[RFC v2 1/1] drm: allow to use mmuless SoC

Daniel Vetter daniel at ffwll.ch
Thu Dec 1 08:21:34 UTC 2016


On Thu, Dec 01, 2016 at 08:21:11AM +0100, Benjamin Gaignard wrote:
> 2016-12-01 7:56 GMT+01:00 Daniel Vetter <daniel at ffwll.ch>:
> > On Wed, Nov 30, 2016 at 02:19:18PM +0100, Benjamin Gaignard wrote:
> >> Some SoC without MMU have display driver where a drm/kms driver
> >> could be implemented.
> >> Before doing such kind of thing drm/kms must allow to use mmuless
> >> devices.
> >> This patch propose to remove MMU configuration flag and add some
> >> cma helpers functions to help implementing mmuless display driver
> >>
> >> version 2:
> >> - define get_fb_unmapped_area() in drivers/video/fbdev/core/fbmem.c
> >>   if HAVE_ARCH_FB_UNMAPPED_AREA is set and get_fb_unmapped_area not
> >>   already defined in the architecture
> >>
> >> Signed-off-by: Benjamin Gaignard <benjamin.gaignard at linaro.org>
> >> ---
> >>  drivers/gpu/drm/Kconfig              |  4 +--
> >>  drivers/gpu/drm/drm_fb_cma_helper.c  |  4 +++
> >>  drivers/gpu/drm/drm_gem_cma_helper.c | 62 ++++++++++++++++++++++++++++++++++++
> >>  drivers/gpu/drm/drm_vm.c             |  4 +++
> >>  drivers/video/fbdev/core/fbmem.c     | 15 +++++++++
> >>  include/drm/drm_gem_cma_helper.h     |  6 ++++
> >>  6 files changed, 93 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
> >> index 8d9cf73..36c102b 100644
> >> --- a/drivers/gpu/drm/Kconfig
> >> +++ b/drivers/gpu/drm/Kconfig
> >> @@ -6,7 +6,7 @@
> >>  #
> >>  menuconfig DRM
> >>       tristate "Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)"
> >> -     depends on (AGP || AGP=n) && !EMULATED_CMPXCHG && MMU && HAS_DMA
> >> +     depends on (AGP || AGP=n) && !EMULATED_CMPXCHG && HAS_DMA
> >>       select HDMI
> >>       select FB_CMDLINE
> >>       select I2C
> >> @@ -98,7 +98,7 @@ config DRM_LOAD_EDID_FIRMWARE
> >>
> >>  config DRM_TTM
> >>       tristate
> >> -     depends on DRM
> >> +     depends on DRM && MMU
> >>       help
> >>         GPU memory management subsystem for devices with multiple
> >>         GPU memory types. Will be enabled automatically if a device driver
> >> diff --git a/drivers/gpu/drm/drm_fb_cma_helper.c b/drivers/gpu/drm/drm_fb_cma_helper.c
> >> index 81b3558..b1414b6 100644
> >> --- a/drivers/gpu/drm/drm_fb_cma_helper.c
> >> +++ b/drivers/gpu/drm/drm_fb_cma_helper.c
> >> @@ -343,8 +343,12 @@ int drm_fb_cma_debugfs_show(struct seq_file *m, void *arg)
> >>
> >>  static int drm_fb_cma_mmap(struct fb_info *info, struct vm_area_struct *vma)
> >>  {
> >> +#ifdef CONFIG_MMU
> >>       return dma_mmap_writecombine(info->device, vma, info->screen_base,
> >>                                    info->fix.smem_start, info->fix.smem_len);
> >> +#else
> >> +     return vm_iomap_memory(vma, vma->vm_start, info->fix.smem_len);
> >> +#endif
> >>  }
> >>
> >>  static struct fb_ops drm_fbdev_cma_ops = {
> >> diff --git a/drivers/gpu/drm/drm_gem_cma_helper.c b/drivers/gpu/drm/drm_gem_cma_helper.c
> >> index 1d6c335..de13535 100644
> >> --- a/drivers/gpu/drm/drm_gem_cma_helper.c
> >> +++ b/drivers/gpu/drm/drm_gem_cma_helper.c
> >> @@ -318,11 +318,16 @@ static int drm_gem_cma_mmap_obj(struct drm_gem_cma_object *cma_obj,
> >>       vma->vm_flags &= ~VM_PFNMAP;
> >>       vma->vm_pgoff = 0;
> >>
> >> +#ifdef CONFIG_MMU
> >>       ret = dma_mmap_wc(cma_obj->base.dev->dev, vma, cma_obj->vaddr,
> >>                         cma_obj->paddr, vma->vm_end - vma->vm_start);
> >>       if (ret)
> >>               drm_gem_vm_close(vma);
> >>
> >> +#else
> >> +     ret = vm_iomap_memory(vma, vma->vm_start,
> >> +                           (vma->vm_end - vma->vm_start));
> >> +#endif
> >
> > Ok, me asking for a separate helper was probably too much. The #ifdefs
> > here are still ugly, but on 2nd thought this just looks like the dma api
> > is broken on no-mmu? Shouldn't we instead have a patch somewhere which
> > gives us a working dma_mmap_wc on nommu platforms?
> >
> 
> I not sure that dma_mmap_wc is broken for all nommu platforms, but at least
> for ARM nommu it return an error:
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arch/arm/mm/dma-mapping.c#n851
> 
> Could it be an alternative to call vm_iomap_memory if dma_map_wc has
> failed and !CONFIG_MMU ?

Sprinkle #ifdefs all over the place is imo the problem, not what exactly
it looks like. Imo this should be fixed in dma-mapping.c.

And if core maintainers hate that, then we need to define a dma_mmap_wc in
drm headers which works better than the core one ;-) Something like:

#ifdef NOMMU
#define dma_mmap_wc drm_dma_mmap_wc

static inline drm_dma_mmap_wc( .... )
{
	vm_remap_memory( ....);
}
#endif

We should probably create a drm-core.h to collect all these (we have a
metric pile of them in i915 for various things that are too hard to get
merged for no good reason).

> >>       return ret;
> >>  }
> >>
> >> @@ -358,6 +363,63 @@ int drm_gem_cma_mmap(struct file *filp, struct vm_area_struct *vma)
> >>  }
> >>  EXPORT_SYMBOL_GPL(drm_gem_cma_mmap);
> >>
> >> +#ifdef CONFIG_MMU
> >> +#define drm_gem_cma_get_unmapped_area NULL
> >
> > Depending upon how it's used it might be better to stub it out ...
> 
> This function shouldn't be called in MMU case but I can remove the
> flag and compile it for any case
> 
> >
> >> +#else
> >> +unsigned long drm_gem_cma_get_unmapped_area(struct file *filp,
> >> +                                         unsigned long addr,
> >> +                                         unsigned long len,
> >> +                                         unsigned long pgoff,
> >> +                                         unsigned long flags)
> >
> > With no user and no kerneldoc I kinda have no idea what this is even for.
> > The name also doesn't help, it says "get_unmapped_area", but very clearly
> > you're returning a kernel vaddr cast to a  long (and with ERR_PTR
> > semantics on top, but non-standard). Why is this not called get_vaddr?
> 
> This function will be used in driver file_operations structure to set
> get_unmapped_area field.
> For example v4l2 driver use it:
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/media/v4l2-core/v4l2-dev.c#n462
> I don't have found so much examples of that but each time they have
> keep "get_unmapped_area"
> in function name so I just do the same.
> 
> In Documentation/nommu-mmap.txt you have few words about get_unmapped_area()

Hm ok. I think it's time to digg out an old idea I've had:

#define DRM_GEM_CMA_FOPS(driver) \
static const struct file_operations driver ## _drm_fops = { \
	.owner		= THIS_MODULE, \
	.open		= drm_open, \
	.release	= drm_release, \
	.unlocked_ioctl	= drm_ioctl, \
	.compat_ioctl	= drm_compat_ioctl, \
	.poll		= drm_poll, \
	.read		= drm_read, \
	.llseek		= no_llseek, \
	.mmap		= drm_gem_cma_mmap, \
}; \

Then you have one nice place to put the NOMMU #ifdeffery. Would be great
to roll this out to drivers.

