Re: [patch V2 08/16] x86/ioperm: Add bitmap sequence numberc

[Thomas Gleixner]

On Tue, 12 Nov 2019, Peter Zijlstra wrote:
> On Mon, Nov 11, 2019 at 11:03:22PM +0100, Thomas Gleixner wrote:
> > Add a globally unique sequence number which is incremented when ioperm() is
> > changing the I/O bitmap of a task. Store the new sequence number in the
> > io_bitmap structure and compare it along with the actual struct pointer
> > with the one which was last loaded on a CPU. Only update the bitmap if
> > either of the two changes. That should further reduce the overhead of I/O
> > bitmap scheduling when there are only a few I/O bitmap users on the system.
> 
> > +	/* Update the sequence number to force an update in switch_to() */
> > +	iobm->sequence = atomic64_add_return(1, &io_bitmap_sequence);
> 
> > +		if (tss->last_bitmap != iobm ||
> > +		    tss->last_sequence != iobm->sequence)
> > +			switch_to_update_io_bitmap(tss, iobm);
> 
> Initially I wondered why we need a globally unique sequence number if we
> already check the struct iobitmap pointer. That ought to make the
> sequence number per-object.
> 
> However, that breaks for memory re-use. So yes, we need that thing to be
> global.

Actually with a global 64bit wide counter we can just avoid the pointer
comparison. Assumed a ioperm() syscall every nanosecond it takes about 584
years of uptime to wrap around. :)

Thanks,

	tglx