| Message ID | 1490194274-30569-25-git-send-email-Dave.Martin@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Mar 22, 2017 at 02:50:54PM +0000, Dave Martin wrote: > The base procedure call standard for the Scalable Vector Extension > defines all of the SVE programmer's model state (Z0-31, P0-15, FFR) > as caller-save, except for that subset of the state that aliases > FPSIMD state. > > System calls from userspace will almost always be made through C > library wrappers -- as a consequence of the PCS there will thus > rarely if ever be any live SVE state at syscall entry in practice. > > This gives us an opportinity to make SVE explicitly caller-save > around SVC and so stop carrying around the SVE state for tasks that > use SVE only occasionally (say, by calling a library). > > Note that FPSIMD state will still be preserved around SVC. > > As a crude heuristic to avoid pathological cases where a thread > that uses SVE frequently has to fault back into the kernel again to > re-enable SVE after a syscall, we switch the thread back to > FPSIMD-only context tracking only if the context is actually > switched out before returning to userspace. > > Signed-off-by: Dave Martin <Dave.Martin@arm.com> > --- > arch/arm64/kernel/fpsimd.c | 17 +++++++++++++++++ > 1 file changed, 17 insertions(+) > > diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c > index 5fb5585..8c18384 100644 > --- a/arch/arm64/kernel/fpsimd.c > +++ b/arch/arm64/kernel/fpsimd.c > @@ -250,6 +250,23 @@ static void task_fpsimd_save(struct task_struct *task) > BUG_ON(task != current); > > if (IS_ENABLED(CONFIG_ARM64_SVE) && > + task_pt_regs(task)->syscallno != ~0UL && > + test_tsk_thread_flag(task, TIF_SVE)) { > + unsigned long tmp; > + > + clear_tsk_thread_flag(task, TIF_SVE); > + > + /* Trap if the task tries to use SVE again: */ > + asm volatile ( > + "mrs %[tmp], cpacr_el1\n\t" > + "bic %[tmp], %[tmp], %[mask]\n\t" > + "msr cpacr_el1, %[tmp]" > + : [tmp] "=r" (tmp) > + : [mask] "i" (CPACR_EL1_ZEN_EL0EN) > + ); Given we're poking this bit in a few places, I think it would make more sense to add enable/disable helpers. Those can also subsume the lazy writeback used for the context switch, e.g. static inline void sve_el0_enable(void) } unsigned long cpacr = read_sysreg(cpacr_el1); if ((cpacr & CPACR_EL1_ZEN_EL0EN) return; cpacr |= CPACR_EL1_ZEN_EL0EN; write_sysreg(cpacr, cpacr_el1); } static inline void sve_el0_disable(void) { unsigned long cpacr = read_sysreg(cpacr_el1); if (!(cpacr & CPACR_EL1_ZEN_EL0EN) return; cpacr &= ~CPACR_EL1_ZEN_EL0EN; write_sysreg(cpacr, cpacr_el1); } Thanks, Mark.
On Wed, Mar 22, 2017 at 05:03:30PM +0000, Mark Rutland wrote: > On Wed, Mar 22, 2017 at 02:50:54PM +0000, Dave Martin wrote: > > The base procedure call standard for the Scalable Vector Extension > > defines all of the SVE programmer's model state (Z0-31, P0-15, FFR) > > as caller-save, except for that subset of the state that aliases > > FPSIMD state. > > > > System calls from userspace will almost always be made through C > > library wrappers -- as a consequence of the PCS there will thus > > rarely if ever be any live SVE state at syscall entry in practice. > > > > This gives us an opportinity to make SVE explicitly caller-save > > around SVC and so stop carrying around the SVE state for tasks that > > use SVE only occasionally (say, by calling a library). > > > > Note that FPSIMD state will still be preserved around SVC. > > > > As a crude heuristic to avoid pathological cases where a thread > > that uses SVE frequently has to fault back into the kernel again to > > re-enable SVE after a syscall, we switch the thread back to > > FPSIMD-only context tracking only if the context is actually > > switched out before returning to userspace. > > > > Signed-off-by: Dave Martin <Dave.Martin@arm.com> > > --- > > arch/arm64/kernel/fpsimd.c | 17 +++++++++++++++++ > > 1 file changed, 17 insertions(+) > > > > diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c > > index 5fb5585..8c18384 100644 > > --- a/arch/arm64/kernel/fpsimd.c > > +++ b/arch/arm64/kernel/fpsimd.c > > @@ -250,6 +250,23 @@ static void task_fpsimd_save(struct task_struct *task) > > BUG_ON(task != current); > > > > if (IS_ENABLED(CONFIG_ARM64_SVE) && > > + task_pt_regs(task)->syscallno != ~0UL && ^ I may add a helper for this occult check (since it duplicated elsewhere), but... > > + test_tsk_thread_flag(task, TIF_SVE)) { > > + unsigned long tmp; > > + > > + clear_tsk_thread_flag(task, TIF_SVE); > > + > > + /* Trap if the task tries to use SVE again: */ > > + asm volatile ( > > + "mrs %[tmp], cpacr_el1\n\t" > > + "bic %[tmp], %[tmp], %[mask]\n\t" > > + "msr cpacr_el1, %[tmp]" > > + : [tmp] "=r" (tmp) > > + : [mask] "i" (CPACR_EL1_ZEN_EL0EN) > > + ); > > Given we're poking this bit in a few places, I think it would make more > sense to add enable/disable helpers. Those can also subsume the lazy > writeback used for the context switch, e.g. > > static inline void sve_el0_enable(void) > } > unsigned long cpacr = read_sysreg(cpacr_el1); > if ((cpacr & CPACR_EL1_ZEN_EL0EN) > return; > > cpacr |= CPACR_EL1_ZEN_EL0EN; > write_sysreg(cpacr, cpacr_el1); > } > > static inline void sve_el0_disable(void) > { > unsigned long cpacr = read_sysreg(cpacr_el1); > if (!(cpacr & CPACR_EL1_ZEN_EL0EN) > return; > > cpacr &= ~CPACR_EL1_ZEN_EL0EN; > write_sysreg(cpacr, cpacr_el1); > } ... although this makes sense and I did consider it, instead I ended up consolidating this in one place in task_fpsimd_load(). ret_to_user is the only time we care about what EL0EN is set to, and task_fpsimd_load is only used on that path. With it in one place, abstraction seemed less necessary -- but I'll still clean up the asm hacks as you suggest. Cheers ---Dave
diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c index 5fb5585..8c18384 100644 --- a/arch/arm64/kernel/fpsimd.c +++ b/arch/arm64/kernel/fpsimd.c @@ -250,6 +250,23 @@ static void task_fpsimd_save(struct task_struct *task) BUG_ON(task != current); if (IS_ENABLED(CONFIG_ARM64_SVE) && + task_pt_regs(task)->syscallno != ~0UL && + test_tsk_thread_flag(task, TIF_SVE)) { + unsigned long tmp; + + clear_tsk_thread_flag(task, TIF_SVE); + + /* Trap if the task tries to use SVE again: */ + asm volatile ( + "mrs %[tmp], cpacr_el1\n\t" + "bic %[tmp], %[tmp], %[mask]\n\t" + "msr cpacr_el1, %[tmp]" + : [tmp] "=r" (tmp) + : [mask] "i" (CPACR_EL1_ZEN_EL0EN) + ); + } + + if (IS_ENABLED(CONFIG_ARM64_SVE) && test_tsk_thread_flag(task, TIF_SVE)) sve_save_state(sve_pffr(task), &task->thread.fpsimd_state.fpsr);
The base procedure call standard for the Scalable Vector Extension defines all of the SVE programmer's model state (Z0-31, P0-15, FFR) as caller-save, except for that subset of the state that aliases FPSIMD state. System calls from userspace will almost always be made through C library wrappers -- as a consequence of the PCS there will thus rarely if ever be any live SVE state at syscall entry in practice. This gives us an opportinity to make SVE explicitly caller-save around SVC and so stop carrying around the SVE state for tasks that use SVE only occasionally (say, by calling a library). Note that FPSIMD state will still be preserved around SVC. As a crude heuristic to avoid pathological cases where a thread that uses SVE frequently has to fault back into the kernel again to re-enable SVE after a syscall, we switch the thread back to FPSIMD-only context tracking only if the context is actually switched out before returning to userspace. Signed-off-by: Dave Martin <Dave.Martin@arm.com> --- arch/arm64/kernel/fpsimd.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)