Re: [PATCH] scsi: st: convert convert get_user_pages() --> pin_user_pages()

[John Hubbard]

On 2020-05-21 12:47, Bart Van Assche wrote:
> On 2020-05-18 21:55, John Hubbard wrote:
> > This code was using get_user_pages*(), in a "Case 2" scenario
> > (DMA/RDMA), using the categorization from [1]. That means that it's
> > time to convert the get_user_pages*() + put_page() calls to
> > pin_user_pages*() + unpin_user_pages() calls.
> >
> > There is some helpful background in [2]: basically, this is a small
> > part of fixing a long-standing disconnect between pinning pages, and
> > file systems' use of those pages.
> >
> > Note that this effectively changes the code's behavior as well: it now
> > ultimately calls set_page_dirty_lock(), instead of SetPageDirty().This
> > is probably more accurate.
> >
> > As Christoph Hellwig put it, "set_page_dirty() is only safe if we are
> > dealing with a file backed page where we have reference on the inode it
> > hangs off." [3]
> >
> > Also, this deletes one of the two FIXME comments (about refcounting),
> > because there is nothing wrong with the refcounting at this point.
> >
> > [1] Documentation/core-api/pin_user_pages.rst
> >
> > [2] "Explicit pinning of user-space pages":
> >      https://lwn.net/Articles/807108/
> >
> > [3] https://lore.kernel.org/r/20190723153640.GB720@xxxxxx
>
> Kai, why is the st driver calling get_user_pages_fast() directly instead
> of calling blk_rq_map_user()? blk_rq_map_user() is already used in
> st_scsi_execute(). I think that the blk_rq_map_user() implementation is
> also based on get_user_pages_fast(). See also iov_iter_get_pages_alloc()
> in lib/iov_iter.c.
>
> John, why are the get_user_pages_fast() calls in the st driver modified
> but not the blk_rq_map_user() call? Are you sure that the modified code
> is a "case 2" scenario and not a "case 1" scenario?

No, I am not sure. I thought this was a DMA case (I'm not a SCSI Tape user,
so it *seemed* reasonable that a DMA engine was involved), but if it's really
direct IO, then we need to just drop this patch entirely. Because: I need to
convert the block/biovec code, including iov_iter_get_pages_alloc() and
friends, in order to handle direct IO. I'm working on that but it's not
ready yet.

(I was trying to get the smaller, non-direct-IO cases converted first.)

Thanks for spotting the discrepancy, and apologies for the confusion on this
end.

Also, I doubt if it's worth it, but do you want a patch to change SetPageDirty()
to set_page_dirty_lock(), meanwhile? It seems like if that's never come up, then
it's mostly a theoretical bug.

thanks,
--
John Hubbard
NVIDIA