Re: [PATCH 7/7] riscv: Add qspinlock support based on Zabha extension

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, May 31, 2024 at 2:22 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
>
> On Fri, May 31, 2024 at 3:57 AM Guo Ren <guoren@xxxxxxxxxx> wrote:
> >
> > On Thu, May 30, 2024 at 1:30 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > >
> > > Hi Guo,
> > >
> > > On Thu, May 30, 2024 at 3:55 AM Guo Ren <guoren@xxxxxxxxxx> wrote:
> > > >
> > > > On Wed, May 29, 2024 at 9:03 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > > >
> > > > > Hi Guo,
> > > > >
> > > > > On Wed, May 29, 2024 at 11:24 AM Guo Ren <guoren@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Tue, May 28, 2024 at 11:18 PM Alexandre Ghiti <alexghiti@xxxxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > In order to produce a generic kernel, a user can select
> > > > > > > CONFIG_QUEUED_SPINLOCKS which will fallback at runtime to the ticket
> > > > > > > spinlock implementation if Zabha is not present.
> > > > > > >
> > > > > > > Note that we can't use alternatives here because the discovery of
> > > > > > > extensions is done too late and we need to start with the qspinlock
> > > > > > > implementation because the ticket spinlock implementation would pollute
> > > > > > > the spinlock value, so let's use static keys.
> > > > > > >
> > > > > > > This is largely based on Guo's work and Leonardo reviews at [1].
> > > > > > >
> > > > > > > Link: https://lore.kernel.org/linux-riscv/20231225125847.2778638-1-guoren@xxxxxxxxxx/ [1]
> > > > > > > Signed-off-by: Alexandre Ghiti <alexghiti@xxxxxxxxxxxx>
> > > > > > > ---
> > > > > > >  .../locking/queued-spinlocks/arch-support.txt |  2 +-
> > > > > > >  arch/riscv/Kconfig                            |  1 +
> > > > > > >  arch/riscv/include/asm/Kbuild                 |  4 +-
> > > > > > >  arch/riscv/include/asm/spinlock.h             | 39 +++++++++++++++++++
> > > > > > >  arch/riscv/kernel/setup.c                     | 18 +++++++++
> > > > > > >  include/asm-generic/qspinlock.h               |  2 +
> > > > > > >  include/asm-generic/ticket_spinlock.h         |  2 +
> > > > > > >  7 files changed, 66 insertions(+), 2 deletions(-)
> > > > > > >  create mode 100644 arch/riscv/include/asm/spinlock.h
> > > > > > >
> > > > > > > diff --git a/Documentation/features/locking/queued-spinlocks/arch-support.txt b/Documentation/features/locking/queued-spinlocks/arch-support.txt
> > > > > > > index 22f2990392ff..cf26042480e2 100644
> > > > > > > --- a/Documentation/features/locking/queued-spinlocks/arch-support.txt
> > > > > > > +++ b/Documentation/features/locking/queued-spinlocks/arch-support.txt
> > > > > > > @@ -20,7 +20,7 @@
> > > > > > >      |    openrisc: |  ok  |
> > > > > > >      |      parisc: | TODO |
> > > > > > >      |     powerpc: |  ok  |
> > > > > > > -    |       riscv: | TODO |
> > > > > > > +    |       riscv: |  ok  |
> > > > > > >      |        s390: | TODO |
> > > > > > >      |          sh: | TODO |
> > > > > > >      |       sparc: |  ok  |
> > > > > > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > > > > > > index 184a9edb04e0..ccf1703edeb9 100644
> > > > > > > --- a/arch/riscv/Kconfig
> > > > > > > +++ b/arch/riscv/Kconfig
> > > > > > > @@ -59,6 +59,7 @@ config RISCV
> > > > > > >         select ARCH_SUPPORTS_SHADOW_CALL_STACK if HAVE_SHADOW_CALL_STACK
> > > > > > >         select ARCH_USE_MEMTEST
> > > > > > >         select ARCH_USE_QUEUED_RWLOCKS
> > > > > > > +       select ARCH_USE_QUEUED_SPINLOCKS if TOOLCHAIN_HAS_ZABHA
> > > > > > Using qspinlock or not depends on real hardware capabilities, not the
> > > > > > compiler flag. That's why I introduced combo-spinlock, ticket-spinlock
> > > > > > & qspinlock three Kconfigs, and the combo-spinlock would compat all
> > > > > > hardware platforms but waste some qspinlock code size.
