On Fri, 13 Nov 2026 at 00:15, <[email protected]> wrote:
>
> Thank you for your contribution! Sashiko AI review found 3 potential
> issues to consider:
> - [High] drm/tiny: trigger5: broken USB reset lifecycle in
> trigger5_reset()
> - [Medium] drm/tiny: trigger5: unclipped src_rect in
> trigger5_plane_atomic_update()
> - [Medium] drm/tiny: trigger5: bulk stream desync on mid-frame
> trigger5_submit_urb() error
> --
>
> commit efc535dee28f1fcc3c108a02e819d423382ef63b
> Author: Ho Jie Feng <[email protected]>
>
> drm/tiny: trigger5: Add MCT Trigger 5 USB display driver
>
> This commit adds a DRM driver for MCT Trigger 5 USB 3 to HDMI/VGA display
> adapters, implementing modesetting, EDID retrieval, and bulk URB frame
> updates.
>
> > diff --git a/drivers/gpu/drm/tiny/trigger5.c
> > b/drivers/gpu/drm/tiny/trigger5.c
> > new file mode 100644
> > index 0000000000000..46fa6e6babe9c
> > --- /dev/null
> > +++ b/drivers/gpu/drm/tiny/trigger5.c
>
> [ ... ]
>
> > +static void trigger5_reset(struct trigger5_device *trigger5, int err)
> > +{
> > + if (!atomic_xchg(&trigger5->needs_reset, 1))
> > + usb_queue_reset_device(trigger5->intf);
> > +
> > + drm_err(&trigger5->drm, "USB transfer failed: %d, resetting
> > adapter\n",
> > + err);
> > +}
> > +
> > +static void trigger5_urb_complete(struct urb *urb)
> > +{
> > + struct trigger5_transfer *transfer = urb->context;
> > + struct trigger5_device *trigger5 = transfer->trigger5;
> > +
> > + switch (urb->status) {
> > + case -ENOENT:
> > + case -ECONNRESET:
> > + case -ESHUTDOWN:
> > + break;
> > + case 0:
> > + if (urb->actual_length == urb->transfer_buffer_length)
> > + break;
> > + fallthrough;
> > + default:
> > + trigger5_reset(trigger5, urb->status);
> > + break;
> > + }
> > +
> > + complete(&transfer->done);
> > +}
> > +
> > +static struct trigger5_transfer *
> > +trigger5_get_next_urb(struct trigger5_device *trigger5)
> > +{
> > + struct trigger5_transfer *transfer =
> > + &trigger5->transfers[trigger5->next_transfer];
> > +
> > + if (atomic_read(&trigger5->needs_reset))
> > + return ERR_PTR(-EIO);
> > +
> > + if (!wait_for_completion_timeout(&transfer->done,
> > +
> > msecs_to_jiffies(TRIGGER5_BULK_TIMEOUT_MS))) {
> > + trigger5_reset(trigger5, -ETIMEDOUT);
> > + return ERR_PTR(-ETIMEDOUT);
> > + }
> > +
> > + trigger5->next_transfer =
> > + (trigger5->next_transfer + 1) % TRIGGER5_NUM_URBS;
> > +
> > + return transfer;
> > +}
>
> [Severity: High]
> Does trigger5_reset() cause the active DRM device to be permanently
> unplugged or stuck returning -EIO after any USB transfer failure?
Don't know. In any case, assuming the device isn't faulty, a user replug
should recover it.
> trigger5_driver does not define .pre_reset and .post_reset callbacks
> (.reset_resume is only used when resuming a suspended device), so
> usb_reset_device() falls back to usb_forced_unbind_intf(), which runs
> trigger5_usb_disconnect() -> drm_dev_unplug() and kills the active DRM
> device instead of resetting the adapter in place.
This is intentionally kept simple for the first revision to be upstreamed.
A reset is simple to implement and correctly recovers the device in my tests.
> In addition, nothing in the driver ever clears trigger5->needs_reset once
> trigger5_reset() sets it to 1, so if the queued reset does not unbind the
> device (for example if usb_lock_device_for_reset() fails or after a
> reset_resume), trigger5_get_next_urb() rejects all future display updates
> with -EIO.
The only reason it fails should be that the device is unplugged, so the point
about reset failing is moot here.
> Two smaller issues in this path: trigger5_get_next_urb() only checks
> trigger5->needs_reset before wait_for_completion_timeout(), so if the URB
> being waited on completes with an error it still returns the transfer to
> the caller; and on a short transfer in trigger5_urb_complete(), falling
> through from case 0 passes urb->status (0) to trigger5_reset().
This passes at most one more URB to a failing device?
trigger5_reset is only put in there in case the dongle fails. However, I have
never observed it to fail in that way, apart from intentionally introducing
some fault condition such as putting a keepalive in the middle of the frame.
Since in this v6, keepalive is disarmed and never sent, it will not happen.
This reset seems to be the most reliable way to recover from something
wrong from my experiments. I do not know enough about the behaviour
when hardware is faulty or some weird condition appears enough to
say that it works in *all* conditions.
