Messages in this thread | | | Subject | Re: *sigh* /proc/*/pagemap | From | Matt Mackall <> | Date | Mon, 07 Jul 2008 13:51:05 -0500 |
| |
On Mon, 2008-07-07 at 11:34 -0700, Linus Torvalds wrote: > > On Mon, 7 Jul 2008, Matt Mackall wrote: > > > > It exists to make the VM stop being a big black box. Before now the VM > > exposed little beyond statistics, many of which are basically > > meaningless (RSS?). With pagemap, you can actually see precisely where > > things are getting allocated, how they're getting shared, etc. Think > > NUMA, think cell phones. > > Umm. How about setting it under CONFIG_DEBUG_VM or something like that > then?
I'm certainly not suggesting it's only or even primarily for NUMA or embedded, just -most obviously- useful for them. Fact is, people have been interested enough in this POS that I've been getting patches from all over. Nor is it about debugging the VM, per se. It's about having transparency in a critical area of the kernel. What pages of my browser are never swapped in? What is the memory footprint of my forking server, really? Why the hell can't I get a huge page? Etc.
> Right now it's not even _asked_ about, unless you're EMBEDDED, and > defaults to 'y'. Which means that it has all the downsides, an none of the > upsides, for 99.99% of all users.
That's what I'd done originally (it's certainly -my- inclination). Various people (Rusty comes to mind) convinced me to go the other way.
-- Mathematics is the supreme nostalgia of our time.
| |