Message-ID: <20180815081539.GN32645@dhcp22.suse.cz>
Date:   Wed, 15 Aug 2018 10:15:39 +0200
From:   Michal Hocko <mhocko@...nel.org>
To:     Mike Rapoport <rppt@...ux.vnet.ibm.com>
Cc:     linux-mm@...ck.org, Jonathan Corbet <corbet@....net>,
        Matthew Wilcox <willy@...radead.org>,
        Vlastimil Babka <vbabka@...e.cz>, linux-doc@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [RFC PATCH] docs/core-api: add memory allocation guide

On Wed 15-08-18 09:36:49, Mike Rapoport wrote:
> (this time with the subject, sorry for the noise)
> 
> On Wed, Aug 15, 2018 at 09:34:47AM +0300, Mike Rapoport wrote:
> > As Vlastimil mentioned at [1], it would be nice to have some guide about
> > memory allocation. I've drafted an initial version that tries to summarize
> > "best practices" for allocation functions and GFP usage.
> > 
> > [1] https://www.spinics.net/lists/netfilter-devel/msg55542.html
> > 
> > From 8027c0d4b750b8dbd687234feda63305d0d5a057 Mon Sep 17 00:00:00 2001
> > From: Mike Rapoport <rppt@...ux.vnet.ibm.com>
> > Date: Wed, 15 Aug 2018 09:10:06 +0300
> > Subject: [RFC PATCH] docs/core-api: add memory allocation guide
> > 
> > Signed-off-by: Mike Rapoport <rppt@...ux.vnet.ibm.com>
> > ---
> >  Documentation/core-api/gfp_mask-from-fs-io.rst |   2 +
> >  Documentation/core-api/index.rst               |   1 +
> >  Documentation/core-api/memory-allocation.rst   | 117 +++++++++++++++++++++++++
> >  Documentation/core-api/mm-api.rst              |   2 +
> >  4 files changed, 122 insertions(+)
> >  create mode 100644 Documentation/core-api/memory-allocation.rst
> > 
> > diff --git a/Documentation/core-api/gfp_mask-from-fs-io.rst b/Documentation/core-api/gfp_mask-from-fs-io.rst
> > index e0df8f4..e7c32a8 100644
> > --- a/Documentation/core-api/gfp_mask-from-fs-io.rst
> > +++ b/Documentation/core-api/gfp_mask-from-fs-io.rst
> > @@ -1,3 +1,5 @@
> > +.. _gfp_mask_from_fs_io:
> > +
> >  =================================
> >  GFP masks used from FS/IO context
> >  =================================
> > diff --git a/Documentation/core-api/index.rst b/Documentation/core-api/index.rst
> > index cdc2020..8afc0da 100644
> > --- a/Documentation/core-api/index.rst
> > +++ b/Documentation/core-api/index.rst
> > @@ -27,6 +27,7 @@ Core utilities
> >     errseq
> >     printk-formats
> >     circular-buffers
> > +   memory-allocation
> >     mm-api
> >     gfp_mask-from-fs-io
> >     timekeeping
> > diff --git a/Documentation/core-api/memory-allocation.rst b/Documentation/core-api/memory-allocation.rst
> > new file mode 100644
> > index 0000000..b1f2ad5
> > --- /dev/null
> > +++ b/Documentation/core-api/memory-allocation.rst
> > @@ -0,0 +1,117 @@
> > +=======================
> > +Memory Allocation Guide
> > +=======================
> > +
> > +Linux supplies variety of APIs for memory allocation. You can allocate
> > +small chunks using `kmalloc` or `kmem_cache_alloc` families, large
> > +virtually contiguous areas using `vmalloc` and it's derivatives, or
> > +you can directly request pages from the page allocator with
> > +`__get_free_pages`. It is also possible to use more specialized

I would rather not mention __get_free_pages. alloc_pages is a more
generic API and less subtle one. If you want to mention __get_free_pages
then please make sure to mention the subtlety (namely that is can
allocate only lowmem memory).

