3, 4, 5 already in the CVS tree.
YH
-----邮件原件----- 发件人: Hendricks David W. [mailto:dwh@lanl.gov] 发送时间: 2004年4月23日 14:52 收件人: YhLu 抄送: Li-Ta Lo; ron minnich; Stefan Reinauer; LinuxBIOS 主题: Re: ´ð¸´: ATI rage xl.init
The only thing that didn't patch cleanly on my setup was a config file for a Tyan board. Was there anything else that didn't apply cleanly to a CVS tree, Ollie? It might be nice to get that ATi and btext stuff in there quickly since there are likely others who are very, very interested in this patch. Though I know Ron wanted to review the superIO and PNP changes a little more after finishing his paperwork :-/
On Fri, 23 Apr 2004, YhLu wrote:
So I need to produce another patch except 3, 4 & 5?
-----ÓʼþÔ¼þ----- ·¢¼þÈË: Li-Ta Lo [mailto:ollie@lanl.gov] ·¢ËÍʱ¼ä: 2004Äê4ÔÂ23ÈÕ 11:00 ÊÕ¼þÈË: YhLu ³ËÍ: ron minnich; Stefan Reinauer; LinuxBIOS Ö÷Ìâ: Re: ATI rage xl.init
On Fri, 2004-04-23 at 11:56, YhLu wrote:
You only commit 4 & 5 ?
I haven't commit anything yet. I am busy working on the ibm/e325 HT enumeration problem. I will look at the patch next week.
Ollie
-----ÓʼþÔ¼þ----- ·¢¼þÈË: Li-Ta Lo [mailto:ollie@lanl.gov] ·¢ËÍʱ¼ä: 2004Äê4ÔÂ21ÈÕ 14:47 ÊÕ¼þÈË: YhLu ³ËÍ: ron minnich; Stefan Reinauer; LinuxBIOS Ö÷Ìâ: Re: ATI rage xl.init
On Tue, 2004-04-20 at 19:38, YhLu wrote:
Ron,
Please commit ati rage xl support.
- ati rage xl support
- add btext_console to use rage xl fb.
- enable pnp init calling
- superio Winbond: real enable/disable device.
- superio Winbond: enable HW
- Tyan S2850 support update
- Tyan S2875 support added.
- Move some PCI master enable driver to /driver dir.
Stefan,
Please test the filo too, I add the btext_console into it. On vocation? Why using stepan@openbios.org ?
there are some conflicts between your tree and the current CVS. Could you do cvs update and send me the patch again ?
Ollie
Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
On Fri, 23 Apr 2004, YhLu wrote:
3, 4, 5 already in the CVS tree.
so 1,2, should go in now I think, and 6,7, clearly go in?
ron
* ron minnich rminnich@lanl.gov [040424 00:35]:
On Fri, 23 Apr 2004, YhLu wrote:
3, 4, 5 already in the CVS tree.
good, then things should be complete now.
so 1,2, should go in now I think, and 6,7, clearly go in?
They're in now. I also included 8. YhLu, are they needed?
Stefan
8.
Sometimes, if you not enable PCI master, the device can not work properly in Linux.
But Eric/Ron didn't enable that in PCI main loop, because some devices don't need that.
Regards
YH
-----邮件原件----- 发件人: Stefan Reinauer [mailto:stepan@openbios.org] 发送时间: 2004年4月24日 16:39 收件人: ron minnich 抄送: YhLu; Hendricks David W.; Li-Ta Lo; LinuxBIOS 主题: Re: ??????: ??????: ATI rage xl.init
* ron minnich rminnich@lanl.gov [040424 00:35]:
On Fri, 23 Apr 2004, YhLu wrote:
3, 4, 5 already in the CVS tree.
good, then things should be complete now.
so 1,2, should go in now I think, and 6,7, clearly go in?
They're in now. I also included 8. YhLu, are they needed?
Stefan
"Yinghai Lu" yhlu@tyan.com writes:
Sometimes, if you not enable PCI master, the device can not work properly in Linux.
But Eric/Ron didn't enable that in PCI main loop, because some devices don't need that.
As a general rule devices don't need that and it is the responsibility of the device driver to enable/os to enable it.
So the master bit should only be enabled for devices that are needed to boot. So we do need to enable this on the ioapic that transmit interrupts over the pci bus for example. Both because it is needed to boot and because it is architectural hardware.
Unless someone can show me a relevant passage of the pci spec, that says differently.
Any linux device driver that can't cope with a device that does not have it's bus master bit clear and needs it set is broken.
Devices with bus master set are allowed to DMA into main memory whenever they feel like it so it is dangerous to enable that bit indiscriminately.
Eric
I will double check if current kernel need that for every onboard device.
YH
-----邮件原件----- 发件人: Eric W. Biederman [mailto:eric@lnxi.com] 代表 Eric W. Biederman 发送时间: 2004年4月25日 11:47 收件人: Yinghai Lu 抄送: 'Stefan Reinauer'; 'ron minnich'; 'Hendricks David W.'; 'Li-Ta Lo'; 'LinuxBIOS' 主题: Re: ATI rage xl.init
"Yinghai Lu" yhlu@tyan.com writes:
Sometimes, if you not enable PCI master, the device can not work properly
in
Linux.
But Eric/Ron didn't enable that in PCI main loop, because some devices
don't
need that.
As a general rule devices don't need that and it is the responsibility of the device driver to enable/os to enable it.
So the master bit should only be enabled for devices that are needed to boot. So we do need to enable this on the ioapic that transmit interrupts over the pci bus for example. Both because it is needed to boot and because it is architectural hardware.
Unless someone can show me a relevant passage of the pci spec, that says differently.
Any linux device driver that can't cope with a device that does not have it's bus master bit clear and needs it set is broken.
Devices with bus master set are allowed to DMA into main memory whenever they feel like it so it is dangerous to enable that bit indiscriminately.
Eric
* Eric W. Biederman ebiederman@lnxi.com [040425 20:46]:
As a general rule devices don't need that and it is the responsibility of the device driver to enable/os to enable it.
How do other OSes besides Linux do this? Are they broken as well?
Unless someone can show me a relevant passage of the pci spec, that says differently.
I think we can allow fixes for OS level problems in LinuxBIOS to some small extent, even though fixing the OS is highest priority if the OS is the problem.
Any linux device driver that can't cope with a device that does not have it's bus master bit clear and needs it set is broken.
s/linux//. There's a lot of things broken that we cope with, ie the need for both mptable and pirq table seems like such a thing.
Devices with bus master set are allowed to DMA into main memory whenever they feel like it so it is dangerous to enable that bit indiscriminately.
--> If at all, it should be done with careful checking.
Stefan
On Mon, 26 Apr 2004, Stefan Reinauer wrote:
Devices with bus master set are allowed to DMA into main memory whenever they feel like it so it is dangerous to enable that bit indiscriminately.
--> If at all, it should be done with careful checking.
Here is a possible scenario. You are in linuxbios after somehow falling into it -- i.e., no power up, no reset has happened. Devices for whatever reason are not totally reset, including the ethernet device. The ethernet device is the kind with on-chip buffer queue pointerss. You enable bus master. Chip gets a broadcast packet. It writes it somewhere to memory, over code that won't get run in 2 or 3 weeks. 2 or 3 weeks later, the code is executed.
Your kernel goes kaboom!
How do you figure out what happened? This is a possible scenario that can happen once you start enabling bus master in the bios.
ron
-
ebiederman＠lnxi.com -
ron minnich -
Stefan Reinauer -
YhLu -
Yinghai Lu