> > > > >
> > > > > You're right, and I think your comment matches what Conor mentioned
> > > > > about the lack of clarity with some extensions: TOOLCHAIN_HAS_ZABHA
> > > > > will allow a platform with Zabha capability to use qspinlocks. But if
> > > > > the hardware does not, it will fallback to the ticket spinlocks.
> > > > >
> > > > > But I agree that looking at the config alone may be misleading, even
> > > > > though it will work as expected at runtime. So I agree with you:
> > > > > unless anyone is strongly against the combo spinlocks, I will do what
> > > > > you suggest and add them.
> > > > The problem with the v12 combo-spinlock is using a static_branch
> > > > instead of the full ALTERNATIVE. Frankly, that's a bad example that
> > > > costs more code space. I found that your cmpxchg32/64 also uses a
> > > > condition branch, which has a similar problem, right?
> > > >
> > > > Anyway, your patch series inspired me to update the v13
> > > > combo-spinlock. My plan is:
> > > > 1. Separate native-qspinlock out of paravirt-qspinlock.
> > > > 2. Re-design an ALTERNATIVE(asm) code instead of static_branch generic
> > > > ticket-lock or qspinlock.
> > >
> > > What's your plan to make use of alternatives here? The alternatives
> > > patching depends on the discovery of the extensions, which is done too
> > > late, at least after the first use of a spinlock (the printk
> > > spinlock). So you'd need to find a way to first use qspinlocks (but
> > > without knowing Zabha is available) and then do the correct patching:
> > I do that in v12:
> > 1. Use qspinlock as init.
> > 2. Change to ticket-lock or not.
> > (Only qspinlock -> ticket-lock, No reverse direction)
> >
> > If there is no contention, Qspinlock is okay for all platforms before
> > smp bringup & no-irq environment.
> >
>
> Yes, by using static keys not alternatives. My question was: how do
> you plan to use alternatives here instead of static keys? To me, it's
> not that simple, hence my suggestions in my previous answer.
Yes, it's not that simple. The current framework doesn't support that
and has two problems:
1. We need to re-implement ticket-lock & qspinlock-fast-path with assembly code.
2. Current alternatives patching only for extensions, but qspinlock is
not a formal extension. Could we accept
__RISCV_ISA_EXT_DATA(xqspinlock, RISCV_ISA_EXT_XQSPINLOCK)?

>
> Thanks,
>
> Alex
>
> > > an idea here could be to add an "init" value to the alternatives and
> > > let the patching process do the right thing when the extensions are
> > > known.
> > >
> > > Another solution would be the early discovery of the extensions, but I
> > > took a look and it's easy with a device tree, but not with ACPI.
> > >
> > > Let me know what you plan to do and how I can help!
> > >
> > > Thanks,
> > >
> > > Alex
> > >
> > > >
> > > > What do you think?