Still need kerneldoc for this new hook, with a reference to
Documentation/nommu-mmap.txt.

I think with that bit of polish&abstraction it makes sense to keep this
all in cma helpers itself.
-Daniel


> 
> > -Daniel
> >
> >> +{
> >> +     struct drm_gem_cma_object *cma_obj;
> >> +     struct drm_gem_object *obj = NULL;
> >> +     struct drm_file *priv = filp->private_data;
> >> +     struct drm_device *dev = priv->minor->dev;
> >> +     struct drm_vma_offset_node *node;
> >> +
> >> +     if (drm_device_is_unplugged(dev))
> >> +             return -ENODEV;
> >> +
> >> +     drm_vma_offset_lock_lookup(dev->vma_offset_manager);
> >> +     node = drm_vma_offset_exact_lookup_locked(dev->vma_offset_manager,
> >> +                                               pgoff,
> >> +                                               len >> PAGE_SHIFT);
> >> +     if (likely(node)) {
> >> +             obj = container_of(node, struct drm_gem_object, vma_node);
> >> +             /*
> >> +              * When the object is being freed, after it hits 0-refcnt it
> >> +              * proceeds to tear down the object. In the process it will
> >> +              * attempt to remove the VMA offset and so acquire this
> >> +              * mgr->vm_lock.  Therefore if we find an object with a 0-refcnt
> >> +              * that matches our range, we know it is in the process of being
> >> +              * destroyed and will be freed as soon as we release the lock -
> >> +              * so we have to check for the 0-refcnted object and treat it as
> >> +              * invalid.
> >> +              */
> >> +             if (!kref_get_unless_zero(&obj->refcount))
> >> +                     obj = NULL;
> >> +     }
> >> +
> >> +     drm_vma_offset_unlock_lookup(dev->vma_offset_manager);
> >> +
> >> +     if (!obj)
> >> +             return -EINVAL;
> >> +
> >> +     if (!drm_vma_node_is_allowed(node, priv)) {
> >> +             drm_gem_object_unreference_unlocked(obj);
> >> +             return -EACCES;
> >> +     }
> >> +
> >> +     cma_obj = to_drm_gem_cma_obj(obj);
> >> +
> >> +     drm_gem_object_unreference_unlocked(obj);
> >> +
> >> +     return cma_obj->vaddr ? (unsigned long)cma_obj->vaddr : -EINVAL;
> >> +}
> >> +#endif
> >> +EXPORT_SYMBOL_GPL(drm_gem_cma_get_unmapped_area);
> >> +
> >>  #ifdef CONFIG_DEBUG_FS
> >>  /**
> >>   * drm_gem_cma_describe - describe a CMA GEM object for debugfs
> >> diff --git a/drivers/gpu/drm/drm_vm.c b/drivers/gpu/drm/drm_vm.c
> >> index caa4e4c..974e02b 100644
> >> --- a/drivers/gpu/drm/drm_vm.c
> >> +++ b/drivers/gpu/drm/drm_vm.c
> >> @@ -489,6 +489,7 @@ static int drm_mmap_dma(struct file *filp, struct vm_area_struct *vma)
> >>  #if defined(__i386__) || defined(__x86_64__)
> >>               pgprot_val(vma->vm_page_prot) &= ~_PAGE_RW;
> >>  #else
> >> +#ifdef CONFIG_MMU
> >>               /* Ye gads this is ugly.  With more thought
> >>                  we could move this up higher and use
> >>                  `protection_map' instead.  */
> >> @@ -497,6 +498,7 @@ static int drm_mmap_dma(struct file *filp, struct vm_area_struct *vma)
> >>                            (pte_wrprotect
