[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [MiNT] Firebee Coldfire MiNT Kernel and NFS
Good observation, I guess we come closer.
Not sure if it’s the root cause, but at least I have a hint: on the Coldfire, DMA (which is used for FEC transfers), CPU and PCI compete for XL bus bandwidth. Bus priority is scheduled by the XL arbiter and can be soft-configured with different priorities. By default (and _I think_ FireTOS doesn’t touch this defaults), all three bus masters have equal priority.
I can easily imagine that DMA runs into bandwidth problems if the CPU is hogging the XL bus. This could explain why you see improved transfer rates during screen activity: in this situation, the CPU transfers data to/from video RAM (which is located on the FlexBus and much slower than SDRAM), with video RAM accesses forcing wait states to the CPU.
This will free XL bus bandwidth in favour of DMA.
You might remember we’ve seen higher network throughput with the EmuTOS BaS-based network drivers? BaS reverts XL bus priorities, giving DMA transfers highest.
I’ve done this after I saw that DMA memory to memory transfers were faster on the m5484 LITE board than on the Firebee despite its lower clock rate.
Does throughput increase also if you have screen output on EmuTOS?
Am 01.09.2014 um 22:32 schrieb Vincent Rivière <email@example.com>:
> On 01/09/2014 22:04, Vincent Rivière wrote:
>> I begin to wonder if FireTOS and its internal multitasking (FreeRTOS,
>> etc...) could not be part of the problem...
> New approach: EmuTOS on the FireBee.
> I used Markus's BaS_GCC, mcdcook.prg and modified fec.xif.
> Good news: the network works (almost) as well as on FireTOS. I can use OpenSSH and Samba without trouble. Everything 100% ColdFire :-)
> Bad news: NFS still does not work. I get the exact same results with EmuTOS than on FireTOS.
> 1) There is nothing wrong in FireTOS. The situation is the same with EmuTOS.
> 2) NFS is much faster when there is some screen activity. It is not related to mouse interrupts, as I initially thought.
> How can screen activity speed up NFS???
> Probably something in deep multitasking routines...
> And what does NFS try to do when it is stuck?
> Vincent Rivière