[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <gwzis5sso3a5g4a7b5tplyfj473b35jzy2cmabs6lgtwa4bc4b@ez3rlmdxphea>
Date: Sun, 18 May 2025 10:18:30 +0200
From: Alejandro Colomar <alx@...nel.org>
To: Peter Xu <peterx@...hat.com>
Cc: linux-man@...r.kernel.org, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, Andrea Arcangeli <aarcange@...hat.com>,
Kyle Huey <me@...ehuey.com>, Robert O'Callahan <robert@...llahan.org>,
Mike Rapoport <rppt@...nel.org>, Suren Baghdasaryan <surenb@...gle.com>,
Axel Rasmussen <axelrasmussen@...gle.com>, Kyle Huey <khuey@...ehuey.com>
Subject: Re: [PATCH v2 1/2] UFFDIO_API.2const: Update userfaultfd handshake
and feature probe
Hi Peter,
On Wed, May 14, 2025 at 01:26:29PM -0400, Peter Xu wrote:
> There's a confusing paragraph in the man page on two-steps handshake for
> userfaultfd UFFDIO_API ioctl. In reality, after a successful UFFDIO_API
> ioctl, the userfaultfd will be locked up on the features and any further
> UFFDIO_API on top of an initialized userfaultfd would fail.
>
> Modify the UFFDIO_API(2const) man page to reflect the reality. Instead,
> add a paragraph explaining the right way to probe userfaultfd features.
> Add that only after the "Before Linux 4.11" paragraph, as the old kernel
> doesn't support any feature anyway.
>
> Fixes: a252b3345 ("ioctl_userfaultfd.2: Describe two-step feature handshake")
> Reviewed-by: Kyle Huey <khuey@...ehuey.com>
> Signed-off-by: Peter Xu <peterx@...hat.com>
Thanks! I've applied the patch.
Have a lovely day!
Alex
> ---
> man/man2const/UFFDIO_API.2const | 44 +++++++++++++++++++--------------
> 1 file changed, 25 insertions(+), 19 deletions(-)
>
> diff --git a/man/man2const/UFFDIO_API.2const b/man/man2const/UFFDIO_API.2const
> index 54b34a1bc..aca27dc5c 100644
> --- a/man/man2const/UFFDIO_API.2const
> +++ b/man/man2const/UFFDIO_API.2const
> @@ -42,25 +42,6 @@ fields to bit masks representing all the available features and the generic
> .BR ioctl (2)
> operations available.
> .P
> -Since Linux 4.11,
> -applications should use the
> -.I features
> -field to perform a two-step handshake.
> -First,
> -.B UFFDIO_API
> -is called with the
> -.I features
> -field set to zero.
> -The kernel responds by setting all supported feature bits.
> -.P
> -Applications which do not require any specific features
> -can begin using the userfaultfd immediately.
> -Applications which do need specific features
> -should call
> -.B UFFDIO_API
> -again with a subset of the reported feature bits set
> -to enable those features.
> -.P
> Before Linux 4.11, the
> .I features
> field must be initialized to zero before the call to
> @@ -70,6 +51,31 @@ and zero (i.e., no feature bits) is placed in the
> field by the kernel upon return from
> .BR ioctl (2).
> .P
> +Since Linux 4.11,
> +userfaultfd supports features that need to be enabled explicitly.
> +To enable any of the features,
> +one needs to set the corresponding feature bits in
> +.I features
> +when issuing the
> +.B UFFDIO_API
> +ioctl.
> +.P
> +For historical reasons,
> +a temporary userfaultfd is needed to probe
> +what userfaultfd features the kernel supports.
> +The application needs to create a temporary userfaultfd,
> +issue an
> +.B UFFDIO_API
> +ioctl with
> +.I features
> +set to zero.
> +After the
> +.B UFFDIO_API
> +ioctl returns successfully,
> +.I features
> +should contain all the userfaultfd features that the kernel supports.
> +The temporary userfaultfd can be safely closed after the probe.
> +.P
> If the application sets unsupported feature bits,
> the kernel will zero out the returned
> .I uffdio_api
> --
> 2.49.0
>
--
<https://www.alejandro-colomar.es/>
Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)
Powered by blists - more mailing lists