On Fri, Jan 14, 2022 at 10:22:56AM +0800, hammer hsieh wrote: > Jiri Slaby <jirislaby@xxxxxxxxxx> 於 2022年1月13日 週四 下午7:12寫道: > > > > On 13. 01. 22, 11:56, hammer hsieh wrote: > > >> Could you explain me what posted write is and how does it not matter in > > >> this case? > > >> > > > > > > Each UART ISC register contains > > > > No, you still don't follow what I write. Use your favorite web search > > for "posted write" and/or consult with your HW team. > > > > Maybe this time, we are on the same page. > Our SP7021 chipset is designed on ARM Cortex-A7 Quad core. > Register Access through AMBA(AXI bus), and it is non-cached. > > Did you mean > case1 have concern about "posted write", and you want to know why it not matter? > case2 will be safer? > > Case1 : > spin_lock_irq_save() > writel(0, target register) > spin_unlock_irqrestore() A lock does not mean that your write made it to the device. Please talk to the hardware designers to properly determine how to correctly write to the hardware and "know" that the write succeeded or not. This driver does not seem to take that into consideration at all. thanks, greg k-h