CVE-2025-38354

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/msm/gpu: Fix crash when throttling GPU immediately during boot<br /> <br /> There is a small chance that the GPU is already hot during boot. In that<br /> case, the call to of_devfreq_cooling_register() will immediately try to<br /> apply devfreq cooling, as seen in the following crash:<br /> <br /> Unable to handle kernel paging request at virtual address 0000000000014110<br /> pc : a6xx_gpu_busy+0x1c/0x58 [msm]<br /> lr : msm_devfreq_get_dev_status+0xbc/0x140 [msm]<br /> Call trace:<br /> a6xx_gpu_busy+0x1c/0x58 [msm] (P)<br /> devfreq_simple_ondemand_func+0x3c/0x150<br /> devfreq_update_target+0x44/0xd8<br /> qos_max_notifier_call+0x30/0x84<br /> blocking_notifier_call_chain+0x6c/0xa0<br /> pm_qos_update_target+0xd0/0x110<br /> freq_qos_apply+0x3c/0x74<br /> apply_constraint+0x88/0x148<br /> __dev_pm_qos_update_request+0x7c/0xcc<br /> dev_pm_qos_update_request+0x38/0x5c<br /> devfreq_cooling_set_cur_state+0x98/0xf0<br /> __thermal_cdev_update+0x64/0xb4<br /> thermal_cdev_update+0x4c/0x58<br /> step_wise_manage+0x1f0/0x318<br /> __thermal_zone_device_update+0x278/0x424<br /> __thermal_cooling_device_register+0x2bc/0x308<br /> thermal_of_cooling_device_register+0x10/0x1c<br /> of_devfreq_cooling_register_power+0x240/0x2bc<br /> of_devfreq_cooling_register+0x14/0x20<br /> msm_devfreq_init+0xc4/0x1a0 [msm]<br /> msm_gpu_init+0x304/0x574 [msm]<br /> adreno_gpu_init+0x1c4/0x2e0 [msm]<br /> a6xx_gpu_init+0x5c8/0x9c8 [msm]<br /> adreno_bind+0x2a8/0x33c [msm]<br /> ...<br /> <br /> At this point we haven&amp;#39;t initialized the GMU at all yet, so we cannot read<br /> the GMU registers inside a6xx_gpu_busy(). A similar issue was fixed before<br /> in commit 6694482a70e9 ("drm/msm: Avoid unclocked GMU register access in<br /> 6xx gpu_busy"): msm_devfreq_init() does call devfreq_suspend_device(), but<br /> unlike msm_devfreq_suspend(), it doesn&amp;#39;t set the df-&gt;suspended flag<br /> accordingly. This means the df-&gt;suspended flag does not match the actual<br /> devfreq state after initialization and msm_devfreq_get_dev_status() will<br /> end up accessing GMU registers, causing the crash.<br /> <br /> Fix this by setting df-&gt;suspended correctly during initialization.<br /> <br /> Patchwork: https://patchwork.freedesktop.org/patch/650772/