On Thu, 2020-02-06 at 06:46 +0100, Christophe Leroy wrote:
> 
> Le 06/02/2020 à 04:08, Leonardo Bras a écrit :
> > On powerpc, we need to do some lockless pagetable walks from functions
> > that already have disabled interrupts, specially from real mode with
> > MSR[EE=0].
> > 
> > In these contexts, disabling/enabling interrupts can be very troubling.
> 
> When interrupts are already disabled, the flag returned when disabling 
> it will be such that when we restore it later, interrupts remain 
> disabled, so what's the problem ?

There are places in code, like patch 8, where it explicitly avoids
doing irq_save/restore by using a function parameter (realmode).
http://patchwork.ozlabs.org/patch/1234130/

I am not sure why it's that way, but I decided to keep it as is. 
It was introduced by Aneesh Kumar in 2015
(691e95fd7396905a38d98919e9c150dbc3ea21a3). 

> > So, this arch-specific implementation features functions with an extra
> > argument that allows interrupt enable/disable to be skipped:
> > __begin_lockless_pgtbl_walk() and __end_lockless_pgtbl_walk().
> > 
> > Functions similar to the generic ones are also exported, by calling
> > the above functions with parameter {en,dis}able_irq = true.
> > 
> > Signed-off-by: Leonardo Bras <leonardo@linux.ibm.com>
> > ---
> >   arch/powerpc/include/asm/book3s/64/pgtable.h |  6 ++
> >   arch/powerpc/mm/book3s64/pgtable.c           | 86 +++++++++++++++++++-
> >   2 files changed, 91 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
> > index 201a69e6a355..78f6ffb1bb3e 100644
> > --- a/arch/powerpc/include/asm/book3s/64/pgtable.h
> > +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
> > @@ -1375,5 +1375,11 @@ static inline bool pgd_is_leaf(pgd_t pgd)
> >   	return !!(pgd_raw(pgd) & cpu_to_be64(_PAGE_PTE));
> >   }
> >   
> > +#define __HAVE_ARCH_LOCKLESS_PGTBL_WALK_CONTROL
> > +unsigned long begin_lockless_pgtbl_walk(void);
> > +unsigned long __begin_lockless_pgtbl_walk(bool disable_irq);
> > +void end_lockless_pgtbl_walk(unsigned long irq_mask);
> > +void __end_lockless_pgtbl_walk(unsigned long irq_mask, bool enable_irq);
> > +
> 
> Why not make them static inline just like the generic ones ?
> 

Sure, can be done.  It would save some function calls.
For that I will define the per-cpu variable in .c and declare it in .h
All new function can be moved to .h, while changing adding the inline
modifier.

> >   #endif /* __ASSEMBLY__ */
> >   #endif /* _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ */
> > diff --git a/arch/powerpc/mm/book3s64/pgtable.c b/arch/powerpc/mm/book3s64/pgtable.c
> > index 2bf7e1b4fd82..535613030363 100644
> > --- a/arch/powerpc/mm/book3s64/pgtable.c
> > +++ b/arch/powerpc/mm/book3s64/pgtable.c
> > @@ -82,6 +82,7 @@ static void do_nothing(void *unused)
> >   {
> >   
> >   }
> > +
> 
> Is this blank line related to the patch ?
> 

Nope, just something I 'fixed' while reading, and gone past my pre-send 
patch reviewing. If it bothers, I can remove.