> > > >
> > > >
> > > > >
> > > > > Thanks again for your initial work,
> > > > >
> > > > > Alex
> > > > >
> > > > > >
> > > > > > >         select ARCH_USES_CFI_TRAPS if CFI_CLANG
> > > > > > >         select ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH if SMP && MMU
> > > > > > >         select ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT if MMU
> > > > > > > diff --git a/arch/riscv/include/asm/Kbuild b/arch/riscv/include/asm/Kbuild
> > > > > > > index 504f8b7e72d4..ad72f2bd4cc9 100644
> > > > > > > --- a/arch/riscv/include/asm/Kbuild
> > > > > > > +++ b/arch/riscv/include/asm/Kbuild
> > > > > > > @@ -2,10 +2,12 @@
> > > > > > >  generic-y += early_ioremap.h
> > > > > > >  generic-y += flat.h
> > > > > > >  generic-y += kvm_para.h
> > > > > > > +generic-y += mcs_spinlock.h
> > > > > > >  generic-y += parport.h
> > > > > > > -generic-y += spinlock.h
> > > > > > >  generic-y += spinlock_types.h
> > > > > > > +generic-y += ticket_spinlock.h
> > > > > > >  generic-y += qrwlock.h
> > > > > > >  generic-y += qrwlock_types.h
> > > > > > > +generic-y += qspinlock.h
> > > > > > >  generic-y += user.h
> > > > > > >  generic-y += vmlinux.lds.h
> > > > > > > diff --git a/arch/riscv/include/asm/spinlock.h b/arch/riscv/include/asm/spinlock.h
> > > > > > > new file mode 100644
> > > > > > > index 000000000000..e00429ac20ed
> > > > > > > --- /dev/null
> > > > > > > +++ b/arch/riscv/include/asm/spinlock.h
> > > > > > > @@ -0,0 +1,39 @@
> > > > > > > +/* SPDX-License-Identifier: GPL-2.0 */
> > > > > > > +
> > > > > > > +#ifndef __ASM_RISCV_SPINLOCK_H
> > > > > > > +#define __ASM_RISCV_SPINLOCK_H
> > > > > > > +
> > > > > > > +#ifdef CONFIG_QUEUED_SPINLOCKS
> > > > > > > +#define _Q_PENDING_LOOPS       (1 << 9)
> > > > > > > +
> > > > > > > +#define __no_arch_spinlock_redefine
> > > > > > > +#include <asm/ticket_spinlock.h>
> > > > > > > +#include <asm/qspinlock.h>
> > > > > > > +#include <asm/alternative.h>
> > > > > > > +
> > > > > > > +DECLARE_STATIC_KEY_TRUE(qspinlock_key);
> > > > > > > +
> > > > > > > +#define SPINLOCK_BASE_DECLARE(op, type, type_lock)                     \
> > > > > > > +static __always_inline type arch_spin_##op(type_lock lock)             \
> > > > > > > +{                                                                      \
> > > > > > > +       if (static_branch_unlikely(&qspinlock_key))                     \
> > > > > > > +               return queued_spin_##op(lock);                          \
> > > > > > > +       return ticket_spin_##op(lock);                                  \
> > > > > > > +}
> > > > > > > +
> > > > > > > +SPINLOCK_BASE_DECLARE(lock, void, arch_spinlock_t *)
> > > > > > > +SPINLOCK_BASE_DECLARE(unlock, void, arch_spinlock_t *)
> > > > > > > +SPINLOCK_BASE_DECLARE(is_locked, int, arch_spinlock_t *)
> > > > > > > +SPINLOCK_BASE_DECLARE(is_contended, int, arch_spinlock_t *)
> > > > > > > +SPINLOCK_BASE_DECLARE(trylock, bool, arch_spinlock_t *)
> > > > > > > +SPINLOCK_BASE_DECLARE(value_unlocked, int, arch_spinlock_t)
> > > > > > > +
> > > > > > > +#else
> > > > > > > +
> > > > > > > +#include <asm/ticket_spinlock.h>
> > > > > > > +
> > > > > > > +#endif
> > > > > > > +
> > > > > > > +#include <asm/qrwlock.h>
> > > > > > > +
> > > > > > > +#endif /* __ASM_RISCV_SPINLOCK_H */
> > > > > > > diff --git a/arch/riscv/kernel/setup.c b/arch/riscv/kernel/setup.c
> > > > > > > index 4f73c0ae44b2..31ce75522fd4 100644
> > > > > > > --- a/arch/riscv/kernel/setup.c
> > > > > > > +++ b/arch/riscv/kernel/setup.c
> > > > > > > @@ -244,6 +244,23 @@ static void __init parse_dtb(void)
> > > > > > >  #endif
> > > > > > >  }
> > > > > > >
> > > > > > > +DEFINE_STATIC_KEY_TRUE(qspinlock_key);
> > > > > > > +EXPORT_SYMBOL(qspinlock_key);
