RE: [RFC 0/2] Reenable might_sleep() checks for might_fault() when atomic

[David Laight <David.Laight@ACULAB.COM>]

From: David Hildenbrand [mailto:dahi@linux.vnet.ibm.com]
> > From: David Hildenbrand
> > ...
> > > Although it might not be optimal, but keeping a separate counter for
> > > pagefault_disable() as part of the preemption counter seems to be the=
 only
> > > doable thing right now. I am not sure if a completely separated count=
er is even
> > > possible, increasing the size of thread_info.
> >
> > What about adding (say) 0x10000 for the more restrictive test?
> >
> > 	David
> >
>=20
> You mean as part of the preempt counter?
>=20
> The current layout (on my branch) is
>=20
>  * PREEMPT_MASK:        0x000000ff
>  * SOFTIRQ_MASK:        0x0000ff00
>  * HARDIRQ_MASK:        0x000f0000
>  *     NMI_MASK:        0x00100000
>  * PREEMPT_ACTIVE:      0x00200000
>=20
> I would have added
>  * PAGEFAULT_MASK:      0x03C00000

I'm not sure where you'd need to add the bits.

I think the above works because disabling 'HARDIRQ' implicitly
disables 'SOFTIRQ' and 'PREEMPT' (etc), so if 256+ threads
disable PREEMPT everything still works.

So if disabling pagefaults implies that pre-emption is disabled
(but SOFTIRQ is still allowed) then you need to insert your bit(s)
between 0xff00 and 0x00ff.
OTOH if disabling pre-emption implies that pagefaults are disabled
then you'd need to use the lsb and change all the above values.

Which makes me think that 'PREEMPT_ACTIVE' isn't right at all.
Two threads disabling NMIs (or 32  disabling HARDIRQ) won't DTRT.

OTOH I'm only guessing at how this is used.

	David