Re: [lwip-devel] core/timers.c with NO_SYS?

[address@hidden]

Hi Leon,

Leon Woestenberg wrote:
> I still don't understand; how can it be we *must* implement a sys_arch
> layer function when NO_SYS is set?
As I said, please don't let the name of that function confuse you. It's 
actually not a sys_arch layer function but a port function. In an 
embedded target, it's enough to set up a hardware counter to tick every 
millisecond and let sys_now() return that counter value. That has 
absolutely *nothing* to do with an OS layer. As such, the actual "bug" 
here, to me, is only a question of which name we give that function. I 
only chose sys_now() because it was already there and I didn't want to 
break old ports unnecessarily (which already might have implemented 
sys_now()).

Aside from that, how do you propose to implement a generic timer 
framework without knowing the time? You do need a port-dependent 
function to get the time for that.
> This makes a lower layer (the core API) depend on an optional higher
> API abstraction layer (sys_arch). This breaks the architecture.
>
> This probably explains why NO_SYS_NO_TIMERS is there.... core/timers.c
> doesn't seem to be a core architecture module this way.
If you see it that way, the architecture is already broken as 
SYS_ARCH_PROTECT() is used in mem.c, memp.c and pbuf.c (and is included 
in many more, where it does not seem to be needed). Mem.c even uses 
semaphores, so I guess it's not a core architecture module, as well. 
However, I think that even for NO_SYS==1 ports, SYS_ARCH_PROTECT may be 
needed when locking application against interrupt level. So to me, this 
is also not a strict OS function, rather a port function that locks 
interrupts. It may be implemented differently when using an OS (i.e. 
locking the scheduler, too), but it is also useful without an OS.

As such, it seems to me that we should sort out our sys.h/sys_arch.h 
handling instead of trying to argument away the need for a function 
returning the time or where it is defined.

BTW: Bug #33544 is also a result of a suboptimal sys.h layout which 
defines some things to u8_t or void for NO_SYS==1.

Simon