On 1/9/26 12:29 AM, Nathan Chancellor wrote:
> An (admittedly problematic) optimization change in LLVM 20 [1] turns
> known division by zero into the equivalent of __builtin_unreachable(),
> which invokes undefined behavior if it is encountered in a control flow
> graph, destroying code generation. When compile testing for x86_64,
> objtool flags an instance of this optimization triggering in
> msm_dp_ctrl_config_msa(), inlined into msm_dp_ctrl_on_stream():
> 
>   drivers/gpu/drm/msm/msm.o: warning: objtool: msm_dp_ctrl_on_stream(): 
> unexpected end of section .text.msm_dp_ctrl_on_stream
> 
> The zero division happens if the else branch in the first if statement
> in msm_dp_ctrl_config_msa() is taken because pixel_div is initialized to
> zero and it is not possible for LLVM to eliminate the else branch since
> rate is still not known after inlining into msm_dp_ctrl_on_stream().
> Change pixel_div to one to make the division well defined in the
> presence of an unsupported rate, relying on the DRM_ERROR print to
> indicate the error to the user.
> 
> Fixes: c943b4948b58 ("drm/msm/dp: add displayPort driver support")
> Link: 
> https://github.com/llvm/llvm-project/commit/37932643abab699e8bb1def08b7eb4eae7ff1448
>  [1]
> Reported-by: kernel test robot <[email protected]>
> Closes: 
> https://lore.kernel.org/oe-kbuild-all/[email protected]/
> Signed-off-by: Nathan Chancellor <[email protected]>
> ---
>  drivers/gpu/drm/msm/dp/dp_ctrl.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/msm/dp/dp_ctrl.c 
> b/drivers/gpu/drm/msm/dp/dp_ctrl.c
> index cbcc7c2f0ffc..e4731c059ed8 100644
> --- a/drivers/gpu/drm/msm/dp/dp_ctrl.c
> +++ b/drivers/gpu/drm/msm/dp/dp_ctrl.c
> @@ -2395,7 +2395,7 @@ static void msm_dp_ctrl_config_msa(struct 
> msm_dp_ctrl_private *ctrl,
>                              bool is_ycbcr_420)
>  {
>       u32 pixel_m, pixel_n;
> -     u32 mvid, nvid, pixel_div = 0, dispcc_input_rate;
> +     u32 mvid, nvid, pixel_div = 1, dispcc_input_rate;
>       u32 const nvid_fixed = DP_LINK_CONSTANT_N_VALUE;
>       u32 const link_rate_hbr2 = 540000;
>       u32 const link_rate_hbr3 = 810000;
> 
> ---
> base-commit: 66691e272e40c91305f1704695e0cb340cd162ff
> change-id: 20260108-drm-msm-dp_ctrl-avoid-zero-div-be5dc40cbc18

Dmitry, would it be beneficial to throw an actual error when the rate is
is mangled? i.e.

diff --git a/drivers/gpu/drm/msm/dp/dp_ctrl.c b/drivers/gpu/drm/msm/dp/dp_ctrl.c
index aa2303d0e148..4f710b8e6bc6 100644
--- a/drivers/gpu/drm/msm/dp/dp_ctrl.c
+++ b/drivers/gpu/drm/msm/dp/dp_ctrl.c
@@ -2404,9 +2404,9 @@ static int msm_dp_ctrl_link_retrain(struct 
msm_dp_ctrl_private *ctrl)
        return msm_dp_ctrl_setup_main_link(ctrl, &training_step);
 }
 
-static void msm_dp_ctrl_config_msa(struct msm_dp_ctrl_private *ctrl,
-                              u32 rate, u32 stream_rate_khz,
-                              bool is_ycbcr_420)
+static int msm_dp_ctrl_config_msa(struct msm_dp_ctrl_private *ctrl,
+                                 u32 rate, u32 stream_rate_khz,
+                                 bool is_ycbcr_420)
 {
        u32 pixel_m, pixel_n;
        u32 mvid, nvid, pixel_div = 0, dispcc_input_rate;
@@ -2415,14 +2415,21 @@ static void msm_dp_ctrl_config_msa(struct 
msm_dp_ctrl_private *ctrl,
        u32 const link_rate_hbr3 = 810000;
        unsigned long den, num;
 
-       if (rate == link_rate_hbr3)
+       switch (rate) {
+       case link_rate_hbr3:
                pixel_div = 6;
-       else if (rate == 162000 || rate == 270000)
-               pixel_div = 2;
-       else if (rate == link_rate_hbr2)
+               break;
+       case link_rate_hbr2:
                pixel_div = 4;
-       else
+               break;
+       case 270000:
+       case 162000:
+               pixel_div = 2;
+               break;
+       default:
                DRM_ERROR("Invalid pixel mux divider\n");
+               return -EINVAL;
+       }
 
        dispcc_input_rate = (rate * 10) / pixel_div;
 
@@ -2458,6 +2465,8 @@ static void msm_dp_ctrl_config_msa(struct 
msm_dp_ctrl_private *ctrl,
        drm_dbg_dp(ctrl->drm_dev, "mvid=0x%x, nvid=0x%x\n", mvid, nvid);
        msm_dp_write_link(ctrl, REG_DP_SOFTWARE_MVID, mvid);
        msm_dp_write_link(ctrl, REG_DP_SOFTWARE_NVID, nvid);
+
+       return 0;
 }
 
 int msm_dp_ctrl_on_stream(struct msm_dp_ctrl *msm_dp_ctrl, bool 
force_link_train)
@@ -2525,10 +2534,11 @@ int msm_dp_ctrl_on_stream(struct msm_dp_ctrl 
*msm_dp_ctrl, bool force_link_train
 
        msm_dp_ctrl_configure_source_params(ctrl);
 
-       msm_dp_ctrl_config_msa(ctrl,
-               ctrl->link->link_params.rate,
-               pixel_rate_orig,
-               ctrl->panel->msm_dp_mode.out_fmt_is_yuv_420);
+       ret = msm_dp_ctrl_config_msa(ctrl, ctrl->link->link_params.rate,
+                                    pixel_rate_orig,
+                                    
ctrl->panel->msm_dp_mode.out_fmt_is_yuv_420);
+       if (ret)
+               return ret;
 
        msm_dp_panel_clear_dsc_dto(ctrl->panel);
 


Konrad

Reply via email to