RE: [PATCH] PCI serial card support

[Dugger, Donald D]

>
>There are some serial boards where you have to give N * IO 
>base addresses in order to support all N COMs. Actually there 
>are even more complex cards than this one. Some supporting 
>much higher than 921600 as maximum speed. Some needing even 
>different kinds of formulas in order to get proper baud 
>rates... But I think this should be good short term solution. 
>More complex cards can have their own driver then.

My goal was to support 1 port to be used as the serial console, I don't see a 
great need for grub to support all the ports on a multi-port card, leave that 
to the OS that gets booted.

Of course, it gets worse, I believe there are some cards that use a 
non-standard I/O register layout, if we want to support those it will indeed 
require a special driver.

>
>Thou "base" could be something different. As this is advanced 
>feature it could be something more descriptive and harder to 
>type... in example divbase, or maxbaud... Of course if we can 
>detect some cards automatically we can store some table for 
>those. For detection of PCI cases we could use PCI ID's.
>
>> +  return ((base << 4) + (speed << 3)) / (speed << 4);
>
>Interesting formula. However. Whats the big gain against base/speed ?
>Perhaps I am missing something? This formula just gives +0.5 
>(rounding?) for resulting value which is then rounded down 
>with integer maths.
>

I stole that formula from the Linux PCI serial driver.  My analysis matches 
yours, it only causes a round up of 1 for some cases where the base baud is not 
a multiple of 2.  Unfortunately, if you look at all the base bauds in the Linux 
driver, there are some weird ones where the round up will change the divisor so 
I would assume this formula is necessary.

--
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
address@hidden
Ph: (303)443-3786