Message-ID: <alpine.DEB.2.10.1806081025030.14699@sstabellini-ThinkPad-X260>
Date:   Fri, 8 Jun 2018 10:59:07 -0700 (PDT)
From:   Stefano Stabellini <sstabellini@...nel.org>
To:     Oleksandr Andrushchenko <andr2000@...il.com>
cc:     Boris Ostrovsky <boris.ostrovsky@...cle.com>,
        xen-devel@...ts.xenproject.org, linux-kernel@...r.kernel.org,
        dri-devel@...ts.freedesktop.org, linux-media@...r.kernel.org,
        jgross@...e.com, konrad.wilk@...cle.com,
        Stefano Stabellini <sstabellini@...nel.org>,
        daniel.vetter@...el.com, matthew.d.roper@...el.com,
        dongwon.kim@...el.com,
        Oleksandr Andrushchenko <oleksandr_andrushchenko@...m.com>,
        julien.grall@....com
Subject: Re: [Xen-devel] [PATCH v2 5/9] xen/gntdev: Allow mappings for DMA
 buffers

On Fri, 8 Jun 2018, Oleksandr Andrushchenko wrote:
> On 06/08/2018 12:46 AM, Boris Ostrovsky wrote:
> > (Stefano, question for you at the end)
> > 
> > On 06/07/2018 02:39 AM, Oleksandr Andrushchenko wrote:
> > > On 06/07/2018 12:19 AM, Boris Ostrovsky wrote:
> > > > On 06/06/2018 04:14 AM, Oleksandr Andrushchenko wrote:
> > > > > On 06/04/2018 11:12 PM, Boris Ostrovsky wrote:
> > > > > > On 06/01/2018 07:41 AM, Oleksandr Andrushchenko wrote:
> > > > > > @@ -121,8 +146,27 @@ static void gntdev_free_map(struct grant_map
> > > > > > *map)
> > > > > >         if (map == NULL)
> > > > > >             return;
> > > > > >     +#ifdef CONFIG_XEN_GRANT_DMA_ALLOC
> > > *Option 1: kfree(map->frames);*
> > > > > > +    if (map->dma_vaddr) {
> > > > > > +        struct gnttab_dma_alloc_args args;
> > > > > > +
> > > > > > +        args.dev = map->dma_dev;
> > > > > > +        args.coherent = map->dma_flags & GNTDEV_DMA_FLAG_COHERENT;
> > > > > > +        args.nr_pages = map->count;
> > > > > > +        args.pages = map->pages;
> > > > > > +        args.frames = map->frames;
> > > > > > +        args.vaddr = map->dma_vaddr;
> > > > > > +        args.dev_bus_addr = map->dma_bus_addr;
> > > > > > +
> > > > > > +        gnttab_dma_free_pages(&args);
> > > *Option 2: kfree(map->frames);*
> > > > > > +    } else
> > > > > > +#endif
> > > > > >         if (map->pages)
> > > > > >             gnttab_free_pages(map->count, map->pages);
> > > > > > +
> > > > > > +#ifdef CONFIG_XEN_GRANT_DMA_ALLOC
> > > > > > +    kfree(map->frames);
> > > > > > +#endif
> > > > > > 
> > > > > > Can this be done under if (map->dma_vaddr) ?
> > > > > >     In other words, is it
> > > > > > possible for dma_vaddr to be NULL and still have unallocated frames
> > > > > > pointer?
> > > > > It is possible to have vaddr == NULL and frames != NULL as we
> > > > > allocate frames outside of gnttab_dma_alloc_pages which
> > > > > may fail. Calling kfree on NULL pointer is safe,
> > > > I am not questioning safety of the code, I would like avoid another
> > > > ifdef.
> > > Ah, I now understand, so you are asking if we can have
> > > that kfree(map->frames); in the place *Option 2* I marked above.
> > > Unfortunately no: map->frames is allocated before we try to
> > > allocate DMA memory, e.g. before dma_vaddr is set:
> > > [...]
> > >          add->frames = kcalloc(count, sizeof(add->frames[0]),
> > >                        GFP_KERNEL);
> > >          if (!add->frames)
> > >              goto err;
> > > 
> > > [...]
> > >          if (gnttab_dma_alloc_pages(&args))
> > >              goto err;
> > > 
> > >          add->dma_vaddr = args.vaddr;
> > > [...]
> > > err:
> > >      gntdev_free_map(add);
> > > 
> > > So, it is possible to enter gntdev_free_map with
> > > frames != NULL and dma_vaddr == NULL. Option 1 above cannot be used
> > > as map->frames is needed for gnttab_dma_free_pages(&args);
> > > and Option 2 cannot be used as frames != NULL and dma_vaddr == NULL.
> > > Thus, I think that unfortunately we need that #ifdef.
> > > Option 3 below can also be considered, but that seems to be not good
> > > as we free resources in different places which looks inconsistent.
> > 
> > I was only thinking of option 2. But if it is possible to have frames !=
> > NULL and dma_vaddr == NULL then perhaps we indeed will have to live with
> > the extra ifdef.
> ok
> > 
> > > Sorry if I'm still missing your point.
> > > > > so
> > > > > I see no reason to change this code.
> > > > > > >         kfree(map->pages);
> > > > > > >         kfree(map->grants);
> > > > > > >         kfree(map->map_ops);
> > > > > > > @@ -132,7 +176,8 @@ static void gntdev_free_map(struct grant_map
> > > > > > > *map)
> > > > > > >         kfree(map);
> > > > > > >     }
> > > > > > >     -static struct grant_map *gntdev_alloc_map(struct gntdev_priv
> > > > > > > *priv, int count)
> > > > > > > +static struct grant_map *gntdev_alloc_map(struct gntdev_priv
> > > > > > > *priv,
> > > > > > > int count,
> > > > > > > +                      int dma_flags)
> > > > > > >     {
> > > > > > >         struct grant_map *add;
> > > > > > >         int i;
> > > > > > > @@ -155,6 +200,37 @@ static struct grant_map
> > > > > > > *gntdev_alloc_map(struct gntdev_priv *priv, int count)
> > > > > > >             NULL == add->pages)
> > > > > > >             goto err;
> > > > > > >     +#ifdef CONFIG_XEN_GRANT_DMA_ALLOC
> > > > > > > +    add->dma_flags = dma_flags;
> > > > > > > +
> > > > > > > +    /*
> > > > > > > +     * Check if this mapping is requested to be backed
> > > > > > > +     * by a DMA buffer.
> > > > > > > +     */
> > > > > > > +    if (dma_flags & (GNTDEV_DMA_FLAG_WC |
> > > > > > > GNTDEV_DMA_FLAG_COHERENT)) {
> > > > > > > +        struct gnttab_dma_alloc_args args;
> > > > > > > +
> > > > > > > +        add->frames = kcalloc(count, sizeof(add->frames[0]),
> > > > > > > +                      GFP_KERNEL);
> > > > > > > +        if (!add->frames)
> > > > > > > +            goto err;
> > > > > > > +
> > > > > > > +        /* Remember the device, so we can free DMA memory. */
> > > > > > > +        add->dma_dev = priv->dma_dev;
> > > > > > > +
> > > > > > > +        args.dev = priv->dma_dev;
> > > > > > > +        args.coherent = dma_flags & GNTDEV_DMA_FLAG_COHERENT;
> > > > > > > +        args.nr_pages = count;
> > > > > > > +        args.pages = add->pages;
> > > > > > > +        args.frames = add->frames;
> > > > > > > +
> > > > > > > +        if (gnttab_dma_alloc_pages(&args))
> > > *Option 3: kfree(map->frames);*
> > > > > > > +            goto err;
> > > > > > > +
> > > > > > > +        add->dma_vaddr = args.vaddr;
> > > > > > > +        add->dma_bus_addr = args.dev_bus_addr;
> > > > > > > +    } else
> > > > > > > +#endif
> > > > > > >         if (gnttab_alloc_pages(count, add->pages))
> > > > > > >             goto err;
> > > > > > >     @@ -325,6 +401,14 @@ static int map_grant_pages(struct
> > > > > > > grant_map
> > > > > > > *map)
> > > > > > >             map->unmap_ops[i].handle = map->map_ops[i].handle;
> > > > > > >             if (use_ptemod)
> > > > > > >                 map->kunmap_ops[i].handle =
> > > > > > > map->kmap_ops[i].handle;
> > > > > > > +#ifdef CONFIG_XEN_GRANT_DMA_ALLOC
> > > > > > > +        else if (map->dma_vaddr) {
> > > > > > > +            unsigned long mfn;
> > > > > > > +
> > > > > > > +            mfn = __pfn_to_mfn(page_to_pfn(map->pages[i]));
> > > > > > Not pfn_to_mfn()?
> > > > > I'd love to, but pfn_to_mfn is only defined for x86, not ARM: [1]
> > > > > and [2]
> > > > > Thus,
> > > > > 
> > > > > drivers/xen/gntdev.c:408:10: error: implicit declaration of function
> > > > > ‘pfn_to_mfn’ [-Werror=implicit-function-declaration]
> > > > >       mfn = pfn_to_mfn(page_to_pfn(map->pages[i]));
> > > > > 
> > > > > So, I'll keep __pfn_to_mfn
> > > > How will this work on non-PV x86?
> > > So, you mean I need:
> > > #ifdef CONFIG_X86
> > > mfn = pfn_to_mfn(page_to_pfn(map->pages[i]));
> > > #else
> > > mfn = __pfn_to_mfn(page_to_pfn(map->pages[i]));
> > > #endif
> > > 
> > I'd rather fix it in ARM code. Stefano, why does ARM uses the
> > underscored version?
> Do you want me to add one more patch for ARM to wrap __pfn_to_mfn
> with static inline for ARM? e.g.
> static inline ...pfn_to_mfn(...)
> {
>     __pfn_to_mfn();
> }


A Xen on ARM guest doesn't actually know the mfns behind its own
pseudo-physical pages. This is why we stopped using pfn_to_mfn and
started using pfn_to_bfn instead, which will generally return "pfn",
unless the page is a foreign grant. See include/xen/arm/page.h.
pfn_to_bfn was also introduced on x86. For example, see the usage of
pfn_to_bfn in drivers/xen/swiotlb-xen.c. Otherwise, if you don't care
about other mapped grants, you can just use pfn_to_gfn, that always
returns pfn.


Also, for your information, we support different page granularities in
Linux as a Xen guest, see the comment at include/xen/arm/page.h:

  /*
   * The pseudo-physical frame (pfn) used in all the helpers is always based
   * on Xen page granularity (i.e 4KB).
   *
   * A Linux page may be split across multiple non-contiguous Xen page so we
   * have to keep track with frame based on 4KB page granularity.
   *
   * PV drivers should never make a direct usage of those helpers (particularly
   * pfn_to_gfn and gfn_to_pfn).
   */

A Linux page could be 64K, but a Xen page is always 4K. A granted page
is also 4K. We have helpers to take into account the offsets to map
multiple Xen grants in a single Linux page, see for example
drivers/xen/grant-table.c:gnttab_foreach_grant. Most PV drivers have
been converted to be able to work with 64K pages correctly, but if I
remember correctly gntdev.c is the only remaining driver that doesn't
support 64K pages yet, so you don't have to deal with it if you don't
want to.

Powered by blists - more mailing lists