Hi, Tzung-Bi:
On Wed, 2020-02-12 at 19:31 +0800, Tzung-Bi Shih wrote:
On Wed, Feb 12, 2020 at 4:19 PM CK Hu ck.hu@mediatek.com wrote:
I think hdmi_conn_detect() and mtk_hdmi_audio_hook_plugged_cb() would be called in different thread. So it's necessary to use a mutex to protect this.
Thanks for the reminder. I feel using mutex here is overkill. Please see https://patchwork.kernel.org/patch/11378413/ for my proposed solution.
I'm not only consider the race condition of plugged_cb and codec_dev. I also care about the atomic of mtk_cec_hpd_high() and hdmi->plugged_cb(). If these two function is not an atomic operation, below is an example of problem:
<Status disconnected> 1. Thread A call mtk_hdmi_audio_hook_plugged_cb() 2. Thread A call mtk_cec_hpd_high() and get disconnected. <Status connected> 3. Thread B call hdmi_conn_detect() 4. Thread B call mtk_cec_hpd_high() and get connected 5. Thread B callback plugged_cb() with connected 6. Thread A callback plugged_cb() with disconnected (Bug here)
Regards, CK
Linux-mediatek mailing list Linux-mediatek@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-mediatek