Input: ps2-gpio - use ktime for IRQ timekeeping
Using jiffies for the IRQ timekeeping is not sufficient for two reasons: (1) Usually jiffies have a resolution of 1ms to 10ms. The IRQ intervals based on the clock frequency of PS2 protocol specification (10kHz - 16.7kHz) are between ~60us and 100us only. Therefore only those IRQ intervals can be detected which are either at the end of a transfer or are overly delayed. While this is sufficient in most cases, since we have quite a lot of ways to detect faulty transfers, it can produce false positives in rare cases: When the jiffies value changes right between two interrupt that are in time, we wrongly assume that we missed one or more clock cycles. (2) Some gpio controllers (e.g. the one in the bcm283x chips) may generate spurious IRQs when processing interrupts in the frequency given by PS2 devices. Both issues can be fixed by using ktime resolution for IRQ timekeeping. However, it is still possible to miss clock cycles without detecting them. When the PS2 device generates the falling edge of the clock signal we have between ~30us and 50us to sample the data line, because after this time we reach the next rising edge at which the device changes the data signal already. But, the only thing we can detect is whether the IRQ interval is within the given period. Therefore it is possible to have an IRQ latency greater than ~30us to 50us, sample the wrong bit on the data line and still be on time with the next IRQ. However, this can only happen when within a given transfer the IRQ latency increases slowly. ___ ______ ______ ______ ___ \ / \ / \ / \ / \ / \ / \ / \ / \______/ \______/ \______/ \______/ |-----------------| |--------| 60us/100us 30us/50us Signed-off-by:Danilo Krummrich <danilokrummrich@dk-develop.de> Link: https://lore.kernel.org/r/20220215160208.34826-3-danilokrummrich@dk-develop.deSigned-off-by:
Dmitry Torokhov <dmitry.torokhov@gmail.com>
Showing
Please register or sign in to comment