> >   /*
> >    * Serialize against find_current_mm_pte which does lock-less
> >    * lookup in page tables with local interrupts disabled. For huge pages
> > @@ -98,6 +99,89 @@ void serialize_against_pte_lookup(struct mm_struct *mm)
> >   	smp_call_function_many(mm_cpumask(mm), do_nothing, NULL, 1);
> >   }
> >   
> > +/* begin_lockless_pgtbl_walk: Must be inserted before a function call that does
> > + *   lockless pagetable walks, such as __find_linux_pte().
> > + * This version allows setting disable_irq=false, so irqs are not touched, which
> > + *   is quite useful for running when ints are already disabled (like real-mode)
> > + */
> > +inline
> > +unsigned long __begin_lockless_pgtbl_walk(bool disable_irq)
> > +{
> > +	unsigned long irq_mask = 0;
> > +
> > +	/*
> > +	 * Interrupts must be disabled during the lockless page table walk.
> > +	 * That's because the deleting or splitting involves flushing TLBs,
> > +	 * which in turn issues interrupts, that will block when disabled.
> > +	 *
> > +	 * When this function is called from realmode with MSR[EE=0],
> > +	 * it's not needed to touch irq, since it's already disabled.
> > +	 */
> > +	if (disable_irq)
> > +		local_irq_save(irq_mask);
> > +
> > +	/*
> > +	 * This memory barrier pairs with any code that is either trying to
> > +	 * delete page tables, or split huge pages. Without this barrier,
> > +	 * the page tables could be read speculatively outside of interrupt
> > +	 * disabling or reference counting.
> > +	 */
> > +	smp_mb();
> > +
> > +	return irq_mask;
> > +}
> > +EXPORT_SYMBOL(__begin_lockless_pgtbl_walk);
> > +
> > +/* begin_lockless_pgtbl_walk: Must be inserted before a function call that does
> > + *   lockless pagetable walks, such as __find_linux_pte().
> > + * This version is used by generic code, and always assume irqs will be disabled
> > + */
> > +unsigned long begin_lockless_pgtbl_walk(void)
> > +{
> > +	return __begin_lockless_pgtbl_walk(true);
> > +}
> > +EXPORT_SYMBOL(begin_lockless_pgtbl_walk);
> 
> Even more than begin_lockless_pgtbl_walk(), this one is worth being 
> static inline in the H file.

Same as above, can be done. 

> 
> > +
> > +/*
> > + * __end_lockless_pgtbl_walk: Must be inserted after the last use of a pointer
> > + *   returned by a lockless pagetable walk, such as __find_linux_pte()
> > + * This version allows setting enable_irq=false, so irqs are not touched, which
> > + *   is quite useful for running when ints are already disabled (like real-mode)
> > + */
> > +inline void __end_lockless_pgtbl_walk(unsigned long irq_mask, bool enable_irq)
> > +{
> > +	/*
> > +	 * This memory barrier pairs with any code that is either trying to
> > +	 * delete page tables, or split huge pages. Without this barrier,
> > +	 * the page tables could be read speculatively outside of interrupt
> > +	 * disabling or reference counting.
> > +	 */
> > +	smp_mb();
> > +
> > +	/*
> > +	 * Interrupts must be disabled during the lockless page table walk.
> > +	 * That's because the deleting or splitting involves flushing TLBs,
> > +	 * which in turn issues interrupts, that will block when disabled.
> > +	 *
> > +	 * When this function is called from realmode with MSR[EE=0],
> > +	 * it's not needed to touch irq, since it's already disabled.
> > +	 */
> > +	if (enable_irq)
> > +		local_irq_restore(irq_mask);
> > +}
> > +EXPORT_SYMBOL(__end_lockless_pgtbl_walk);
> > +
> > +/*
> > + * end_lockless_pgtbl_walk: Must be inserted after the last use of a pointer
> > + *   returned by a lockless pagetable walk, such as __find_linux_pte()
> > + * This version is used by generic code, and always assume irqs will be enabled
> > + */
> > +void end_lockless_pgtbl_walk(unsigned long irq_mask)
> > +{
> > +	__end_lockless_pgtbl_walk(irq_mask, true);
> > +}
> > +EXPORT_SYMBOL(end_lockless_pgtbl_walk);
> > +
> >   /*
> >    * We use this to invalidate a pmdp entry before switching from a
> >    * hugepte to regular pmd entry.
> > @@ -487,7 +571,7 @@ static int __init setup_disable_tlbie(char *str)
> >   	tlbie_capable = false;
> >   	tlbie_enabled = false;
> >   
> > -        return 1;
> > +	return 1;
> 
> Is that related to this patch at all ?

Nope, just another something I 'fixed' while reading, and gone past my
pre-send patch reviewing. If it bothers, I can remove.

> 
> >   }
> >   __setup("disable_tlbie", setup_disable_tlbie);
> >   
> > 
> 
> Christophe

Found other unbalanced begin/end in kvmppc_do_h_enter(). I will change
that too.

Thanks for the feedback,

Leonardo Bras