Re: [PATCH bpf-next v7 32/34] bpf: eliminate rlimit-based memory accounting infra for bpf maps

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Nov 20, 2020 at 06:52:27PM -0800, Alexei Starovoitov wrote:
> On Thu, Nov 19, 2020 at 09:37:52AM -0800, Roman Gushchin wrote:
> >  static void bpf_map_put_uref(struct bpf_map *map)
> > @@ -619,7 +562,7 @@ static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
> >  		   "value_size:\t%u\n"
> >  		   "max_entries:\t%u\n"
> >  		   "map_flags:\t%#x\n"
> > -		   "memlock:\t%llu\n"
> > +		   "memlock:\t%llu\n" /* deprecated */
> >  		   "map_id:\t%u\n"
> >  		   "frozen:\t%u\n",
> >  		   map->map_type,
> > @@ -627,7 +570,7 @@ static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
> >  		   map->value_size,
> >  		   map->max_entries,
> >  		   map->map_flags,
> > -		   map->memory.pages * 1ULL << PAGE_SHIFT,
> > +		   0LLU,
> 
> The set looks great to me overall, but above change is problematic.
> There are tools out there that read this value.
> Returning zero might cause oncall alarms to trigger.
> I think we can be more accurate here.
> Instead of zero the kernel can return
> round_up(max_entries * round_up(key_size + value_size, 8), PAGE_SIZE)
> It's not the same as before, but at least the numbers won't suddenly
> go to zero and comparison between maps is still relevant.
> Of course we can introduce a page size calculating callback per map type,
> but imo that would be overkill. These monitoring tools don't care about
> precise number, but rather about relative value and growth from one
> version of the application to another.
> 
> If Daniel doesn't find other issues this can be fixed in the follow up.

Makes total sense. I'll prepare a follow-up patch.

Thanks!




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux