2002-02-16 08:52:24

by viking

[permalink] [raw]
Subject: What? Clock Slowdown Again?


==== Problem One ====
Hi all. I started noticing the "clock slowdown" problem when I started with
2.4.x series of kernels, and described it in a past post. In summary, under
framebuffer mode, when the interrupts are disabled, they don't get re-enabled
quickly enough to make sure the system clock keeps up. I was told about a
patch that would help to solve this, I applied said patch (2.4.3) then, when
the kernel updated to 2.4.11, I was pleased to see the Andrew Morton patch
integrated into the kernel.

However, I'm noticing the slowdown again, though not to the same extent. I
took the kernel up to 2.4.17 but I need to do some further tests to see what
else I can ferret out.

==== Problem Two ====

Also, what are the plans for replacing the virt_to_bus / bus_to_virt
functions? The sourcecode basically says to use pci_map functions, but, being
almost a kernel illiterate (I know how to compile, apply patches, and not much
else) I don't know how to get started on converting. I thought that
pci_resource_{start,end,len} may also be needed too. Am I right?

Again, thanks for all your time. I read KT and found out about why my 2.5.4
kernel refused to compile, fixed that, fixed another couple of things, then
came up against this bus_to_virt problem, which I'd like to get fixed before I
go anywhere with 2.5 - understandable as I just aout live for my 160x64 text
screen 8-) I use a VESA framebuffer, as the SiS framebuffer code hasn't been
extended to include the 5597/5598 yet.

Must go...now I can FINALLY send this off, six months after originally
composeng it.

Regards, The Viking
(Please, email me as well, the list takes a VERY long time to read, and I
might miss your post...)

--
/| _,.:*^*:., |\ Cheers from the Viking family,
| |_/' viking@ `\_| | including Pippin, our cat
| flying-brick | $FunnyMail Bilbo : Now far ahead the Road has gone,
\_.caverock.net.nz_/ 5.39 in LOTR : Let others follow it who can!







2002-02-16 21:44:55

by Francois Romieu

[permalink] [raw]
Subject: Re: What? Clock Slowdown Again?

Eric Gillespie <[email protected]> :
[...]
> Also, what are the plans for replacing the virt_to_bus / bus_to_virt
> functions? The sourcecode basically says to use pci_map functions, but, being
> almost a kernel illiterate (I know how to compile, apply patches, and not much
> else) I don't know how to get started on converting. I thought that
> pci_resource_{start,end,len} may also be needed too. Am I right?

Look at Documentation/DMA-mapping.txt, section "Types of DMA mappings"
Read it again from the start. If you aren't familiar with PCI,
http://ww.google.com/search?hl=en&q=PCI+dual+address+cycle+explanation may
help. Then look at drivers/net/{epic/eepro100/3c59x}.c and grep for the
functions quoted in DMA-mapping.txt. After that, glue all the pieces in
your head.

--
Ueimor