel@redhat.com>
> ---
> arch/x86/entry/common.c | 4 ++++
> arch/x86/include/asm/fpu/api.h | 5 +++++
> arch/x86/include/asm/fpu/internal.h | 44 +++++++++----------------------------
> arch/x86/include/asm/thread_info.h | 4 +++-
> arch/x86/kernel/fpu/core.c | 17 ++++++++------
> arch/x86/kernel/process.c | 35 +++++++++++++++++++++++++++++
> arch/x86/kernel/process_32.c | 5 ++---
> arch/x86/kernel/process_64.c | 5 ++---
> 8 files changed, 71 insertions(+), 48 deletions(-)
>
> diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
> index 1433f6b4607d..a69bbefa3408 100644
> --- a/arch/x86/entry/common.c
> +++ b/arch/x86/entry/common.c
> @@ -27,6 +27,7 @@
> #include <asm/vdso.h>
> #include <asm/uaccess.h>
> #include <asm/cpufeature.h>
> +#include <asm/fpu/api.h>
>
> #define CREATE_TRACE_POINTS
> #include <trace/events/syscalls.h>
> @@ -197,6 +198,9 @@ __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
> if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
> exit_to_usermode_loop(regs, cached_flags);
>
> + if (unlikely(test_and_clear_thread_flag(TIF_LOAD_FPU)))
> + switch_fpu_return();
> +
How about:
if (unlikely(...)) {
exit_to_usermode_loop(regs, cached_flags);
cached_flags = READ_ONCE(ti->flags);
}
if (ti->flags & _TIF_LOAD_FPU) {
clear_thread_flag(TIF_LOAD_FPU);
switch_fpu_return();
}
or something along those lines. The issue is that
test_and_clear_thread_flag is unconditionally atomic, which means that
it will slow down the common fast case by quite a bit.
Alternatively, you could try to jam it into thread_struct::status,
which would let you avoid an atomic operation when you clear it, but
you'd still need an atomic op to set it, so this might not be a win.
> +static inline void switch_fpu_finish(void)
> {
> - bool preload;
> - /*
> - * If the task has used the math, pre-load the FPU on xsave processors
> - * or if the past 5 consecutive context-switches used math.
> - */
> - preload = static_cpu_has(X86_FEATURE_FPU) &&
> - new_fpu->fpstate_active &&
> - (use_eager_fpu() || new_fpu->counter > 5);
> -
> - if (preload) {
> - prefetch(&new_fpu->state);
> - new_fpu->counter++;
> - __fpregs_activate(new_fpu);
> - trace_x86_fpu_regs_activated(new_fpu);
> -
> - /* Don't change CR0.TS if we just switch! */
> - if (!__this_cpu_read(fpu_active)) {
> - __fpregs_activate_hw();
> - __this_cpu_write(fpu_active, true);
> - }
> -
> - copy_kernel_to_fpregs(&new_fpu->state);
> - } else if (__this_cpu_read(fpu_active)) {
> - __this_cpu_write(fpu_active, false);
> - __fpregs_deactivate_hw();
> - }
> + set_thread_flag(TIF_LOAD_FPU);
> }
I can imagine this causing problems with kernel code that accesses
current's FPU state, e.g. get_xsave_field_ptr(). I wonder if it would
make sense to make your changes deeper into the FPU core innards so
that, for example, we'd have explicit functions that cause the
in-memory state for current to be up-to-date and readable, to cause
the in-memory state to be up-to-date and writable (which is the same
thing + TIF_FPU_LOAD + whatever other bookkeeping), and causing the
in-CPU state to be up-to-date (possibly readable and writable).
TIF_LOAD_FPU would trigger the latter.
I've often found it confusing that fpu__save by itself has somewhat
ill-defined effects.
> +/*
> + * Set up the userspace FPU context before returning to userspace.
> + */
> +void switch_fpu_return(void)
> +{
> + struct fpu *fpu = ¤t->thread.fpu;
> + bool preload;
> + /*
> + * If the task has used the math, pre-load the FPU on xsave processors
> + * or if the past 5 consecutive context-switches used math.
> + */
> + preload = static_cpu_has(X86_FEATURE_FPU) &&
> + fpu->fpstate_active &&
> + (use_eager_fpu() || fpu->counter > 5);
> +
> + if (preload) {
> + prefetch(&fpu->state);
> + fpu->counter++;
> + __fpregs_activate(fpu);
> + trace_x86_fpu_regs_activated(fpu);
> +
> + /* Don't change CR0.TS if we just switch! */
> + if (!__this_cpu_read(fpu_active)) {
> + __fpregs_activate_hw();
> + __this_cpu_write(fpu_active, true);
> + }
We should just finish getting rid of all TS uses.
next prev parent reply other threads:[~2016-10-01 23:44 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-10-01 20:31 [PATCH RFC 0/5] x86,fpu: make FPU context switching much lazier riel
2016-10-01 20:31 ` [PATCH RFC 1/5] x86,fpu: split prev/next task fpu state handling riel
2016-10-01 23:26 ` Andy Lutomirski
2016-10-02 0:02 ` Rik van Riel
2016-10-01 20:31 ` [PATCH RFC 2/5] x86,fpu: delay FPU register loading until switch to userspace riel
2016-10-01 23:44 ` Andy Lutomirski [this message]
2016-10-02 0:08 ` Rik van Riel
2016-10-03 20:54 ` Andy Lutomirski
2016-10-03 21:21 ` Rik van Riel
2016-10-03 21:36 ` Andy Lutomirski
2016-10-04 1:29 ` Rik van Riel
2016-10-04 2:09 ` Andy Lutomirski
2016-10-04 2:47 ` Rik van Riel
2016-10-04 3:02 ` Andy Lutomirski
2016-10-04 6:35 ` Ingo Molnar
2016-10-04 12:48 ` Rik van Riel
2016-10-04 2:11 ` Rik van Riel
2016-10-04 3:02 ` Andy Lutomirski
2016-10-02 0:42 ` Rik van Riel
2016-10-03 16:23 ` Dave Hansen
2016-10-01 20:31 ` [PATCH RFC 3/5] x86,fpu: add kernel fpu argument to __kernel_fpu_begin riel
2016-10-01 20:31 ` [PATCH RFC 4/5] x86,fpu: lazily skip FPU restore when still loaded riel
2016-10-03 20:04 ` Dave Hansen
2016-10-03 20:22 ` Rik van Riel
2016-10-03 20:49 ` Dave Hansen
2016-10-03 21:02 ` Rik van Riel
2016-10-01 20:31 ` [PATCH RFC 5/5] x86,fpu: kinda sorta fix up signal path riel
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CALCETrWnhA4Gd5AxDEJz7d-VQNDN7O8ayyna7fz3=HNtNnARxQ@mail.gmail.com' \
--to=luto@amacapital.net \
--cc=bp@suse.de \
--cc=dave.hansen@linux.intel.com \
--cc=hpa@zytor.com \
--cc=linux-kernel@vger.kernel.org \
--cc=luto@kernel.org \
--cc=mingo@redhat.com \
--cc=pbonzini@redhat.com \
--cc=riel@redhat.com \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).