On Tue, Aug 15, 2017 at 06:31:05PM +0100, Ard Biesheuvel wrote:
> Hi Dave,
>
> On 9 August 2017 at 13:05, Dave Martin <Dave.Martin@xxxxxxx> wrote:
> > This patch adds the core support for switching and managing the SVE
> > architectural state of user tasks.
> >
> > Calls to the existing FPSIMD low-level save/restore functions are
> > factored out as new functions task_fpsimd_{save,load}(), since SVE
> > now dynamically may or may not need to be handled at these points
> > depending on the kernel configuration, hardware features discovered
> > at boot, and the runtime state of the task. To make these
> > decisions as fast as possible, const cpucaps are used where
> > feasible, via the system_supports_sve() helper.
> >
> > The SVE registers are only tracked for threads that have explicitly
> > used SVE, indicated by the new thread flag TIF_SVE. Otherwise, the
> > FPSIMD view of the architectural state is stored in
> > thread.fpsimd_state as usual.
> >
> > When in use, the SVE registers are not stored directly in
> > thread_struct due to their potentially large and variable size.
> > Because the task_struct slab allocator must be configured very
> > early during kernel boot, it is also tricky to configure it
> > correctly to match the maximum vector length provided by the
> > hardware, since this depends on examining secondary CPUs as well as
> > the primary. Instead, a pointer sve_state in thread_struct points
> > to a dynamically allocated buffer containing the SVE register data,
> > and code is added to allocate, duplicate and free this buffer at
> > appropriate times.
> >
> > TIF_SVE is set when taking an SVE access trap from userspace, if
> > suitable hardware support has been detected. This enables SVE for
> > the thread: a subsequent return to userspace will disable the trap
> > accordingly. If such a trap is taken without sufficient hardware
> > support, SIGILL is sent to the thread instead as if an undefined
> > instruction had been executed: this may happen if userspace tries
> > to use SVE in a system where not all CPUs support it for example.
> >
> > The kernel may clear TIF_SVE and disable SVE for the thread
> > whenever an explicit syscall is made by userspace, though this is
> > considered an optimisation opportunity rather than a deterministic
> > guarantee: the kernel may not do this on every syscall, but it is
> > permitted to do so. For backwards compatibility reasons and
> > conformance with the spirit of the base AArch64 procedure call
> > standard, the subset of the SVE register state that aliases the
> > FPSIMD registers is still preserved across a syscall even if this
> > happens.
> >
> > TIF_SVE is also cleared, and SVE disabled, on exec: this is an
> > obvious slow path and a hint that we are running a new binary that
> > may not use SVE.
> >
> > Code is added to sync data between thread.fpsimd_state and
> > thread.sve_state whenever enabling/disabling SVE, in a manner
> > consistent with the SVE architectural programmer's model.
> >
> > Signed-off-by: Dave Martin <Dave.Martin@xxxxxxx>
> > ---
> > arch/arm64/include/asm/fpsimd.h | 19 +++
> > arch/arm64/include/asm/processor.h | 2 +
> > arch/arm64/include/asm/thread_info.h | 1 +
> > arch/arm64/include/asm/traps.h | 2 +
> > arch/arm64/kernel/entry.S | 14 +-
> > arch/arm64/kernel/fpsimd.c | 241 ++++++++++++++++++++++++++++++++++-
> > arch/arm64/kernel/process.c | 6 +-
> > arch/arm64/kernel/traps.c | 4 +-
> > 8 files changed, 279 insertions(+), 10 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/fpsimd.h b/arch/arm64/include/asm/fpsimd.h
