On 1/5/2022 1:34 PM, Stephen Boyd wrote:

Quoting Kuogee Hsieh (2021-12-29 11:17:02)

...

There is kernel crashed due to unable to handle kernel NULL pointer dereference of dp_panel->connector while running DP link layer compliance test case 4.2.2.6 (EDID Corruption Detection).

Can you explain how we get into that situation? Like

"We never assign struct dp_panel::connector, instead the connector is stored in struct msm_dp::connector. When we run compliance testing test case 4.2.2.6 dp_panel_handle_sink_request() won't have a valid edid set in struct dp_panel::edid so we'll try to use the connectors real_edid_checksum and hit a NULL pointer deref error because the connector pointer is never assigned."

...

This patch fixes this problem by populating connector of dp_panel.

[drm:dp_panel_read_sink_caps] *ERROR* panel edid read failed Unable to handle kernel NULL pointer dereference at virtual address 00000000000006e1 Mem abort info: ESR = 0x96000006 EC = 0x25: DABT (current EL), IL = 32 bits SET = 0, FnV = 0 EA = 0, S1PTW = 0 Data abort info: ISV = 0, ISS = 0x00000006 CM = 0, WnR = 0 user pgtable: 4k pages, 39-bit VAs, pgdp=0000000115f25000 [00000000000006e1] pgd=00000001174fe003, p4d=00000001174fe003, pud=00000001174fe003, pmd=0000000000000000 Internal error: Oops: 96000006 [#1] PREEMPT SMP

This sort of stuff isn't really useful because it takes quite a few lines to say "We hit a NULL pointer deref" which was already stated. I'd rather have a clear description of what goes wrong and how setting the pointer in msm_dp_modeset_init() fixes it.

...

{...]

Changes in V2: -- populate panel connector at msm_dp_modeset_init() instead of at dp_panel_read_sink_caps()

Fixes: 7948fe12d47 ("drm/msm/dp: return correct edid checksum after corrupted edid checksum read") Signee-off-by: Kuogee Hsieh quic_khsieh@quicinc.com

---

drivers/gpu/drm/msm/dp/dp_display.c | 25 +++++++++++++++---------- 1 file changed, 15 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c index 3449d3f..c282bbf 100644 --- a/drivers/gpu/drm/msm/dp/dp_display.c +++ b/drivers/gpu/drm/msm/dp/dp_display.c @@ -1495,36 +1495,41 @@ void msm_dp_debugfs_init(struct msm_dp *dp_display, struct drm_minor *minor) } }

-int msm_dp_modeset_init(struct msm_dp *dp_display, struct drm_device *dev, +int msm_dp_modeset_init(struct msm_dp *dp, struct drm_device *dev, struct drm_encoder *encoder) { struct msm_drm_private *priv;

•   struct dp_display_private *dp_display;
     int ret;
  
  

•   if (WARN_ON(!encoder) || WARN_ON(!dp_display) || WARN_ON(!dev))
  

•   if (WARN_ON(!encoder) || WARN_ON(!dp) || WARN_ON(!dev))
             return -EINVAL;
  
     priv = dev->dev_private;
  

•   dp_display->drm_dev = dev;
  

•   dp->drm_dev = dev;
  

•   dp_display = container_of(dp, struct dp_display_private, dp_display);
  
  

•   ret = dp_display_request_irq(dp_display);
  

•   ret = dp_display_request_irq(dp);
     if (ret) {
             DRM_ERROR("request_irq failed, ret=%d\n", ret);
             return ret;
     }
  
  

•   dp_display->encoder = encoder;
  

•   dp->encoder = encoder;
  
  

•   dp_display->connector = dp_drm_connector_init(dp_display);
  

•   if (IS_ERR(dp_display->connector)) {
  

•           ret = PTR_ERR(dp_display->connector);
  

•   dp->connector = dp_drm_connector_init(dp);
  

•   if (IS_ERR(dp->connector)) {
  

•           ret = PTR_ERR(dp->connector);
             DRM_DEV_ERROR(dev->dev,
                     "failed to create dp connector: %d\n", ret);
  

•           dp_display->connector = NULL;
  

•           dp->connector = NULL;
             return ret;
     }
  
  

•   priv->connectors[priv->num_connectors++] = dp_display->connector;
  

•   dp_display->panel->connector = dp->connector;
  

This is the one line that matters I think? Can we reach the connector for the dp device without going through the panel in dp_panel_handle_sink_request()? That would reduce the number of struct elements if possible.

I tried, but very difficulty. It will take more text section space.

...

•   priv->connectors[priv->num_connectors++] = dp->connector;
  

Can we not rename all the local variables in this patch and do it later or never? Reading this patch takes a long time because we have to make sure nothing has actually changed with the rename of 'dp_display' to 'dp'.

...

     return 0;

}

-- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project