On 6/22/20 2:21 AM, Andrew Jeffery wrote: > On Wed, 17 Jun 2020, at 13:11, Philippe Mathieu-Daudé wrote: >> Hi Andrew, >> >> On 6/17/20 3:18 AM, Andrew Jeffery wrote: >>> On Tue, 16 Jun 2020, at 17:21, Philippe Mathieu-Daudé wrote: >>>> The current implementation uses nano-second precision, while >>>> the watchdog can not be more precise than a micro-second. >>> >>> What's the basis for this assertion? It's true for the AST2500 and AST2600, >>> but >>> the AST2400 can run the watchdog from either a 1MHz clock source or the APB >>> clock (which must be at least 16.5MHz on palmetto). The reset state on the >>> AST2400 configures the watchdog for the APB clock rate. >>> >>> The Linux driver will eventually configure the watchdog for 1MHz mode >>> regardless so perhaps the AST2400 reset state is a bit of a corner case, but >>> I feel the assertion should be watered down a bit? >> >> What about this description? >> >> "The current implementation uses nano-second precision, but >> is not more precise than micro-second precision. >> Simplify by using a micro-second based timer. >> Rename the timer 'timer_us' to have the unit explicit." > > So is this a limitation of QEMUTimer? I was establishing that the hardware > can > operate at greater than 1 micro-second precision.
No, I misread your comment about the AST2400 timer which can run at more than 1Mhz. The QEMUTimer doesn't have a such limitation; this patch aimed to simplify the code for reviewers, but you proved it incorrect, so let's disregard it. Thanks for your careful review! > > Andrew >