> > index 026a7c7..72090a1 100644
> > --- a/arch/arm64/include/asm/fpsimd.h
> > +++ b/arch/arm64/include/asm/fpsimd.h
> > @@ -20,6 +20,8 @@
> >
> > #ifndef __ASSEMBLY__
> >
> > +#include <linux/stddef.h>
> > +
> > /*
> > * FP/SIMD storage area has:
> > * - FPSR and FPCR
> > @@ -72,6 +74,23 @@ extern void sve_load_state(void const *state, u32 const *pfpsr,
> > unsigned long vq_minus_1);
> > extern unsigned int sve_get_vl(void);
> >
> > +#ifdef CONFIG_ARM64_SVE
> > +
> > +extern size_t sve_state_size(struct task_struct const *task);
> > +
> > +extern void sve_alloc(struct task_struct *task);
> > +extern void fpsimd_release_thread(struct task_struct *task);
> > +extern void fpsimd_dup_sve(struct task_struct *dst,
> > + struct task_struct const *src);
> > +
> > +#else /* ! CONFIG_ARM64_SVE */
> > +
> > +static void __maybe_unused sve_alloc(struct task_struct *task) { }
> > +static void __maybe_unused fpsimd_release_thread(struct task_struct *task) { }
> > +static void __maybe_unused fpsimd_dup_sve(struct task_struct *dst,
> > + struct task_struct const *src) { }
> > +#endif /* ! CONFIG_ARM64_SVE */
> > +
> > /* For use by EFI runtime services calls only */
> > extern void __efi_fpsimd_begin(void);
> > extern void __efi_fpsimd_end(void);
> > diff --git a/arch/arm64/include/asm/processor.h b/arch/arm64/include/asm/processor.h
> > index b7334f1..969feed 100644
> > --- a/arch/arm64/include/asm/processor.h
> > +++ b/arch/arm64/include/asm/processor.h
> > @@ -85,6 +85,8 @@ struct thread_struct {
> > unsigned long tp2_value;
> > #endif
> > struct fpsimd_state fpsimd_state;
> > + void *sve_state; /* SVE registers, if any */
> > + u16 sve_vl; /* SVE vector length */
> > unsigned long fault_address; /* fault info */
> > unsigned long fault_code; /* ESR_EL1 value */
> > struct debug_info debug; /* debugging */
> > diff --git a/arch/arm64/include/asm/thread_info.h b/arch/arm64/include/asm/thread_info.h
> > index 46c3b93..1a4b30b 100644
> > --- a/arch/arm64/include/asm/thread_info.h
> > +++ b/arch/arm64/include/asm/thread_info.h
> > @@ -96,6 +96,7 @@ struct thread_info {
> > #define TIF_RESTORE_SIGMASK 20
> > #define TIF_SINGLESTEP 21
> > #define TIF_32BIT 22 /* 32bit process */
> > +#define TIF_SVE 23 /* Scalable Vector Extension in use */
> >
> > #define _TIF_SIGPENDING (1 << TIF_SIGPENDING)
> > #define _TIF_NEED_RESCHED (1 << TIF_NEED_RESCHED)
> > diff --git a/arch/arm64/include/asm/traps.h b/arch/arm64/include/asm/traps.h
> > index 02e9035..f058c07 100644
> > --- a/arch/arm64/include/asm/traps.h
> > +++ b/arch/arm64/include/asm/traps.h
> > @@ -34,6 +34,8 @@ struct undef_hook {
> >
> > void register_undef_hook(struct undef_hook *hook);
> > void unregister_undef_hook(struct undef_hook *hook);
> > +void force_signal_inject(int signal, int code, struct pt_regs *regs,
> > + unsigned long address);
> >
> > void arm64_notify_segfault(struct pt_regs *regs, unsigned long addr);
> >
> > diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
> > index cace76d..c33069c 100644
> > --- a/arch/arm64/kernel/entry.S
> > +++ b/arch/arm64/kernel/entry.S
> > @@ -524,6 +524,8 @@ el0_sync:
> > b.eq el0_ia
> > cmp x24, #ESR_ELx_EC_FP_ASIMD // FP/ASIMD access
> > b.eq el0_fpsimd_acc
> > + cmp x24, #ESR_ELx_EC_SVE // SVE access
> > + b.eq el0_sve_acc
> > cmp x24, #ESR_ELx_EC_FP_EXC64 // FP/ASIMD exception
> > b.eq el0_fpsimd_exc
> > cmp x24, #ESR_ELx_EC_SYS64 // configurable trap
> > @@ -622,9 +624,19 @@ el0_fpsimd_acc:
> > mov x1, sp