> >>                             (__pte(pgprot_val(vma->vm_page_prot)))));
> >>  #endif
> >> +#endif
> >>       }
> >>
> >>       vma->vm_ops = &drm_vm_dma_ops;
> >> @@ -573,6 +575,7 @@ static int drm_mmap_locked(struct file *filp, struct vm_area_struct *vma)
> >>  #if defined(__i386__) || defined(__x86_64__)
> >>               pgprot_val(vma->vm_page_prot) &= ~_PAGE_RW;
> >>  #else
> >> +#ifdef CONFIG_MMU
> >>               /* Ye gads this is ugly.  With more thought
> >>                  we could move this up higher and use
> >>                  `protection_map' instead.  */
> >> @@ -581,6 +584,7 @@ static int drm_mmap_locked(struct file *filp, struct vm_area_struct *vma)
> >>                            (pte_wrprotect
> >>                             (__pte(pgprot_val(vma->vm_page_prot)))));
> >>  #endif
> >> +#endif
> >>       }
> >>
> >>       switch (map->type) {
> >> diff --git a/drivers/video/fbdev/core/fbmem.c b/drivers/video/fbdev/core/fbmem.c
> >> index 76c1ad9..54488ee 100644
> >> --- a/drivers/video/fbdev/core/fbmem.c
> >> +++ b/drivers/video/fbdev/core/fbmem.c
> >> @@ -1492,6 +1492,21 @@ static long fb_compat_ioctl(struct file *file, unsigned int cmd,
> >>       return 0;
> >>  }
> >>
> >> +#if defined(HAVE_ARCH_FB_UNMAPPED_AREA) && !defined(get_fb_unmapped_area)
> >> +unsigned long get_fb_unmapped_area(struct file *filp,
> >> +                                 unsigned long addr, unsigned long len,
> >> +                                 unsigned long pgoff, unsigned long flags)
> >> +{
> >> +     struct fb_info * const info = filp->private_data;
> >> +     unsigned long fb_size = PAGE_ALIGN(info->fix.smem_len);
> >> +
> >> +     if (pgoff > fb_size || len > fb_size - pgoff)
> >> +             return -EINVAL;
> >> +
> >> +     return (unsigned long)info->screen_base + pgoff;
> >> +}
> >> +#endif
> >> +
> >>  static const struct file_operations fb_fops = {
> >>       .owner =        THIS_MODULE,
> >>       .read =         fb_read,
> >> diff --git a/include/drm/drm_gem_cma_helper.h b/include/drm/drm_gem_cma_helper.h
> >> index acd6af8..177a20d 100644
> >> --- a/include/drm/drm_gem_cma_helper.h
> >> +++ b/include/drm/drm_gem_cma_helper.h
> >> @@ -53,6 +53,12 @@ struct drm_gem_cma_object *drm_gem_cma_create(struct drm_device *drm,
> >>
> >>  extern const struct vm_operations_struct drm_gem_cma_vm_ops;
> >>
> >> +unsigned long drm_gem_cma_get_unmapped_area(struct file *filp,
> >> +                                          unsigned long addr,
> >> +                                          unsigned long len,
> >> +                                          unsigned long pgoff,
> >> +                                          unsigned long flags);
> >> +
> >>  #ifdef CONFIG_DEBUG_FS
> >>  void drm_gem_cma_describe(struct drm_gem_cma_object *obj, struct seq_file *m);
> >>  #endif
> >> --
> >> 1.9.1
> >>
> >
> > --
> > Daniel Vetter
> > Software Engineer, Intel Corporation
> > http://blog.ffwll.ch
> 
> 
> 
> -- 
> Benjamin Gaignard
> 
> Graphic Study Group
> 
> Linaro.org │ Open source software for ARM SoCs
> 
> Follow Linaro: Facebook | Twitter | Blog

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch


More information about the dri-devel mailing list