[PATCH v11 18/19] media: platform: Add mali-c55 parameters video node
Dan Scally
dan.scally at ideasonboard.com
Wed Jul 30 14:11:22 PDT 2025
Hi Sakari, thanks for the comments
On 30/07/2025 15:36, Sakari Ailus wrote:
> Hi Daniel,
>
> Thanks for the update.
>
> On Mon, Jul 14, 2025 at 04:06:44PM +0100, Daniel Scally wrote:
>> +static int
>> +mali_c55_params_validate_buffer(struct device *dev,
>> + const struct v4l2_params_buffer *buffer)
>> +{
>> + /* Only v1 is supported at the moment. */
>> + if (buffer->version != MALI_C55_PARAM_BUFFER_V1) {
>> + dev_dbg(dev, "Unsupported extensible format version: %u\n",
>> + buffer->version);
>> + return -EINVAL;
>> + }
> Is there anything else to validate here?
Not at this stage; the rest of the validation is done by Jacopo's new framework
>
> I guess nothing is done with the information other than the value is being
> checked above, but if it had an effect on something, one would need to copy
> the information to memory not accessible to the user.
Thats done later in .buf_prepare()
>
>> +
>> + return 0;
>> +}
>> +
>> +static int mali_c55_params_buf_prepare(struct vb2_buffer *vb)
>> +{
>> + struct mali_c55_params *params = vb2_get_drv_priv(vb->vb2_queue);
>> + struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
>> + struct mali_c55_params_buf *buf = to_mali_c55_params_buf(vbuf);
>> + struct mali_c55 *mali_c55 = params->mali_c55;
>> + struct v4l2_params_buffer *config;
>> + int ret;
>> +
>> + ret = v4l2_params_buffer_validate(
>> + mali_c55->dev, vb,
>> + v4l2_params_buffer_size(MALI_C55_PARAMS_MAX_SIZE),
>> + mali_c55_params_validate_buffer);
>> + if (ret)
>> + return ret;
>> +
>> + /*
>> + * Copy the parameters buffer provided by userspace to the internal
>> + * scratch buffer. This protects against the chance of userspace making
>> + * changed to the buffer content whilst the driver processes it.
>> + */
>> + config = vb2_plane_vaddr(vb, 0);
>> + memcpy(buf->config, config, v4l2_params_buffer_size(MALI_C55_PARAMS_MAX_SIZE));
>> +
>> + return v4l2_params_blocks_validate(mali_c55->dev, buf->config,
>> + mali_c55_block_handlers,
>> + ARRAY_SIZE(mali_c55_block_handlers),
>> + NULL);
>> +}
>> +
>> +static void mali_c55_params_buf_queue(struct vb2_buffer *vb)
>> +{
>> + struct mali_c55_params *params = vb2_get_drv_priv(vb->vb2_queue);
>> + struct vb2_v4l2_buffer *vbuf = to_vb2_v4l2_buffer(vb);
>> + struct mali_c55_params_buf *buf = to_mali_c55_params_buf(vbuf);
>> +
>> + spin_lock(¶ms->buffers.lock);
>> + list_add_tail(&buf->queue, ¶ms->buffers.queue);
>> + spin_unlock(¶ms->buffers.lock);
>> +}
>> +
>> +static void mali_c55_params_return_buffers(struct mali_c55_params *params,
>> + enum vb2_buffer_state state)
>> +{
>> + struct mali_c55_params_buf *buf, *tmp;
>> +
>> + guard(spinlock)(¶ms->buffers.lock);
>> +
>> + list_for_each_entry_safe(buf, tmp, ¶ms->buffers.queue, queue) {
>> + list_del(&buf->queue);
>> + vb2_buffer_done(&buf->vb.vb2_buf, state);
>> + }
>> +}
>> +
>> +static int mali_c55_params_start_streaming(struct vb2_queue *q,
>> + unsigned int count)
>> +{
>> + struct mali_c55_params *params = vb2_get_drv_priv(q);
>> + struct mali_c55 *mali_c55 = params->mali_c55;
>> + int ret;
>> +
>> + ret = pm_runtime_resume_and_get(mali_c55->dev);
>> + if (ret)