> > bl do_fpsimd_acc
> > b ret_to_user
> > +el0_sve_acc:
> > + /*
> > + * Scalable Vector Extension access
> > + */
> > + enable_dbg
> > + ct_user_exit
> > + mov x0, x25
> > + mov x1, sp
> > + bl do_sve_acc
> > + b ret_to_user
> > el0_fpsimd_exc:
> > /*
> > - * Floating Point or Advanced SIMD exception
> > + * Floating Point, Advanced SIMD or SVE exception
> > */
> > enable_dbg
> > ct_user_exit
> > diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
> > index 9c1f268e..37dd1b2 100644
> > --- a/arch/arm64/kernel/fpsimd.c
> > +++ b/arch/arm64/kernel/fpsimd.c
> > @@ -24,12 +24,17 @@
> > #include <linux/init.h>
> > #include <linux/percpu.h>
> > #include <linux/preempt.h>
> > +#include <linux/ptrace.h>
> > #include <linux/sched/signal.h>
> > #include <linux/signal.h>
> > +#include <linux/slab.h>
> >
> > #include <asm/fpsimd.h>
> > #include <asm/cputype.h>
> > #include <asm/simd.h>
> > +#include <asm/sigcontext.h>
> > +#include <asm/sysreg.h>
> > +#include <asm/traps.h>
> >
> > #define FPEXC_IOF (1 << 0)
> > #define FPEXC_DZF (1 << 1)
> > @@ -38,6 +43,10 @@
> > #define FPEXC_IXF (1 << 4)
> > #define FPEXC_IDF (1 << 7)
> >
> > +/* Forward declarations for local functions used by both SVE and FPSIMD */
> > +static void task_fpsimd_load(void);
> > +static void task_fpsimd_save(void);
> > +
>
> We usually try to avoid forward declarations for functions with static
> linkage. Is it possible to reorder them and get rid of this?
So do I :) There was some sort of historical reason for this, to do
with some forward and backward dependencies and trying to corral
affected code into a single #ifdef rather than having multiple #ifdefs.
These two declarations were the most stubborn, and I gave up on them.
Now that the ifdeffery is mostly gone it looks like these _can_ probably
be got rid of, so I'll try to do that.
>
> > /*
> > * In order to reduce the number of times the FPSIMD state is needlessly saved
> > * and restored, we need to keep track of two things:
> > @@ -99,6 +108,160 @@
> > */
> > static DEFINE_PER_CPU(struct fpsimd_state *, fpsimd_last_state);
> >
> > +static void sve_free(struct task_struct *task)
> > +{
> > + kfree(task->thread.sve_state);
> > + task->thread.sve_state = NULL;
> > +}
> > +
> > +/* Offset of FFR in the SVE register dump */
> > +static size_t sve_ffr_offset(int vl)
> > +{
> > + BUG_ON(!sve_vl_valid(vl));
>
> BUG_ON() is a pretty heavy hammer, so we should not use it unless the
> kernel state is so corrupted that there is no way to carry on. I have
> a feeling this may not be the case for some of the occurrences in this
> patch.
These were useful during development, but you're right, some of these
are probably superfluous.
Bugs aside, there should be no way for an invalid value to get into
thread.sve_vl, sve_max_vl etc., with the one exception that the init
task and kernel tasks forked from it have thread.sve_vl = 0 (which
doesn't matter because SVE state is never loaded, saved or otherwise
manipulated for these tasks due to never entering userspace). (The
logic that attempts to ensure that these values are only set validly is
in patches 14 (thread.sve_vl), 15 (sve_max_vl), and 12 and 20
(sve_default_vl).)
I think if we try to enter userspace with an invalid vl that may be
worth a BUG(), but elsewhere we can probably just use SVE_MIN_VL
instead of "invalid" values.
In particular, poisoning sve_max_vl and sve_default_vl with -1 is
probably not needed now the development phase is over. These could be
initially SVE_MIN_VL without ill effects.
I'll take a look at this.
>
> > + return SVE_SIG_FFR_OFFSET(sve_vq_from_vl(vl)) - SVE_SIG_REGS_OFFSET;
> > +}
> > +
> > +static void *sve_pffr(struct task_struct *task)
> > +{
> > + unsigned int vl = task->thread.sve_vl;
> > +
> > + BUG_ON(!sve_vl_valid(vl) || !task->thread.sve_state);
> > + return (char *)task->thread.sve_state + sve_ffr_offset(vl);
> > +}
> > +
> > +static u64 sve_cpacr_trap_on(u64 cpacr)
> > +{
> > + return cpacr & ~(u64)CPACR_EL1_ZEN_EL0EN;
> > +}
> > +
> > +static u64 sve_cpacr_trap_off(u64 cpacr)
> > +{
> > + return cpacr | CPACR_EL1_ZEN_EL0EN;
> > +}
> > +
> > +static void change_cpacr(u64 old, u64 new)
> > +{
> > + if (old != new)
> > + write_sysreg(new, CPACR_EL1);
> > +}
> > +
> > +#ifdef CONFIG_ARM64_SVE
> > +
> > +#define ZREG(sve_state, vq, n) ((char *)(sve_state) + \
> > + (SVE_SIG_ZREG_OFFSET(vq, n) - SVE_SIG_REGS_OFFSET))
> > +
> > +static void sve_to_fpsimd(struct task_struct *task)
> > +{
> > + unsigned int vl = task->thread.sve_vl;
> > + unsigned int vq;
> > + void const *sst = task->thread.sve_state;
> > + struct fpsimd_state *fst = &task->thread.fpsimd_state;
> > + unsigned int i;
> > +
> > + if (!system_supports_sve())
> > + return;
> > +
> > + BUG_ON(!sve_vl_valid(vl));
> > + vq = sve_vq_from_vl(vl);
> > +
> > + for (i = 0; i < 32; ++i)
> > + memcpy(&fst->vregs[i], ZREG(sst, vq, i),
> > + sizeof(fst->vregs[i]));
> > +}
> > +
> > +static void fpsimd_to_sve(struct task_struct *task)
> > +{
> > + unsigned int vl = task->thread.sve_vl;
> > + unsigned int vq;
> > + void *sst = task->thread.sve_state;
> > + struct fpsimd_state const *fst = &task->thread.fpsimd_state;
> > + unsigned int i;
> > +
> > + if (!system_supports_sve())
> > + return;
> > +
> > + BUG_ON(!sve_vl_valid(vl));
> > + vq = sve_vq_from_vl(vl);
> > +
> > + for (i = 0; i < 32; ++i)
> > + memcpy(ZREG(sst, vq, i), &fst->vregs[i],
> > + sizeof(fst->vregs[i]));
> > +}
> > +
> > +size_t sve_state_size(struct task_struct const *task)
> > +{
> > + unsigned int vl = task->thread.sve_vl;
> > +
> > + BUG_ON(!sve_vl_valid(vl));
> > + return SVE_SIG_REGS_SIZE(sve_vq_from_vl(vl));
> > +}
> > +
> > +void sve_alloc(struct task_struct *task)
> > +{
> > + if (task->thread.sve_state) {
> > + memset(task->thread.sve_state, 0, sve_state_size(current));
> > + return;
> > + }
> > +
> > + /* This is a small allocation (maximum ~8KB) and Should Not Fail. */
> > + task->thread.sve_state =
> > + kzalloc(sve_state_size(task), GFP_KERNEL);
> > +
> > + /*
> > + * If future SVE revisions can have larger vectors though,
> > + * this may cease to be true:
> > + */
> > + BUG_ON(!task->thread.sve_state);
> > +}
> > +
> > +/*
> > + * Handle SVE state across fork():
> > + *
> > + * dst and src must not end up with aliases of the same sve_state.
> > + * Because a task cannot fork except in a syscall, we can discard SVE
> > + * state for dst here: reallocation will be deferred until dst tries
> > + * to use SVE.
