Re: [RFC 3/3] drm/vkms: Implement CRC debugfs API

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Jun 28, 2018 at 10:07 AM, Daniel Vetter <daniel@xxxxxxxx> wrote:
> On Thu, Jun 28, 2018 at 12:24:35AM +0300, Haneen Mohammed wrote:
>> Implement the .set_crc_source() callback.
>> Compute CRC using crc32 on the visible part of the framebuffer.
>> Use work_struct to compute and add CRC at the end of a vblank.
>>
>> Signed-off-by: Haneen Mohammed <hamohammed.sa@xxxxxxxxx>
>
> Ok locking review. I still don't have a clear idea how to best fix this,
> but oh well.
>
>> ---
>>  drivers/gpu/drm/vkms/vkms_crtc.c | 76 ++++++++++++++++++++++++++++++++
>>  drivers/gpu/drm/vkms/vkms_drv.h  | 16 +++++++
>>  2 files changed, 92 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/vkms/vkms_crtc.c b/drivers/gpu/drm/vkms/vkms_crtc.c
>> index 73aae129c37d..d78934b76e33 100644
>> --- a/drivers/gpu/drm/vkms/vkms_crtc.c
>> +++ b/drivers/gpu/drm/vkms/vkms_crtc.c
>> @@ -7,8 +7,78 @@
>>   */
>>
>>  #include "vkms_drv.h"
>> +#include <linux/crc32.h>
>>  #include <drm/drm_atomic_helper.h>
>>  #include <drm/drm_crtc_helper.h>
>> +#include <drm/drm_gem_framebuffer_helper.h>
>> +
>> +uint32_t _vkms_get_crc(struct drm_framebuffer *fb)
>> +{
>> +     struct drm_gem_object *gem_obj = drm_gem_fb_get_obj(fb, 0);
>> +     struct vkms_gem_object *vkms_obj = container_of(gem_obj,
>> +                                                     struct vkms_gem_object,
>> +                                                     gem);
>> +     u32 crc = 0;
>> +     int i = 0;
>> +     struct drm_plane_state *state = vkms_obj->plane_state;
>
> vkms_obj->plane_state is the first locking issue: You store the pointer
> without any locks or synchronization, which means the crc worker here
> could access is while it's being changed, resulting in an inconsistent
> crc. Note that the compiler/cpu is allowed to read a pointer multiple
> times if it's not protected by locking/barriers, so all kinds of funny
> stuff can happen here.
>
> Second issue is that the memory you're pointing to isn't owned by the crc
> subsystem, but by the atomic commit worker. That could release the memory
> (and it might get reused for something else meanwhile) before the crc
> worker here is done with it.
>
>> +     unsigned int x = state->src.x1 >> 16;
>> +     unsigned int y = state->src.y1 >> 16;
>> +     unsigned int height = drm_rect_height(&state->src) >> 16;
>> +     unsigned int width = drm_rect_width(&state->src) >> 16;
>> +     unsigned int cpp = fb->format->cpp[0];
>> +     unsigned int src_offset;
>> +     unsigned int size_byte = width * cpp;
>> +     void *vaddr = vkms_obj->vaddr;
>> +
>> +     if (WARN_ON(!vaddr))
>> +             return crc;
>> +
>> +     for (i = y; i < y + height; i++) {
>> +             src_offset = fb->offsets[0] + (i * fb->pitches[0]) + (x * cpp);
>> +             crc = crc32_le(crc, vaddr + src_offset, size_byte);
>> +     }
>> +
>> +     return crc;
>> +}
>> +
>> +void vkms_crc_work_handle(struct work_struct *work)
>> +{
>> +     struct vkms_crtc_state *crtc_state = container_of(work,
>> +                                          struct vkms_crtc_state,
>> +                                          crc_workq);
>> +     struct drm_crtc *crtc = crtc_state->crtc;
>> +     struct drm_framebuffer *fb = (crtc->primary ? crtc->primary->fb : NULL);
>
> The drm_plane->fb pointer has similar issues: No locking (the only thing
> that's really allowed to change this is the atomic commit worker for
> atomic drivers, nothing else), so same issues with inconsistent data and
> using possibly freed memory.
>
> On top Ville has a patch series to set drm_plane->fb == NULL for all
> atomic drivers - it's really a leftover from the pre-atomic age and
> doesn't fit for atomic.

Ville just said that he pushed all that. Are you sure that you're on
latest drm-tip? On latest drm-tip primary->fb should be always NULL
for an atomic driver like vkms ... If you're not on latest drm-tip
probably good idea to rebase everything.
-Daniel

