Re: SuSE 9.1 packages imported (Linux emulation)

(moving thread to tech-kern)

> I've imported an incomplete set of SuSE 9.1 packages a few minutes ago.
> These package do *not* work stable with NetBSD's Linux emulation.
> Mozilla (taken from the "mozilla-bin" package) crashes in one of two
> attempts to start it.

Okay, I can repeat the problem here. Starting Mozilla-linux several
times makes it eventually start properly and run (apparently) fine.
Appears some file pages may not be loaded properly, leading to spurious
SIGSEGV.  It may be the same problem which has been occurring with
the native threads and native Mozilla some months ago.

I do not recall what was the exact fix to the 'native' Mozilla case.
Perhaps that needs to be extended to handle also address space
shared via clone(2)?

The new libs also use new Linux syscall exit_group(), but that happens
only in fatal signal handling (such as SIGSEGV), and is not causing
the problem AFAICS from kdump.

Jaromir
--

-- 
Jaromir Dolecek <jdolecek <at> NetBSD.org>            http://www.NetBSD.cz/
-=- We should be mindful of the potential goal, but as the Buddhist -=-
-=- masters say, ``You may notice during meditation that you        -=-
-=- sometimes levitate or glow.   Do not let this distract you.''   -=-

Re: SuSE 9.1 packages imported (Linux emulation)

> (moving thread to tech-kern)

Just FYI in case it's not already known problem: The sun-jdk14 also seem to 
have problems with the suse91 libs. (Tested on 2.0_BETA.)

Marc

ata bus speed

Hi,
I'd like to add IOCTLs to retrieve and set the speed status of an ATA device.
I have data defined for PATA and SATA, should I also add something for USB
(or others) ? If so, what ?

-- 
Manuel Bouyer <bouyer <at> antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--

struct atabusiosp_args {
	int type;      /* bus type (PATA, SATA, ...)
#define ATABUSIOSP_TYPE_PATA 0x00
#define ATABUSIOSP_TYPE_SATA 0x01
	union {
		struct _pata {
			/*
			 * current/wanted speed,
			 * -1 = not known/supported
			 */
			int pio_level;
			int dma_level;
			int udma_level;
		} pata;
		struct _sata {
			uint32_t sstatus; /* the sstatus register */
		} sata;
	} speed;
};

IOCTL names [Re: CVS commit: src/sys/sys]

On Mon, Aug 02, 2004 at 07:47:11PM +0000, Manuel Bouyer wrote:
> Rename the ATABUSIO* ioctls from ('A', x) to ('A', 5x), as the former collide
> with the APM ioctl. Reported by Havard Eidnes.

Hi,
is there a policy about IOCTL names, or a registery of the names somewhere ?
When I picked a name for this, I didn't think there could be IOCTLs
definitions outside of src/sys/sys :(

--

-- 
Manuel Bouyer <bouyer <at> antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--

network interface statistics under sysctl

Greetings.
Is it possible to retrieve network interface statistics through the sysctl
interface or kvm is the only way to go ? (I don't really want to develop a kmem
setgid program)

Thanks in advance.

--

-- 
  "Simplicity is the ultimate 
    sophistication." 
    -- Leonardo da Vinci      

Re: network interface statistics under sysctl

phlox wrote:

> Is it possible to retrieve network interface statistics through the sysctl
> interface or kvm is the only way to go ? (I don't really want to develop a kmem
> setgid program)

There is currently no sysctl interface for this, but there are distant
plans to make convert all the current setgid statistics programs to
sysctl one day.  Unfortunately there's no real time line for this...

Simon.
--
Simon Burge                            <simonb <at> wasabisystems.com>
NetBSD Support and Service:         http://www.wasabisystems.com/

Re: network interface statistics under sysctl

When we do get to this, it would be good to try and more or less 
match the standard SNMP MIB in content, if not in name.

	Erik <fair <at> netbsd.org>

load average and xmms

Greetings.
I have a question about load averages and xmms. (I know this is not xmms
specific but I use it as an example) 
On my system (2.0_BETA) xmms uses normally 1% of WCPU and CPU, which makes the
load go above 1.00. With this load the CPU idle percent is above 90% which 
is normal. But the load is high anyway.
The definition I'm aware about the load is based on currently running processes
and their cpu usage, so since xmms uses 1% of cpu why do I get such a high load
?

Thanks in advance.

--

-- 
  "Simplicity is the ultimate 
    sophistication." 
    -- Leonardo da Vinci      

ideas for loadable drivers and hot-plug - PCI and everything

This continues the ideas brought in in the "lkm for pci device" thread
some weeks ago.
(Actually, the basic ideas are much older. I've beem playing with this
since NetBSD-1.3.)

While my implementation examples might look mature - there might be
wrong assumtions from the beginning, so consider them proof-of concept
only for now.

So what needs to be done, or can be done: We want to attach/detach devices
 dynamically at runtime. There are some cases coming to mind:
 a)The driver is already present and the hardware gets added/removed. This
  works well already. Subcases are whether the hardware provides
  some notification on hot-plug events (as USB and PCMCIA) or needs
  to be manually notified (SCSI). Polling would be in-between.
 b)Some hardware is present but not supported by the boot kernel. There
  are some ad-hoc solutions to add drivers (actually there is only the
  "pcilkm" I'm aware of), but this works around the autoconf framework
  and should be replaced by something well-designed.
 c)Hardware gets added, the driver is not yet present but will be loaded
  "on demand". This needs some serious effort to get it "right", in
  particular because the driver matching now done in the kernel (in the
  individual drivers) needs to be duplicated in userland.
 d)A newer driver for some already supported hardware gets available,
  I want to replace the existing one. Needs proper detaching, otherwise
  it is like the second case.

What's dealt here with is case (b) primarily. Case (d) too as far as
the bus drivers keep track of their childs. There are some ideas for

vdevgone() and pending I/O

Hi,
is vdevgone() supposed to wait for pending I/O to complete ?

I have a problem with the wd driver; if I call config_detach() on
it with a dd on the raw device running, I get a kernel trap. I couldn't get
anything usefull from ddb, the trap address is 0 :(

--

-- 
Manuel Bouyer <bouyer <at> antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--