> > +allocators, for instance `cma_alloc` or `zs_malloc`.
> > +
> > +Most of the memory allocations APIs use GFP flags to express how that
> > +memory should be allocated. The GFP acronym stands for "get free
> > +pages", the underlying memory allocation function.
> > +
> > +Diversity of the allocation APIs combined with the numerous GFP flags
> > +makes the question "How should I allocate memory?" not that easy to
> > +answer, although very likely you should use
> > +
> > +::
> > +
> > +  kzalloc(<size>, GFP_KERNEL);
> > +
> > +Of course there are cases when other allocation APIs and different GFP
> > +flags must be used.
> > +
> > +Get Free Page flags
> > +===================
> > +
> > +The GFP flags control the allocators behavior. They tell what memory
> > +zones can be used, how hard the allocator should try to find a free
> > +memory, whether the memory can be accessed by the userspace etc. The
> > +:ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>` provides
> > +reference documentation for the GFP flags and their combinations and
> > +here we briefly outline their recommended usage:
> > +
> > +  * Most of the times ``GFP_KERNEL`` is what you need. Memory for the
> > +    kernel data structures, DMAable memory, inode cache, all these and
> > +    many other allocations types can use ``GFP_KERNEL``. Note, that
> > +    using ``GFP_KERNEL`` implies ``GFP_RECLAIM``, which means that
> > +    direct reclaim may be triggered under memory pressure; the calling
> > +    context must be allowed to sleep.
> > +  * If the allocation is performed from an atomic context, e.g
> > +    interrupt handler, use ``GFP_ATOMIC``.

GFP_NOWAIT please. GFP_ATOMIC should be only used if accessing memory
reserves is justified. E.g. fallback allocation would be too costly. It
should be also noted that these allocation are quite likely to fail
especially under memory pressure.

> > +  * Untrusted allocations triggered from userspace should be a subject
> > +    of kmem accounting and must have ``__GFP_ACCOUNT`` bit set. There
> > +    is handy ``GFP_KERNEL_ACCOUNT`` shortcut for ``GFP_KERNEL``
> > +    allocations that should be accounted.
> > +  * Userspace allocations should use either of the ``GFP_USER``,
> > +    ``GFP_HIGHUSER`` and ``GFP_HIGHUSER_MOVABLE`` flags. The longer
> > +    the flag name the less restrictive it is.
> > +
> > +    The ``GFP_HIGHUSER_MOVABLE`` does not require that allocated
> > +    memory will be directly accessible by the kernel or the hardware
> > +    and implies that the data may move.

@may move@is movable@

> > +    The ``GFP_HIGHUSER`` means that the allocated memory is not
> > +    movable, but it is not required to be directly accessible by the
> > +    kernel or the hardware. An example may be a hardware allocation
> > +    that maps data directly into userspace but has no addressing
> > +    limitations.
> > +
> > +    The ``GFP_USER`` means that the allocated memory is not movable
> > +    and it must be directly accessible by the kernel or the
> > +    hardware. It is typically used by hardware for buffers that are
> > +    mapped to userspace (e.g. graphics) that hardware still must DMA
> > +    to.
> > +
> > +You may notice that quite a few allocations in the existing code
> > +specify ``GFP_NOIO`` and ``GFP_NOFS``. Historically, they were used to
> > +prevent recursion deadlocks caused by direct memory reclaim calling
> > +back into the FS or IO paths and blocking on already held
> > +resources. Since 4.12 the preferred way to address this issue is to
> > +use new scope APIs described in
> > +:ref:`Documentation/core-api/gfp_mask-from-fs-io.rst <gfp_mask_from_fs_io>`.
> > +
> > +Another legacy GFP flags are ``GFP_DMA`` and ``GFP_DMA32``. They are
> > +used to ensure that the allocated memory is accessible by hardware
> > +with limited addressing capabilities. So unless you are writing a
> > +driver for a device with such restrictions, avoid using these flags.

And even with HW with restrictions it is preferable to use dma_alloc*
APIs

Looks nice otherwise. Thanks! With the above changes feel free to add
Acked-by: Michal Hocko <mhocko@...e.com>
-- 
Michal Hocko
SUSE Labs

Powered by blists - more mailing lists