This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH -tip 3/3] kprobes: cleanup: use list instead of hlist for insn_pages


* Masami Hiramatsu <mhiramat@redhat.com> wrote:

> Ingo Molnar wrote:
> > * Masami Hiramatsu <mhiramat@redhat.com> wrote:
> > 
> >> Ingo Molnar wrote:
> >>> * Masami Hiramatsu <mhiramat@redhat.com> wrote:
> >>>
> >>>> Use struct list instead of struct hlist for managing insn_pages, 
> >>>> because insn_pages doesn't use hash table.
> >>>>  struct kprobe_insn_page {
> >>>> -	struct hlist_node hlist;
> >>>> +	struct list_head list;
> >>> Hm, you know that this increases the size of kprobe_insn_page by 4/8 
> >>> bytes, right?
> >> Sure, that will increase size.
> >>
> >>> hlists are not just used for hashes - but also when we want a more 
> >>> compact / smaller list head.
> >> Oh, I thought hlists are used for hash tables...
> > 
> > ... because they are smaller, hence the hash table of list 
> > heads becomes twice as dense as with list_head.
> > 
> > But otherwise it's an (almost) equivalent primitive to list_head, 
> > with a slightly higher runtime cost versus better RAM footprint.
> > 
> >>> How many kprobe_insn_page's can be allocated in the system, 
> >>> maximally?
> >> It's depends on how many probes you will use, but logically, 1 
> >> kprobe_insn_pages is allocated per 4096/16 = 256 probes. So, if 
> >> you use 25,600 probes on your system, memory consumption will 
> >> increase 400/800 bytes.
> > 
> > it's your call really - just wanted to react on the 'because it 
> > should be used for hash tables' comment in the changelog.
> 
> Hi Ingo,
> 
> Would I might be misunderstood?
> 
> struct list_head {
>         struct list_head *next, *prev;
> };
> 
> struct hlist_node {
>         struct hlist_node *next, **pprev;
> };
> 
> Both of list_head and hlist_node are the same size...

ahhh ... a light goes up: i read it as hlist_head, while it's 
hlist_node.

You are right, hlist_node is a needless complication so your cleanup 
is correct.

	Ingo


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]