lkml.org 
[lkml]   [2018]   [Apr]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3] fs: dax: Adding new return type vm_fault_t
On Sun 22-04-18 19:25:05, Matthew Wilcox wrote:
> On Mon, Apr 23, 2018 at 01:09:48AM +0200, Jan Kara wrote:
> > > -int vm_insert_mixed_mkwrite(struct vm_area_struct *vma, unsigned long addr,
> > > - pfn_t pfn)
> > > +vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
> > > + unsigned long addr, pfn_t pfn)
> > > {
> > > - return __vm_insert_mixed(vma, addr, pfn, true);
> > > + int err;
> > > +
> > > + err = __vm_insert_mixed(vma, addr, pfn, true);
> > > + if (err == -ENOMEM)
> > > + return VM_FAULT_OOM;
> > > + if (err < 0 && err != -EBUSY)
> > > + return VM_FAULT_SIGBUS;
> > > + return VM_FAULT_NOPAGE;
> > > }
> > > -EXPORT_SYMBOL(vm_insert_mixed_mkwrite);
> > > +EXPORT_SYMBOL(vmf_insert_mixed_mkwrite);
> >
> > So are we sure that all the callers of this function (and also of
> > vmf_insert_mixed()) are OK with EBUSY? Because especially in the
> > vmf_insert_mixed() case other page than the caller provided is in page
> > tables and thus possibly the caller needs to do some error recovery (such
> > as drop page refcount) in such case...
>
> I went through all the users and didn't find any that did anything
> with -EBUSY other than turn it into VM_FAULT_NOPAGE. I agree that it's
> possible that there might have been someone who wanted to do that, but
> we tend to rely on mapcount (through rmap) rather than refcount (ie we
> use refcount to mean the number of kernel references to the page and then
> use mapcount for the number of times it's mapped into a process' address
> space). All the drivers I audited would allocagte the page first, store
> it in their own data structures, then try to insert it into the virtual
> address space. So an EBUSY always meant "the same page was inserted".
>
> If we did want to support "This happened already" in the future, we
> could define a VM_FAULT flag for that.

OK, fair enough and thanks for doing an audit! So possibly just add a
comment above vmf_insert_mixed() and vmf_insert_mixed_mkwrite() like:

/*
* If the insertion of PTE failed because someone else already added a
* different entry in the mean time, we treat that as success as we assume
* the same entry was actually inserted.
*/

After that feel free to add:

Reviewed-by: Jan Kara <jack@suse.cz>

to the patch.

Honza

--
Jan Kara <jack@suse.com>
SUSE Labs, CR

\
 
 \ /
  Last update: 2018-04-23 16:00    [W:0.086 / U:0.264 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site