> > + */
> > +void fpsimd_dup_sve(struct task_struct *dst, struct task_struct const *src)
> > +{
> > + BUG_ON(!in_syscall(task_pt_regs(dst)));
> > +
> > + if (test_and_clear_tsk_thread_flag(dst, TIF_SVE)) {
> > + sve_to_fpsimd(dst);
> > + dst->thread.sve_state = NULL;
Looking at this in reponse to your comment on arch_dup_task_struct(), I
think there's a bug here: if the forking task has TIF_SVE clear, then a
non-NULL sve_state pointer will remain cloned from the parent task.
The assignment of NULL needs to move outside the if.
This bug would pop up with programs that use SVE and then go
multithreaded, or programs that use SVE and then fork+exec. In the
former case we'd end up with aliased state; in the latter, the state
would get freed by the exec while the parent task still owns it.
> > + }
> > +}
> > +
> > +void fpsimd_release_thread(struct task_struct *dead_task)
> > +{
> > + sve_free(dead_task);
> > +}
> > +
> > +#endif /* CONFIG_ARM64_SVE */
> > +
> > +/*
> > + * Trapped SVE access
> > + */
> > +void do_sve_acc(unsigned int esr, struct pt_regs *regs)
> > +{
> > + BUG_ON(is_compat_task());
> > +
> > + /* Even if we chose not to use SVE, the hardware could still trap: */
> > + if (unlikely(!system_supports_sve())) {
> > + force_signal_inject(SIGILL, ILL_ILLOPC, regs, 0);
> > + return;
> > + }
> > +
> > + task_fpsimd_save();
> > +
> > + sve_alloc(current);
> > + fpsimd_to_sve(current);
> > + if (test_and_set_thread_flag(TIF_SVE))
> > + WARN_ON(1); /* SVE access shouldn't have trapped */
> > +
> > + task_fpsimd_load();
> > +}
> > +
> > /*
> > * Trapped FP/ASIMD access.
> > */
> > @@ -135,6 +298,55 @@ void do_fpsimd_exc(unsigned int esr, struct pt_regs *regs)
> > send_sig_info(SIGFPE, &info, current);
> > }
> >
> > +static void task_fpsimd_load(void)
> > +{
> > + if (system_supports_sve() && test_thread_flag(TIF_SVE)) {
> > + unsigned int vl = current->thread.sve_vl;
> > +
> > + BUG_ON(!sve_vl_valid(vl));
> > + sve_load_state(sve_pffr(current),
> > + ¤t->thread.fpsimd_state.fpsr,
> > + sve_vq_from_vl(vl) - 1);
> > + } else
> > + fpsimd_load_state(¤t->thread.fpsimd_state);
> > +
>
> Please use braces consistently on all branches of an if ()
Arg, will fix.
> > + if (system_supports_sve()) {
> > + u64 cpacr = read_sysreg(CPACR_EL1);
> > + u64 new_cpacr;
> > +
> > + /* Toggle SVE trapping for userspace if needed */
> > + if (test_thread_flag(TIF_SVE))
> > + new_cpacr = sve_cpacr_trap_off(cpacr);
> > + else
> > + new_cpacr = sve_cpacr_trap_on(cpacr);
> > +
> > + change_cpacr(cpacr, new_cpacr);
>
> I understand you want to avoid setting CPACR to the same value, but
> this does look a bit clunky IMO. Wouldn't it be much better to have a
> generic accessor with a mask and a value that encapsulates this?
Hmmm, this was a replacement for some more obscure logic that was there
previously and that Mark Rutland suggested cleaning up.
Because of the way this is used, I think it would be reasonable to have
static void change_cpacr(u64 new_bits, u64 mask)
which elides the cpacr write if no change would result. Then the _off()
and _on() could be folded into the call sites as 0 and
CPACR_EL1_ZEN_EL0EN respectively.
But since this function would be used for nothing other than toggling
CPACR_EL1_ZEN_EL0EN this looks pointlessly general.
