Re: [tip:perfcounters/core] x86: Add NMI types for kmap_atomic

[Ingo Molnar <mingo@elte.hu>]

* Peter Zijlstra <peterz@infradead.org> wrote:

> On Mon, 2009-06-15 at 20:15 +0200, Ingo Molnar wrote:
> > * Peter Zijlstra <peterz@infradead.org> wrote:
> > 
> > > +static int kmap_type_to_context(enum km_type type)
> > > +{
> > > +	switch (type) {
> > > +	case KM_BOUNCE_READ:
> > > +		return KM_CTX_USER;
> > > +	case KM_SKB_SUNRPC_DATA:
> > > +		return KM_CTX_USER;
> > > +	case KM_SKB_DATA_SOFTIRQ:
> > > +		return KM_CTX_SOFTIRQ;
> > > +	case KM_USER0:
> > > +		return KM_CTX_USER;
> > > +	case KM_USER1:
> > > +		return KM_CTX_USER;
> > > +	case KM_BIO_SRC_IRQ:
> > > +		return KM_CTX_IRQ;
> > > +	case KM_BIO_DST_IRQ:
> > > +		return KM_CTX_IRQ;
> > > +	case KM_PTE0:
> > > +		return KM_CTX_USER;
> > > +	case KM_PTE1:
> > > +		return KM_CTX_USER;
> > > +	case KM_IRQ0:
> > > +		return KM_CTX_IRQ;
> > > +	case KM_IRQ1:
> > > +		return KM_CTX_IRQ;
> > > +	case KM_SOFTIRQ0:
> > > +		return KM_CTX_SOFTIRQ;
> > > +	case KM_SOFTIRQ1:
> > > +		return KM_CTX_SOFTIRQ;
> > > +	case KM_NMI:
> > > +		return KM_CTX_NMI;
> > > +	case KM_NMI_PTE:
> > > +		return KM_CTX_NMI;
> > > +	}
> > > +
> > > +	return KM_CTX_MAX;
> > 
> > why not do a very simple stack of atomic kmaps, like Hugh suggested?
> > 
> > That would mean a much simpler interface:
> > 
> > 	kaddr = kmap_atomic(page);
> > 
> > no index needed. The kunmap pops the entry off the stack:
> > 
> > 	kunmap_atomic(kaddr);
> > 
> > This becomes simpler too.
> > 
> > Now, a stack can be overflown by imbalance - but that's easy to 
> > detect and existing entries are easily printed and thus the source 
> > of the leak is easily identified.
> > 
> > In my book this design beats the current enumeration of kmap types 
> > indices hands down ... It would likely be much more robust as well, 
> > and much more easy to extend.
> > 
> > Am i missing any subtlety?
> 
> The above is mostly debug code used to validate the kmap_atomic
> conditions.
> 
> KM_CTX_NMI nests in KM_CTX_IRQ nests in KM_CTX_SOFTIRQ nests in
> KM_CTX_USER.
> 
> And validate that we indeed are in the context specified by the type.
> That is, it will warn if we use KM_IRQ1 with KM_CTX_IRQ from user
> context.
> 
> Some of this was already captured in the old kmap debug code which I
> removed.
> 
> But yes, I should write that nicer..

but ... look at the APIs i propose above. We dont need _any_ 
'types'.

That type enumeration is basically an open-coded allocator. If we do 
a _real_ allocator (a balanced stack of atomic kmaps) we dont need 
any of those indices, and all the potential for mismatch goes away 
as well - a stack nests trivially with IRQ and NMI and arbitrary 
other contexts.

	Ingo