On Sun, Jan 10, 2021 at 03:32:47PM +0100, AngeloGioacchino Del Regno wrote: > Then, the set_bit/clear_bit in aw9523_irq_mask(), aw9523_irq_unmask were > replaced with calls to regmap_update_bits_async, example: Why are you trying to use the _async() versions here? Don't you need the I/O to complete synchronously, and what would be the benefit of dealing with the complexity of making things async? The genirq framework is definitely going to expect things to be complete before you return to it... Note also that this only really does anything for SPI, I2C doesn't offer an aynchronous API. > <4>[ 1.064060] regmap_i2c_write+0x1c/0x50 regmap will only use spinlocks if you explictly ask it to do so by setting fast_io. You can't use spinlock based locking for your register cache if your device is attached via I2C, I2C is a very slow bus and I2C controllers will need to sleep. I'd be very surprised if there were any benefit from using a flat cache with such a device frankly, the I/O is going to be so slow that any performance gain from using a flat cache is lost in the noise.
Attachment:
signature.asc
Description: PGP signature
