* Peter Zijlstra <peterz@xxxxxxxxxxxxx> [2011-11-28 16:29:54]:
> On Fri, 2011-11-18 at 16:37 +0530, Srikar Dronamraju wrote:
> > +static void __unregister_uprobe(struct inode *inode, loff_t offset,
> > + struct uprobe *uprobe)
> > +{
> > + struct list_head try_list;
> > + struct address_space *mapping;
> > + struct vma_info *vi, *tmpvi;
> > + struct vm_area_struct *vma;
> > + struct mm_struct *mm;
> > + loff_t vaddr;
> > +
> > + mapping = inode->i_mapping;
> > + INIT_LIST_HEAD(&try_list);
> > + while ((vi = find_next_vma_info(&try_list, offset,
> > + mapping, false)) != NULL) {
> > + if (IS_ERR(vi))
> > + break;
>
> So what kind of half-assed state are we left in if we try an unregister
> under memory pressure and how do we deal with that?
>
Agree, Even I had this concern and wanted to see if there are ways to
deal with this.
- One approach would be pass extra GFG flags while we do allocations
atleast in the unregister_uprobe.
Drawback of this approach: if the system is already under memory
pressure we shouldnt exert more pressure by asking it to repeat.
- The other approach would be to cache these temporary objects while we
insert probes. i.e keep these metadata around.
I am sure you wouldnt want to add additional metadata.
- Third approach would be to have a completion/worker routine kick in if
unregister_uprobe fails due to memory allocations.
This looks better than the rest.
Do you have any other approaches that we could try?
--
thanks and regards
Srikar
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. 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>