Alternatively we could have two functions
static void sve_el0_trap_on(void);
static void sve_el0_trap_off(void);
which contain the CPACR manipulation so the caller doesn't have to spell
it out.
Any thoughts?
>
> > + /* Serialised by exception return to user */
> > + }
> > +}
> > +
> > +static void task_fpsimd_save(void)
> > +{
> > + if (system_supports_sve() &&
> > + in_syscall(current_pt_regs()) &&
> > + test_thread_flag(TIF_SVE)) {
> > + u64 cpacr = read_sysreg(CPACR_EL1);
> > +
> > + clear_thread_flag(TIF_SVE);
> > +
> > + /* Trap if the task tries to use SVE again: */
> > + change_cpacr(cpacr, sve_cpacr_trap_on(cpacr));
> > + }
> > +
> > + if (!test_thread_flag(TIF_FOREIGN_FPSTATE)) {
> > + if (system_supports_sve() && test_thread_flag(TIF_SVE))
> > + sve_save_state(sve_pffr(current),
> > + ¤t->thread.fpsimd_state.fpsr);
> > + else
> > + fpsimd_save_state(¤t->thread.fpsimd_state);
> > + }
> > +}
> > +
> > void fpsimd_thread_switch(struct task_struct *next)
> > {
> > if (!system_supports_fpsimd())
> > @@ -144,8 +356,8 @@ void fpsimd_thread_switch(struct task_struct *next)
> > * the registers is in fact the most recent userland FPSIMD state of
> > * 'current'.
> > */
> > - if (current->mm && !test_thread_flag(TIF_FOREIGN_FPSTATE))
> > - fpsimd_save_state(¤t->thread.fpsimd_state);
> > + if (current->mm)
> > + task_fpsimd_save();
> >
> > if (next->mm) {
> > /*
> > @@ -172,8 +384,25 @@ void fpsimd_flush_thread(void)
> >
> > local_bh_disable();
> >
> > - memset(¤t->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
> > fpsimd_flush_task_state(current);
> > +
> > + memset(¤t->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
> > +
>
> Any reason in particular this needs to be reordered?
This seems to have been left behind from some overlapping fixes.
The recent commit cb84d11e ("arm64: neon: Remove support for nested or
hardirq kernel-mode NEON") adds the local_bh_disable(), which prevents
this code from racing with fpsimd_thread_switch().
Previously there was a bug, which commit 674c242c ("arm64: flush
FP/SIMD state correctly after execve()") tries to fix by adding the
fpsimd_flush_task_state(). However, I think this is still buggy:
context switch can occur during the memset and replace the zeros with
data from the registers again (which is what the patch was intended to
prevent). It's important in this case that the
fpsimd_flush_task_state() is first, so that a racing context switch
doesn't try to save anything.
I think I reordered these calls earlier in development while debugging
some race problems, but this is not really an SVE fix. The kernel-mode
NEON simplifications make this irrelevant for mainline since preemption
is disabled anyway, but a separate fix is probably needed for stable,
either reordering those calls or adding preempt_disable().
For this patch, I think the reordering can be dropped.
Does this logic sound correct to you?
[...]
> > int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
> > {
> > if (current->mm)
> > fpsimd_preserve_current_state();
> > - *dst = *src;
> > + memcpy(dst, src, arch_task_struct_size);
Looking at this, we can go back to the assignment here. The memcpy()
is from the days when I reserved extra space off the end of task_struct
to store the SVE state, and hadn't thought about whether the SVE state
really needed to be copied across fork.
> > +
> > + fpsimd_dup_sve(dst, src);
> > +
>
> Is this used for anything except fork()? If not, do we really need to
No, I don't think so.
> duplicate the SVE state?
No, but fpsimd_dup_sve() is perhaps unfortunately named because it
doesn't do that. It dups the fpsimd subset of the state (required by
the pre-existing syscall ABI) which lives in sve_state if TIF_SVE is
currently set. It should also nuke the sve_state pointer in dst, but
this currently looks broken (see my comments on fpsimd_dup_sve above).
Any thoughts from your side?
[...]
Thanks for the review
---Dave
