[microblaze-uclinux] Re: Re: [uClinux-dev] Regarding the xmbserial driver

[anand_12 <anand_12@xxxxxxxxxxxx>]

hi john,

 thanks for your prompt clarification again..yeah!!..a better thing would be to fix the bug itself rather than
find workarounds...:-)..

Cheers

V.Anand
(www.tenet.res.in)













Hi Anand,
>
>anand_12 wrote:
>
>> I am running uclinux on a spartan 2E microblaze board.
>> Sometime back I had posted to the groups regarding a problem
>> I was facing.The microblaze-uclinux system used to crash(atleast apparently) if the user typed commands or text at a very fast rate at the shell prompt over the serial console(hyperterm in my case). 
>
>This is a known bug in the UARTlite driver (xmbserial.c)
>
>> As suspected,the slow response of the serial driver was because of some buffering issue.The interrupt service routine invokes a "receive_chars" function in its handler..
>> that function basically receives every character and puts it into a buffer(if i got it right)..called a "flip buffer"..which as shown below
>> (uclinux-dist/linux-2.4.x/include/linux/tty.h)
>> 
>> has a size # defined to 512 bytes...i changed this to 1024 bytes(also shown below)..and now the problem is solved..I am able to type and issue commands at a very fast rate without the system crashing...
>> 
>> the problem seems to be solved but i do need to look at the implications of having increased the size of this buffer..but as it is..there is big difference now when i issue commands or just type some text and a very fast rate..it does not crash at all!!...
>
>This is probably not an acceptable fix, because that header file is 
>global to all uClinux targets.  There's nothing stopping you or anyone 
>else applying this fix to their local source tree, but I wouldn't try to 
>push it into the CVS.
>
>Much better would be for us to identify the source of the bug in 
>xmbserial (we've got some good ideas) and fix it there.
>
>The problem seems to be related to a code path in the interrupt handler 
>of xmbserial.c - from memory if the flip_buffer is full, the handler 
>exits without stripping the incoming chars in the hardware buffer.  This 
>causes an endless interrupt loop because the interrupt is still 
>asserted, but there's no room in the filp_buf.  Or something like that.
>
>Unfortunately this issue is intertwined with a general unresolved 
>question about the behaviour of the interrupt controller regarding 
>level-sensitive interrupts.  There was much discussion about this on the 
>microblaze list a while back, a search of the archives should track it down.
>
>Cheers,
>
>John
>
****************************************


___________________________
microblaze-uclinux mailing list
microblaze-uclinux@itee.uq.edu.au
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/