>
> On top of these issues for ->plane_state and ->fb we have a third issue:
> The crc computation must be atomic wrt the generated vblank event for a
> given atomic update. That means we must make sure that the crc we're
> generating is exactly for the plane_state/fb that also issued the vblank
> event. There's some atomic kms tests in igt that check this. More
> concretely we need to make sure that:
> - if the vblank code sees the new vblank event, then the crc _must_
>   compute the crc for the new configuration.
> - if the vblank code does not yet see the new vblank event, then the crc
>   _must_ compute the crc for the old configuration.
>
> One simple solution, but quite a bit of code to type would be.
>
> 1. add a spinlock to the vkms_crtc structure (probably need that first) to
> protect all the crc/vblank event stuff. Needs to be a spinlock since we
> must look at this from the hrtimer.
>
> 2. Create a new structure where you make a copy (not just pointers) of all
> the data the hrtimer needs. So src offset, vblank event, all that stuff.
> For the fb pointer you need to make sure it's properly reference counted
> (drm_framebuffer_get/put).
>
> 3. For every atomic update, allocate a new such structure and store it in
> the vkms_crtc structure (holding the spinlock while updating the pointer).
>
> 4. In the hrtimer grab that pointer and set it to NULL (again holding the
> spinlock), and then explicitly send out the vblank event (using
> drm_crtc_send_vblank_event). And then pass that structure with everything
> the crc worker needs to the crc worker.
>
> I think this should work, mostly.
> -Daniel
>
>> +
>> +     if (crtc_state->crc_enabled && fb) {
>> +             u32 crc32 = _vkms_get_crc(fb);
>> +             unsigned int n_frame = drm_crtc_accurate_vblank_count(crtc);
>> +
>> +             drm_crtc_add_crc_entry(crtc, true, n_frame, &crc32);
>> +     }
>> +}
>> +
>> +static int vkms_set_crc_source(struct drm_crtc *crtc, const char *src_name,
>> +                            size_t *values_cnt)
>> +{
>> +     struct drm_device *dev = crtc->dev;
>> +     struct vkms_device *vkms_dev = container_of(dev,
>> +                                                 struct vkms_device,
>> +                                                 drm);
>> +     struct vkms_crtc_state *crtc_state = &vkms_dev->output.crtc_state;
>> +
>> +     if (!src_name) {
>> +             crtc_state->crc_enabled = false;
>> +     } else if (strcmp(src_name, "auto") == 0) {
>> +             crtc_state->crc_enabled = true;
>> +     } else {
>> +             crtc_state->crc_enabled = false;
>> +             return -EINVAL;
>> +     }
>> +
>> +     *values_cnt = 1;
>> +
>> +     return 0;
>> +}
>>
>>  static enum hrtimer_restart vkms_vblank_simulate(struct hrtimer *timer)
>>  {
>> @@ -23,6 +93,8 @@ static enum hrtimer_restart vkms_vblank_simulate(struct hrtimer *timer)
>>       if (!ret)
>>               DRM_ERROR("vkms failure on handling vblank");
>>
>> +     vkms_crc_work_handle(&output->crtc_state.crc_workq);
>> +
>>       spin_lock_irqsave(&crtc->dev->event_lock, flags);
>>       if (output->event) {
>>               drm_crtc_send_vblank_event(crtc, output->event);
>> @@ -46,6 +118,9 @@ static int vkms_enable_vblank(struct drm_crtc *crtc)
>>       out->period_ns = ktime_set(0, DEFAULT_VBLANK_NS);
>>       hrtimer_start(&out->vblank_hrtimer, out->period_ns, HRTIMER_MODE_ABS);
>>
>> +     out->crtc_state.crtc = crtc;
>> +     INIT_WORK(&out->crtc_state.crc_workq, vkms_crc_work_handle);
>> +
>>       return 0;
>>  }
>>
>> @@ -65,6 +140,7 @@ static const struct drm_crtc_funcs vkms_crtc_funcs = {
>>       .atomic_destroy_state   = drm_atomic_helper_crtc_destroy_state,
>>       .enable_vblank          = vkms_enable_vblank,
>>       .disable_vblank         = vkms_disable_vblank,
>> +     .set_crc_source         = vkms_set_crc_source,
>>  };
>>
>>  static int vkms_crtc_atomic_check(struct drm_crtc *crtc,
>> diff --git a/drivers/gpu/drm/vkms/vkms_drv.h b/drivers/gpu/drm/vkms/vkms_drv.h
>> index 300e6a05d473..4a1458731dd7 100644
>> --- a/drivers/gpu/drm/vkms/vkms_drv.h
>> +++ b/drivers/gpu/drm/vkms/vkms_drv.h
>> @@ -22,6 +22,18 @@ static const u32 vkms_formats[] = {
>>       DRM_FORMAT_XRGB8888,
>>  };
>>
>> +/**
>> + * struct vkms_crtc_state
>> + * @crtc: backpointer to the CRTC
>> + * @crc_workq: worker that captures CRCs for each frame
>> + * @crc_enabled: flag to test if crc is enabled
>> + */
>> +struct vkms_crtc_state {
>> +     struct drm_crtc *crtc;
>> +     struct work_struct crc_workq;
>> +     bool crc_enabled;
>> +};
>> +
>>  struct vkms_output {
>>       struct drm_crtc crtc;
>>       struct drm_encoder encoder;
>> @@ -29,6 +41,7 @@ struct vkms_output {
>>       struct hrtimer vblank_hrtimer;
>>       ktime_t period_ns;
>>       struct drm_pending_vblank_event *event;
>> +     struct vkms_crtc_state crtc_state;
>>  };
>>
>>  struct vkms_device {
>> @@ -55,6 +68,9 @@ int vkms_output_init(struct vkms_device *vkmsdev);
>>
>>  struct drm_plane *vkms_plane_init(struct vkms_device *vkmsdev);
>>
>> +/* CRC Support */
>> +void vkms_crc_work_handle(struct work_struct *work);
>> +
>>  /* Gem stuff */
>>  struct drm_gem_object *vkms_gem_create(struct drm_device *dev,
>>                                      struct drm_file *file,
>> --
>> 2.17.1
>>
>
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch



-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux