On Tue, 29 Mar 2011, Andrew Morton wrote: > > -struct percpu_counter vm_committed_as; > > +struct percpu_counter vm_committed_as ____cacheline_internodealigned_in_smp; > > Why ____cacheline_internodealigned_in_smp? That's pretty aggressive. > > afacit the main benefit from this will occur if the read-only > vm_committed_as.counters lands in the same cacheline as some > write-frequently storage. > > But that's a complete mad guess and I'd prefer not to have to guess. It would be useful to have some functionality that allows us to give hints as to which variables are accessed together and therefore would be useful to put in the same cacheline. Thus avoiding things like the readmostly segment and the above aberration. Andi had a special pda area in earlier version before the merger of 32 and 64 bit code for x86 that resulted in placement of the most performance critical variables near one another. I am afraid now they are all spread out. So maybe something that allows us to define multiple pdas? Or just structs that are cacheline aligned? -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>