> > > > > > > +
> > > > > > > +static void __init riscv_spinlock_init(void)
> > > > > > > +{
> > > > > > > +       asm goto(ALTERNATIVE("nop", "j %[qspinlock]", 0, RISCV_ISA_EXT_ZABHA, 1)
> > > > > > > +                : : : : qspinlock);
> > > > > > > +
> > > > > > > +       static_branch_disable(&qspinlock_key);
> > > > > > > +       pr_info("Ticket spinlock: enabled\n");
> > > > > > > +
> > > > > > > +       return;
> > > > > > > +
> > > > > > > +qspinlock:
> > > > > > > +       pr_info("Queued spinlock: enabled\n");
> > > > > > > +}
> > > > > > > +
> > > > > > >  extern void __init init_rt_signal_env(void);
> > > > > > >
> > > > > > >  void __init setup_arch(char **cmdline_p)
> > > > > > > @@ -295,6 +312,7 @@ void __init setup_arch(char **cmdline_p)
> > > > > > >         riscv_set_dma_cache_alignment();
> > > > > > >
> > > > > > >         riscv_user_isa_enable();
> > > > > > > +       riscv_spinlock_init();
> > > > > > >  }
> > > > > > >
> > > > > > >  bool arch_cpu_is_hotpluggable(int cpu)
> > > > > > > diff --git a/include/asm-generic/qspinlock.h b/include/asm-generic/qspinlock.h
> > > > > > > index 0655aa5b57b2..bf47cca2c375 100644
> > > > > > > --- a/include/asm-generic/qspinlock.h
> > > > > > > +++ b/include/asm-generic/qspinlock.h
> > > > > > > @@ -136,6 +136,7 @@ static __always_inline bool virt_spin_lock(struct qspinlock *lock)
> > > > > > >  }
> > > > > > >  #endif
> > > > > > >
> > > > > > > +#ifndef __no_arch_spinlock_redefine
> > > > > > >  /*
> > > > > > >   * Remapping spinlock architecture specific functions to the corresponding
> > > > > > >   * queued spinlock functions.
> > > > > > > @@ -146,5 +147,6 @@ static __always_inline bool virt_spin_lock(struct qspinlock *lock)
> > > > > > >  #define arch_spin_lock(l)              queued_spin_lock(l)
> > > > > > >  #define arch_spin_trylock(l)           queued_spin_trylock(l)
> > > > > > >  #define arch_spin_unlock(l)            queued_spin_unlock(l)
> > > > > > > +#endif
> > > > > > >
> > > > > > >  #endif /* __ASM_GENERIC_QSPINLOCK_H */
> > > > > > > diff --git a/include/asm-generic/ticket_spinlock.h b/include/asm-generic/ticket_spinlock.h
> > > > > > > index cfcff22b37b3..325779970d8a 100644
> > > > > > > --- a/include/asm-generic/ticket_spinlock.h
> > > > > > > +++ b/include/asm-generic/ticket_spinlock.h
> > > > > > > @@ -89,6 +89,7 @@ static __always_inline int ticket_spin_is_contended(arch_spinlock_t *lock)
> > > > > > >         return (s16)((val >> 16) - (val & 0xffff)) > 1;
> > > > > > >  }
> > > > > > >
> > > > > > > +#ifndef __no_arch_spinlock_redefine
> > > > > > >  /*
> > > > > > >   * Remapping spinlock architecture specific functions to the corresponding
> > > > > > >   * ticket spinlock functions.
> > > > > > > @@ -99,5 +100,6 @@ static __always_inline int ticket_spin_is_contended(arch_spinlock_t *lock)
> > > > > > >  #define arch_spin_lock(l)              ticket_spin_lock(l)
> > > > > > >  #define arch_spin_trylock(l)           ticket_spin_trylock(l)
> > > > > > >  #define arch_spin_unlock(l)            ticket_spin_unlock(l)
> > > > > > > +#endif
> > > > > > >
> > > > > > >  #endif /* __ASM_GENERIC_TICKET_SPINLOCK_H */
> > > > > > > --
> > > > > > > 2.39.2
> > > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Best Regards
> > > > > >  Guo Ren
> > > >
> > > >
> > > >
> > > > --
> > > > Best Regards
> > > >  Guo Ren
> >
> >
> >
> > --
> > Best Regards
> >  Guo Ren



-- 
Best Regards
 Guo Ren





[Index of Archives]     [Linux Kernel]     [Kernel Newbies]     [x86 Platform Driver]     [Netdev]     [Linux Wireless]     [Netfilter]     [Bugtraq]     [Linux Filesystems]     [Yosemite Discussion]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Device Mapper]

  Powered by Linux