>> + goto err_return_buffers;
>> +
>> + ret = video_device_pipeline_alloc_start(¶ms->vdev);
>> + if (ret)
>> + goto err_pm_put;
>> +
>> + ret = video_device_pipeline_started(¶ms->vdev);
>> + if (ret < 0)
>> + goto err_stop_pipeline;
>> +
>> + return 0;
>> +
>> +err_stop_pipeline:
>> + video_device_pipeline_stop(¶ms->vdev);
>> +err_pm_put:
>> + pm_runtime_put(mali_c55->dev);
>> +err_return_buffers:
>> + mali_c55_params_return_buffers(params, VB2_BUF_STATE_QUEUED);
>> +
>> + return ret;
>> +}
>> +
>> +static void mali_c55_params_stop_streaming(struct vb2_queue *q)
>> +{
>> + struct mali_c55_params *params = vb2_get_drv_priv(q);
>> + struct media_pipeline *pipe;
>> +
>> + pipe = video_device_pipeline(¶ms->vdev);
>> + if (mali_c55_pipeline_ready(pipe))
>> + media_pipeline_stopped(pipe);
>> +
>> + video_device_pipeline_stop(¶ms->vdev);
>> + mali_c55_params_return_buffers(params, VB2_BUF_STATE_ERROR);
>> +}
>> +
>> +static const struct vb2_ops mali_c55_params_vb2_ops = {
>> + .queue_setup = mali_c55_params_queue_setup,
>> + .buf_init = mali_c55_params_buf_init,
>> + .buf_cleanup = mali_c55_params_buf_cleanup,
>> + .buf_queue = mali_c55_params_buf_queue,
>> + .buf_prepare = mali_c55_params_buf_prepare,
>> + .wait_prepare = vb2_ops_wait_prepare,
>> + .wait_finish = vb2_ops_wait_finish,
>> + .start_streaming = mali_c55_params_start_streaming,
>> + .stop_streaming = mali_c55_params_stop_streaming,
>> +};
>> +
>> +void mali_c55_params_write_config(struct mali_c55 *mali_c55)
>> +{
>> + struct mali_c55_params *params = &mali_c55->params;
>> + struct v4l2_params_buffer *config;
>> + struct mali_c55_params_buf *buf;
>> + size_t block_offset = 0;
>> + size_t max_offset;
>> +
>> + spin_lock(¶ms->buffers.lock);
>> +
>> + buf = list_first_entry_or_null(¶ms->buffers.queue,
>> + struct mali_c55_params_buf, queue);
>> + if (buf)
>> + list_del(&buf->queue);
>> + spin_unlock(¶ms->buffers.lock);
>> +
>> + if (!buf)
>> + return;
>> +
>> + buf->vb.sequence = mali_c55->isp.frame_sequence;
>> + config = buf->config;
>> +
>> + max_offset = config->data_size - sizeof(struct v4l2_params_block_header);
>> +
>> + /*
>> + * Walk the list of parameter blocks and process them. No validation is
>> + * done here, as the contents of the config buffer are already checked
>> + * when the buffer is queued.
>> + */
>> + while (block_offset < max_offset) {
>> + const struct v4l2_params_handler *block_handler;
>> + union mali_c55_params_block block;
>> +
>> + block.data = &config->data[block_offset];
>> +
>> + /* We checked the array index already in .buf_queue() */
> Not a lot seems to be done there in terms of validation.
>
> Even if that had been done, you can't trust the buffer contents as it
> remains mapped to the userspace.
Ah; a hangover of an earlier iteration before using the new v4l2-params helpers; the validation is
done by those functions in .buf_prepare() now and the userspace buffer is memcopied to a scratch buffer.
Thanks
Dan
>> + block_handler = &mali_c55_block_handlers[block.header->type];
>> + block_handler->handler(mali_c55, block.header);
>> +
>> + block_offset += block.header->size;
>> + }
>> +
>> + vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_DONE);
>> +}
More information about the linux-arm-kernel
mailing list