From: Jason Gunthorpe <jgg@ziepe.ca>
To: Christoph Hellwig <hch@lst.de>
Cc: Philip Yang <Philip.Yang@amd.com>,
Ralph Campbell <rcampbell@nvidia.com>,
John Hubbard <jhubbard@nvidia.com>,
Felix.Kuehling@amd.com, dri-devel@lists.freedesktop.org,
linux-mm@kvack.org, Jerome Glisse <jglisse@redhat.com>,
amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH hmm 2/8] mm/hmm: don't free the cached pgmap while scanning
Date: Mon, 16 Mar 2020 15:07:13 -0300 [thread overview]
Message-ID: <20200316180713.GI20941@ziepe.ca> (raw)
In-Reply-To: <20200316090250.GB12439@lst.de>
On Mon, Mar 16, 2020 at 10:02:50AM +0100, Christoph Hellwig wrote:
> On Wed, Mar 11, 2020 at 03:35:00PM -0300, Jason Gunthorpe wrote:
> > @@ -694,6 +672,15 @@ long hmm_range_fault(struct hmm_range *range, unsigned int flags)
> > return -EBUSY;
> > ret = walk_page_range(mm, hmm_vma_walk.last, range->end,
> > &hmm_walk_ops, &hmm_vma_walk);
> > + /*
> > + * A pgmap is kept cached in the hmm_vma_walk to avoid expensive
> > + * searching in the probably common case that the pgmap is the
> > + * same for the entire requested range.
> > + */
> > + if (hmm_vma_walk.pgmap) {
> > + put_dev_pagemap(hmm_vma_walk.pgmap);
> > + hmm_vma_walk.pgmap = NULL;
> > + }
> > } while (ret == -EBUSY);
>
> In which case it should only be put on return, and not for every loop.
I chose this to be simple without having to goto unwind it.
So, instead like this:
@@ -683,21 +661,33 @@ long hmm_range_fault(struct hmm_range *range, unsigned int flags)
.flags = flags,
};
struct mm_struct *mm = range->notifier->mm;
- int ret;
+ long ret;
lockdep_assert_held(&mm->mmap_sem);
do {
/* If range is no longer valid force retry. */
if (mmu_interval_check_retry(range->notifier,
- range->notifier_seq))
- return -EBUSY;
+ range->notifier_seq)) {
+ ret = -EBUSY;
+ goto out;
+ }
ret = walk_page_range(mm, hmm_vma_walk.last, range->end,
&hmm_walk_ops, &hmm_vma_walk);
} while (ret == -EBUSY);
if (ret)
- return ret;
- return (hmm_vma_walk.last - range->start) >> PAGE_SHIFT;
+ goto out;
+ ret = (hmm_vma_walk.last - range->start) >> PAGE_SHIFT;
+
+out:
+ /*
+ * A pgmap is kept cached in the hmm_vma_walk to avoid expensive
+ * searching in the probably common case that the pgmap is the
+ * same for the entire requested range.
+ */
+ if (hmm_vma_walk.pgmap)
+ put_dev_pagemap(hmm_vma_walk.pgmap);
+ return ret;
}
EXPORT_SYMBOL(hmm_range_fault);
?
> I still think the right fix is to just delete all the unused and broken
> pgmap handling code. If we ever need to add it back it can be added
> in a proper understood and tested way.
What I want to add is something like
if (pgmap != walk->required_pgmap)
cpu_flags = 0
hmm_range_need_fault(..., cpu_flags, ...)
Which will fix a bug in nouveau where it blindly assumes any device
pages are its own, IIRC.
I think Ralph observed it needs to be here, because if the pgmap
doesn't match then it should trigger migration, in a single call,
rather than iterating.
I'm mostly expecting to replace all the other pgmap code, but keep the
pgmap caching. The existing pgmap stuff seems approx right, but
useless..
Jason
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
next prev parent reply other threads:[~2020-03-17 14:17 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-03-11 18:34 [PATCH hmm 0/8] Various error case bug fixes for hmm_range_fault() Jason Gunthorpe
2020-03-11 18:34 ` [PATCH hmm 1/8] mm/hmm: add missing unmaps of the ptep during hmm_vma_handle_pte() Jason Gunthorpe
2020-03-12 1:28 ` Ralph Campbell
2020-03-12 14:24 ` Jason Gunthorpe
2020-03-11 18:35 ` [PATCH hmm 2/8] mm/hmm: don't free the cached pgmap while scanning Jason Gunthorpe
2020-03-12 1:29 ` Ralph Campbell
[not found] ` <20200316090250.GB12439@lst.de>
2020-03-16 18:07 ` Jason Gunthorpe [this message]
[not found] ` <20200316181324.GA24533@lst.de>
2020-03-16 19:23 ` Jason Gunthorpe
2020-03-11 18:35 ` [PATCH hmm 3/8] mm/hmm: do not call hmm_vma_walk_hole() while holding a spinlock Jason Gunthorpe
2020-03-12 1:31 ` Ralph Campbell
2020-03-12 8:54 ` Steven Price
2020-03-12 10:28 ` [PATCH] mm/hmm: Simplify hmm_vma_walk_pud slightly Steven Price
2020-03-12 14:27 ` Jason Gunthorpe
2020-03-12 14:40 ` Steven Price
2020-03-12 15:11 ` Jason Gunthorpe
2020-03-12 16:16 ` Steven Price
2020-03-12 16:37 ` Jason Gunthorpe
2020-03-12 17:02 ` Steven Price
2020-03-12 17:17 ` Jason Gunthorpe
2020-03-13 19:55 ` Jason Gunthorpe
2020-03-13 21:04 ` Matthew Wilcox
2020-03-13 22:51 ` Jason Gunthorpe
[not found] ` <20200316090503.GC12439@lst.de>
2020-03-16 12:56 ` [PATCH hmm 3/8] mm/hmm: do not call hmm_vma_walk_hole() while holding a spinlock Jason Gunthorpe
2020-03-11 18:35 ` [PATCH hmm 4/8] mm/hmm: add missing pfns set to hmm_vma_walk_pmd() Jason Gunthorpe
2020-03-12 1:33 ` Ralph Campbell
2020-03-11 18:35 ` [PATCH hmm 5/8] mm/hmm: add missing call to hmm_range_need_fault() before returning EFAULT Jason Gunthorpe
2020-03-12 1:34 ` Ralph Campbell
2020-03-11 18:35 ` [PATCH hmm 6/8] mm/hmm: reorganize how !pte_present is handled in hmm_vma_handle_pte() Jason Gunthorpe
2020-03-12 1:36 ` Ralph Campbell
2020-03-11 18:35 ` [PATCH hmm 7/8] mm/hmm: return -EFAULT when setting HMM_PFN_ERROR on requested valid pages Jason Gunthorpe
2020-03-12 1:36 ` Ralph Campbell
2020-03-12 14:35 ` Jason Gunthorpe
2020-03-11 18:35 ` [PATCH hmm 8/8] mm/hmm: add missing call to hmm_pte_need_fault in HMM_PFN_SPECIAL handling Jason Gunthorpe
2020-03-12 1:38 ` Ralph Campbell
[not found] ` <20200316091347.GH12439@lst.de>
2020-03-16 12:10 ` Jason Gunthorpe
[not found] ` <20200316124953.GC17386@lst.de>
2020-03-16 13:04 ` Jason Gunthorpe
[not found] ` <20200316131201.GA17955@lst.de>
[not found] ` <20200317123210.GA12058@lst.de>
2020-03-17 12:53 ` Jason Gunthorpe
[not found] ` <20200317130608.GA13030@lst.de>
2020-03-17 13:25 ` Jason Gunthorpe
2020-03-12 19:33 ` [PATCH hmm 9/8] mm/hmm: do not check pmd_protnone twice in hmm_vma_handle_pmd() Jason Gunthorpe
2020-03-12 23:50 ` Ralph Campbell
2020-03-16 18:25 ` [PATCH hmm 0/8] Various error case bug fixes for hmm_range_fault() Jason Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200316180713.GI20941@ziepe.ca \
--to=jgg@ziepe.ca \
--cc=Felix.Kuehling@amd.com \
--cc=Philip.Yang@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=hch@lst.de \
--cc=jglisse@redhat.com \
--cc=jhubbard@nvidia.com \
--cc=linux-mm@kvack.org \
--cc=rcampbell@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).