On 9/14/21 12:29 PM, Wolfram Sang wrote: > On Mon, Aug 23, 2021 at 11:41:39PM +0200, Marek Vasut wrote: >> Booting ZynqMP with XIIC I2C driver shows multitude of race conditions >> in the XIIC driver. This is because locking is completely missing from >> the driver, and there are odd corner cases where the hardware behaves >> strangely. >> >> Most of these races could be triggered easily when booting on SMP >> machines, like the ZynqMP which has up to 4 cores. It is sufficient >> for the interrupt handler to run on another core than xiic_start_xfer >> and the driver fails completely. >> >> This does not add support for long transfers, this only fixes the >> driver to be usable at all instead of being completely broken. >> >> The V2 fixes a few remaining details which cropped up in deployment >> over the last year or so, so I believe the result should be reasonably >> well tested. >> >> Marek Vasut (6): >> i2c: xiic: Fix broken locking on tx_msg >> i2c: xiic: Drop broken interrupt handler >> i2c: xiic: Defer xiic_wakeup() and __xiic_start_xfer() in >> xiic_process() >> i2c: xiic: Switch from waitqueue to completion >> i2c: xiic: Only ever transfer single message >> i2c: xiic: Fix RX IRQ busy check >> > > Applied to for-next, thanks everyone! > Thx. FYI: Ravi will be sending his series on the top of this one. Thanks, Michal