> [ ... ]
>
> > +static void trigger5_plane_atomic_update(struct drm_plane *plane,
> > + struct drm_atomic_commit
> > *atomic_state)
> > +{
>
> [ ... ]
>
> > + src_rect = drm_plane_state_src(state);
> > +
> > + /* Match drm_atomic_helper_damage_iter_init() rounding. */
> > + src_rect.x1 >>= 16;
> > + src_rect.y1 >>= 16;
> > + src_rect.x2 = (src_rect.x2 >> 16) + !!(src_rect.x2 & 0xffff);
> > + src_rect.y2 = (src_rect.y2 >> 16) + !!(src_rect.y2 & 0xffff);
> > +
> > + /* Clip merged damage to the new resolution. */
> > + if (!drm_rect_intersect(¤t_rect, &src_rect))
> > + goto exit;
> > +
> > + width = drm_rect_width(¤t_rect);
> > + height = drm_rect_height(¤t_rect);
> > + payload_len = array3_size(width, height, 3);
> > + lines_per_urb = TRIGGER5_URB_BUF_SIZE / (width * 3);
> > +
> > + if (drm_gem_fb_begin_cpu_access(state->fb, DMA_FROM_DEVICE) < 0)
> > + goto exit;
> > +
> > + transfer = trigger5_get_next_urb(trigger5);
> > + if (IS_ERR(transfer))
> > + goto exit_end_cpu_access;
> > +
> > + header = transfer->urb->transfer_buffer;
> > + header->magic = 0xfb;
> > + header->length = 0x14;
> > + /* flags 0: uncompressed 24-bit RGB888. */
> > + header->counter =
> > + cpu_to_le16((trigger5->frame_counter++) & 0xfff);
> > + header->horizontal_offset = cpu_to_le16(current_rect.x1 -
> > src_rect.x1);
> > + header->vertical_offset = cpu_to_le16(current_rect.y1 - src_rect.y1);
> > + header->width = cpu_to_le16(width);
> > + header->height = cpu_to_le16(height);
> > + header->payload_length = cpu_to_le32((u32)payload_len);
>
> [Severity: Medium]
> Should src_rect come from the clipped state->src (via
> drm_rect_fp_to_int(&src_rect, &state->src)) rather than the unclipped
> drm_plane_state_src(state)?
>
> In trigger5_plane_atomic_check(), drm_atomic_helper_check_plane_state()
> clips state->dst to the active CRTC mode and stores the visible source
> rectangle in state->src, while leaving state->src_x/y/w/h unclipped.
> Because drm_atomic_helper_damage_iter_init() also initializes its plane_src
> from the unclipped drm_plane_state_src(state), intersecting current_rect
> with drm_plane_state_src(state) here does not clip damage to the active
> CRTC mode when a larger framebuffer (up to 8191x8191) is panned with
> negative crtc_x/crtc_y, and outward rounding of fractional 16.16 src
> coordinates expands src_rect by an extra pixel. This computes
> horizontal_offset and vertical_offset relative to the off-screen origin and
> can send frame dimensions and payload_length exceeding the active display
> mode and TRIGGER5_MAX_FRAME_SIZE.
Real bug, but unlikely there are weird userspace DRM users weird
enough to hit this bug.
> > + header->flags = 0x1;
> > + header->unknown1 = 0;
> > + header->unknown2 = 0;
> > + header->checksum = trigger5_bulk_header_checksum(header);
> > +
> > + if (trigger5_submit_urb(transfer, sizeof(*header)))
> > + goto exit_end_cpu_access;
> > +
> > + clip = current_rect;
> > + for (y = current_rect.y1; y < current_rect.y2; y += lines_per_urb) {
> > + clip.y1 = y;
> > + clip.y2 = min(y + lines_per_urb, current_rect.y2);
> > +
> > + transfer = trigger5_get_next_urb(trigger5);
> > + if (IS_ERR(transfer))
> > + break;
> > +
> > + iosys_map_set_vaddr(&data_map,
> > transfer->urb->transfer_buffer);
> > + drm_fb_xrgb8888_to_rgb888(&data_map, NULL,
> > + &shadow_plane_state->data[0],
> > + state->fb, &clip,
> > + &shadow_plane_state->fmtcnv_state);
> > +
> > + if (trigger5_submit_urb(transfer,
> > + width * 3 * drm_rect_height(&clip)))
> > + break;
> > + }
>
> [Severity: Medium]
> Does breaking out of this loop when trigger5_submit_urb() fails leave the
> adapter's bulk stream out of sync?
>
> At that point the 20-byte trigger5_bulk_header URB declaring
> header->payload_length bytes has already been submitted, and unlike
> trigger5_urb_complete() or trigger5_get_next_urb(), trigger5_submit_urb()
> does not call trigger5_reset() on usb_submit_urb() failure. The device
> will keep waiting for the remaining bytes of the aborted frame and treat
> subsequent frame headers and pixel data as part of the old payload.
Known bug, however, on submit urb failure it should queue a reset.
I am not sure what is the best way to recover the device
beyond queuing a reset, since further URB submissions to pad the
frame to its entirety might fail too.
> --
> Sashiko AI review ยท
> https://sashiko.dev/#/patchset/[email protected]?part=1