Skip to contentSkip to conversation

TODO list

Linux-ia64 Don Dugger 5 participants 11 messages Feb 9, 2001 – Feb 17, 2001

lore
Share & tools
lore

Don Dugger

author
#0 lore raw

All-

As promised, here is the Linux IA64 To-Do list, current as of 2/9/01. The pretty printed HTML version is always available at `http://www.linuxia64.org/todo.html/'. Note that there are a few things that no one has signed up to do yet. If you are interested in taking responsibility for:

EEPro100
Hot plug SCSI
GLIBC unwind API
setjmp cleanup

just let me know (here is your chance for fame and fortune :-)

Linux IA64 To-Do List
Februrary 9, 2001

Kernel

Task
Owner
Details

Unaligned handler
David Mosberger - DONE
Works for user procs, need support for kernel

CS4281 Sound
Thomas Woller
Should be OK, requires updated MPG128 and/or XMMS

EEPro100
?
Hangs on >180MB xfers

Hot plug SCSI
?
Hotplugging doesn't crash the kernel but the kernel doesn't detect the change.

Other HW Support

UNIX98 test suite
Doug Beattie
Fix test suite failures, need simple test cases

CPU state on execve
David Mosberger
Clear CPU state to avoid info leak to new program. Alan Cox thinks it would be sufficient to do this when current->dumpable is zero

AGP Support
Chris Ahna

DRI support
Chris Ahna
Need DRI module for ATI Rage128

llseek
Erich Focht - DONE
Need llseek for fs/proc/base.c:proc_mem operations. Patch submitted.

Libraries

Task
Owner
Details

gprof support
Jes Sorensen
Most of the GLIBC bits should be there, hope to get the M4 autoconf done soon.

GLIBC unwind API
?
Define API that works with signal handling, probably
starting from the kernel unwinder.
Implement this API
Contact David Mosberger for details.

setjmp cleanup
?
Bring setjmp into compliance with SW conventions
(flushrs must be done on a setjmp)
Remove extraneous loadrs in __longjmp. Contact David Mosberger for details.

Applications

Task Owner Details

gdb
Kevin Buettner
NaT bits fail
pthread support
coredump support broken
inferior function calls fail
HW watchpoints fail

strace
Don Dugger
new system calls added
strace -f hangs on a fork

Signature · 4 lines
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
[email protected]
Ph: 303/938-9838
#1 lore raw

Do we want to support epc-based system calls? I know some other IA-64 OS uses that mechanism. I'm not sure how much perfermance gain we can get by that in Linux, though.

Quoted 134 lines · Don Dugger

All-

As promised, here is the Linux IA64 To-Do list, current as of 2/9/01. The pretty printed HTML version is always available at `http://www.linuxia64.org/todo.html/'. Note that there are a few things that no one has signed up to do yet. If you are interested in taking responsibility for:

EEPro100
Hot plug SCSI
GLIBC unwind API
setjmp cleanup

just let me know (here is your chance for fame and fortune :-)

Linux IA64 To-Do List
Februrary 9, 2001

Kernel

Task
Owner
Details

Unaligned handler
David Mosberger - DONE
Works for user procs, need support for kernel

CS4281 Sound
Thomas Woller
Should be OK, requires updated MPG128 and/or XMMS

EEPro100
?
Hangs on >180MB xfers

Hot plug SCSI
?
Hotplugging doesn't crash the kernel but the kernel doesn't detect the change.

Other HW Support

UNIX98 test suite
Doug Beattie
Fix test suite failures, need simple test cases

CPU state on execve
David Mosberger
Clear CPU state to avoid info leak to new program. Alan Cox thinks it would be sufficient to do this when current->dumpable is zero

AGP Support
Chris Ahna

DRI support
Chris Ahna
Need DRI module for ATI Rage128

llseek
Erich Focht - DONE
Need llseek for fs/proc/base.c:proc_mem operations. Patch submitted.

Libraries

Task
Owner
Details

gprof support
Jes Sorensen
Most of the GLIBC bits should be there, hope to get the M4 autoconf done soon.

GLIBC unwind API
?
Define API that works with signal handling, probably
starting from the kernel unwinder.
Implement this API
Contact David Mosberger for details.

setjmp cleanup
?
Bring setjmp into compliance with SW conventions
(flushrs must be done on a setjmp)
Remove extraneous loadrs in __longjmp. Contact David Mosberger for details.

Applications

Task Owner Details

gdb
Kevin Buettner
NaT bits fail
pthread support
coredump support broken
inferior function calls fail
HW watchpoints fail

strace
Don Dugger
new system calls added
strace -f hangs on a fork
--
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale [email protected]
Ph: 303/938-9838

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64

Signature · 4 lines
Jun U Nakajima
Core OS Development
SCO/Murray Hill, NJ
Email: [email protected], Phone: 908-790-2352 Fax: 908-790-2426
#2 lore raw
In reply to Don Dugger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Jun Nakajima

Jun> Do we want to support epc-based system calls? I know some other Jun> IA-64 OS uses that mechanism. I'm not sure how much perfermance Jun> gain we can get by that in Linux, though.

That's certainly something I'm interested in investigating. It's not nearly as obvious a win as it may seem at first, so it's something that needs to be looked at carefully.

In any case, this would be something for 2.5.x, so I don't think we should clutter the current TODO list with such things (there are LOTS of other performance related things we can do...).

	--david
author
#3 lore raw
In reply to Don Dugger

It also isn't as easy to do as you might think. We were looking into this at one time and you have to be really carefull about stack management because you can wind up running in kernel mode with a user stack pointer. This would cause serious problems when an interrupt comes in.

Quoted 20 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Jun Nakajima

Jun> Do we want to support epc-based system calls? I know some other Jun> IA-64 OS uses that mechanism. I'm not sure how much perfermance Jun> gain we can get by that in Linux, though.

That's certainly something I'm interested in investigating. It's not nearly as obvious a win as it may seem at first, so it's something that needs to be looked at carefully.

In any case, this would be something for 2.5.x, so I don't think we should clutter the current TODO list with such things (there are LOTS of other performance related things we can do...).

	--david

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64

Signature · 4 lines
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
[email protected]
Ph: 303/938-9838
#4 lore raw
In reply to Don Dugger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Don Dugger

Don> It also isn't as easy to do as you might think. We were Don> looking into this at one time and you have to be really Don> carefull about stack management because you can wind up running Don> in kernel mode with a user stack pointer. This would cause Don> serious problems when an interrupt comes in.

While I agree with the sentiment, I'd phrase it a little differently:
it's not difficult to fix the assumption of "psr.cpl=0 => we're on the kernel stack". The difficulty is in proving that you fixed all the code that made that assumption...

	--david
#5 lore raw
In reply to Don Dugger

On Fri, Feb 09, 2001, Don Dugger <[email protected]> wrote:
;
; Linux IA64 To-Do List
; Februrary 9, 2001
;
; EEPro100
; ?
; Hangs on >180MB xfers

Is this the same issue as the NFS transfer problem or is this something else?

Signature · 3 lines
No Kernel Hackers were harmed during writing of this email.

                                                -Dragan
author
#6 lore raw
In reply to Don Dugger

Dragan-

I can't comment as I have not been able to duplicate this problem on my BigSur. Maybe someone who is seeing this problem could create a 200MB file of zeros and see if the problem persists (hint, hint).

Quoted 21 lines · Dragan Stancevic

On Fri, Feb 09, 2001, Don Dugger <[email protected]> wrote:
;
; Linux IA64 To-Do List
; Februrary 9, 2001
;
; EEPro100
; ?
; Hangs on >180MB xfers

Is this the same issue as the NFS transfer problem or is this something else?

Signature · 8 lines
No Kernel Hackers were harmed during writing of this email.

                                                -Dragan

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64
Signature · 4 lines
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
[email protected]
Ph: 303/938-9838
#7 lore raw
In reply to Don Dugger

I copied a 300MB of zeros to a Lion via nfs using an eepro100. No problems. This is with memory restricted to 1GB.

Quoted 18 lines · Don Dugger

Dragan-

I can't comment as I have not been able to duplicate this problem on my BigSur. Maybe someone who is seeing this problem could create a 200MB file of zeros and see if the problem persists (hint, hint).

Quoted 10 lines · Dragan Stancevic

On Fri, Feb 09, 2001, Don Dugger <[email protected]> wrote:
;
; Linux IA64 To-Do List
; Februrary 9, 2001
;
; EEPro100
; ?
; Hangs on >180MB xfers

Is this the same issue as the NFS transfer problem or is this something else?

Signature · 3 lines
Mike Madore
Software Engineer
TurboLinux, Inc.
#8 lore raw
In reply to Don Dugger

On Fri, Feb 16, 2001, Michael Madore <[email protected]> wrote:
; I copied a 300MB of zeros to a Lion via nfs using an eepro100. No problems. This ; is with memory restricted to 1GB.

It won't work with zeros, there has to be the right data in the right places of the datagram which when multiplexed has to produce a fragment withouth the UDP header and if by chance the right bits are on in the right places the card will confuse it for a TCO message which is transperent of the driver.

Signature · 3 lines
No Kernel Hackers were harmed during writing of this email.

                                                -Dragan
author
#9 lore raw
In reply to Don Dugger

Dragain-

The purpose of the test was to see if we had two problems on Lion, a size issue and a pattern issue. Mike's test indicates that the Lion only has a pattern problem, the size of the transfer is not an issue.

Quoted 19 lines · Dragan Stancevic

On Fri, Feb 16, 2001, Michael Madore <[email protected]> wrote:
; I copied a 300MB of zeros to a Lion via nfs using an eepro100. No problems. This ; is with memory restricted to 1GB.

It won't work with zeros, there has to be the right data in the right places of the datagram which when multiplexed has to produce a fragment withouth the UDP header and if by chance the right bits are on in the right places the card will confuse it for a TCO message which is transperent of the driver.

Signature · 8 lines
No Kernel Hackers were harmed during writing of this email.

                                                -Dragan

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64
Signature · 4 lines
Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
[email protected]
Ph: 303/938-9838
#10 lore raw
In reply to Don Dugger

On Fri, Feb 16, 2001, Don Dugger <[email protected]> wrote:

Oh, I see, Dan :^)

; Dragain-
;
; The purpose of the test was to see if we had two problems on Lion, ; a size issue and a pattern issue. Mike's test indicates that the Lion ; only has a pattern problem, the size of the transfer is not an issue.

Signature · 3 lines
No Kernel Hackers were harmed during writing of this email.

                                                -Dragan