On Thu, Mar 13, 2025 at 09:41:51AM +0800, Miaoqing Pan wrote: > On 3/13/2025 12:43 AM, Johan Hovold wrote: > > I've taken a closer look at the driver and it seems like we're missing a > > read barrier to make sure that the updated descriptor is not read until > > after the head pointer. > > > > Miaoqing, could you try the below patch with your reproducer and see if > > it is enough to fix the corruption? > > Sure, the stress test is running. Thanks. > > If so I can resend with the warning removed and include a corresponding > > fix for ath12k (it looks like there are further places where barriers > > are missing too). > If the DMA read barrier works, do you think my submitted patch series is > still needed? Because the error handling is incorrect. Yeah, it would still be good to fix up the error handling even if you don't expect to ever see a descriptor with length 0. But unless the device is doing something wrong here, there shouldn't be a need for peeking at the descriptor and retrying. Johan