Messages in this thread Patch in this message | | | Date | Tue, 13 Jan 2015 14:32:26 -0500 | From | Rik van Riel <> | Subject | Re: [RFC PATCH 04/11] x86,fpu: defer FPU restore until return to userspace |
| |
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 01/13/2015 12:58 PM, Oleg Nesterov wrote: > On 01/11, riel@redhat.com wrote: >> >> Defer restoring the FPU state, if so desired, until the task >> returns to userspace. > > And yet another concern ;) Although I feel that I am totally > confused and probably wrong. > >> --- a/arch/x86/include/asm/fpu-internal.h +++ >> b/arch/x86/include/asm/fpu-internal.h @@ -382,6 +382,7 @@ static >> inline void drop_init_fpu(struct task_struct *tsk) else >> fxrstor_checking(&init_xstate_buf->i387); } + >> clear_thread_flag(TIF_LOAD_FPU); } > > OK, but shouldn't (say) restore_user_xstate() clear TIF_LOAD_FPU > too? Otherwise, can't switch_fpu_finish() restore the wrong context > later?
I address restore_user_xstate and save_user_xstate later in the patch series. Once I have all the code working, I should refactor that, in order to get a cleanly bisectable series.
> Or. Perhaps switch_fpu_finish() should do nothing if fpu.has_fpu == > T, I dunno.
One of the later patches in the series implements that.
- -- All rights reversed -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iQEcBAEBAgAGBQJUtXLKAAoJEM553pKExN6DCksH/iSAOJ4rqvDsBBY+VG5ZtQEK 7fo5bK1HQb0X8yeJPoR8rSg3PzqHjV1v+eU4d/OWEJHV7U9zxWedn+BRv8DrVZvo 1f1QqHY9ORZD0/NVZRJxlNBBqvUijUE1POTFixZVhU/t0MuWFG0+4a7BgY5B6RIa /hZ4XcZ9fnTpN3cOSIEIOHCT90xZuUMwopRX/7ad8QRhqLaoH0MXqspAEMi8321b 4xwQUw1wD6OtCqWk1fG5dfF69F8ZrSGgNilUudwMeBEg4NODbVis6jw6iOZnjM+3 juZZXZ49LdYTlw5erjPVoZJ3kXZbdT1cIa81+LPDot/uKyAx5j6IKmM4ENQusw0= =OHgf -----END PGP SIGNATURE-----
| |