[parisc-linux] O_DIRECT on devices

[parisc-linux] O_DIRECT on devices

[Patrick Caulfield]

I'm currently working on LVM2 and we want to use O_DIRECT to write the metadata
from userspace. so I open a partition with O_DIRECT and try to read/write the
data from an aligned buffer.

This works fine on Alpha & Intel but seems very unreliable on parisc for some
reason. Sometimes it works, sometimes it oopses "do_page_fault() pid=2120
command='lvm' type=15 address=0x00000001" and sometimes I just get back the
wrong information (at least LVM thinks its wrong I haven't check in detail just
what is up with it).

I haven't tried writing yet - I'd rather not think what might happen to my disk
just ATM :)

This is a C110 with 256meg of memory running 2.4.18-pa46

patrick

Re: [parisc-linux] O_DIRECT on devices

[Patrick Caulfield]

On Thu, Jul 11, 2002 at 11:09:31AM +0200, Rodolfo Baselli wrote:
> Patrick Caulfield wrote:
> >I'm currently working on LVM2 and we want to use O_DIRECT to write the 
> >metadata
> >from userspace. so I open a partition with O_DIRECT and try to read/write 
> >the
> >data from an aligned buffer.
> 
> Hmmm, so is LVM ready for HP PA-RISC Linux? WOW!

LVM2 works fine - I use it all the time on my HP and Alpha boxes. O_DIRECT is an
optimisation I would like to have in there though cos it avoids a lot of
buffer-cache nastiness when clustering (yes, I'm working on that too but that
bit won't be GPLed)

 
> By the way, I'm reading right now "The catcher in the rye" by J.D. 
> Salinger, and as I glanced at your 2nd name I felt like I was still 
> reading the book! :-)))

:-)


If you were a pop-art expert you might have got even more deja-vu !
http://www.postershop.com/Caulfield-Patrick/Caulfield-Patrick-After-Lunch-2405657.html

I blame my parents...

patrick

Re: [parisc-linux] O_DIRECT on devices

[Matthew Wilcox]

On Thu, Jul 11, 2002 at 09:22:59AM +0100, Patrick Caulfield wrote:
> This works fine on Alpha & Intel but seems very unreliable on parisc for some
> reason. Sometimes it works, sometimes it oopses "do_page_fault() pid=2120
> command='lvm' type=15 address=0x00000001" and sometimes I just get back the
> wrong information (at least LVM thinks its wrong I haven't check in detail just
> what is up with it).

pid 2120 did a NULL pointer dereference ... you should be able to trace
it with the IAOQ, perhaps.  I'm not sure whether anyone's really played
with the O_DIRECT code yet, it could well be buggy on PA.  What you're
seeing sounds like bogus cache behaviour.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] O_DIRECT on devices

[Grant Grundler]

Matthew Wilcox wrote:
> What you're seeing sounds like bogus cache behaviour.

Randolph and I think most SMP bugs reported (and we seen ourselves)
suggest a D-cache problem.

My theory is virtual addresses are flushed on one CPU but any data 
accessed through an aliases on another CPU are not flushed. And then
we end up with an inconsistency.

We've been reading Documentation/cachetlb.txt and trying
to understand what it says about virtually indexed caches.

The other thing is we don't hit the problems with PA8500 - only PA8700.
I'm guessing the aliasing or timing is quite different betweem the two.
Maybe someone else knows more?

grant

Re: [parisc-linux] O_DIRECT on devices

[Matthew Wilcox]

On Mon, Jul 15, 2002 at 01:42:19AM -0600, Grant Grundler wrote:
> Randolph and I think most SMP bugs reported (and we seen ourselves)
> suggest a D-cache problem.

Entirely plausible.  PA has bigger virtually indexed caches than anyone
else, so we're more susceptible to cache aliasing bugs than anyone else.
It could be a missing flush somewhere in the arch-independent code.

> My theory is virtual addresses are flushed on one CPU but any data 
> accessed through an aliases on another CPU are not flushed. And then
> we end up with an inconsistency.

it's certainly possible... i don't claim to understand exactly how this
works, but my recollection is that some of the flush instructions are
cpu-local whereas others are global to the system.  you'd have to ask
jsm about it, really.

> We've been reading Documentation/cachetlb.txt and trying
> to understand what it says about virtually indexed caches.

It seems pretty straightforward to me... am I missing something?

> The other thing is we don't hit the problems with PA8500 - only PA8700.
> I'm guessing the aliasing or timing is quite different betweem the two.
> Maybe someone else knows more?

8700 has bigger caches than 8500 and more TLB entries, so it may be
easier to hit problems.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] O_DIRECT on devices

[Patrick Caulfield]

On Mon, Jul 15, 2002 at 12:29:19PM +0100, Matthew Wilcox wrote:
> On Mon, Jul 15, 2002 at 01:42:19AM -0600, Grant Grundler wrote:
> > Randolph and I think most SMP bugs reported (and we seen ourselves)
> > suggest a D-cache problem.
> 
> Entirely plausible.  PA has bigger virtually indexed caches than anyone
> else, so we're more susceptible to cache aliasing bugs than anyone else.
> It could be a missing flush somewhere in the arch-independent code.

My machine isn't SMP but could a similar bug cause the problem to occur between
the CPU and the DMA controller of the SCSI card ?

patrick

Re: [parisc-linux] O_DIRECT on devices

[Matthew Wilcox]

On Mon, Jul 15, 2002 at 02:57:57PM +0100, Patrick Caulfield wrote:
> My machine isn't SMP but could a similar bug cause the problem to occur between
> the CPU and the DMA controller of the SCSI card ?

you're on a C110, right?  that has a ccio IOC so you shouldn't have any
problems, unless there's a driver bug.  and the symbios driver gets a
lot of testing ;-)

what we could be seeing on your machine is incoherency between user &
kernel space.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] O_DIRECT on devices

[Patrick Caulfield]

On Mon, Jul 15, 2002 at 03:10:13PM +0100, Matthew Wilcox wrote:
> On Mon, Jul 15, 2002 at 02:57:57PM +0100, Patrick Caulfield wrote:
> > My machine isn't SMP but could a similar bug cause the problem to occur between
> > the CPU and the DMA controller of the SCSI card ?
> 
> you're on a C110, right?  that has a ccio IOC so you shouldn't have any
> problems, unless there's a driver bug.  and the symbios driver gets a
> lot of testing ;-)

yes.
 
> what we could be seeing on your machine is incoherency between user &
> kernel space.

ah OK. makes sense.

patrick

[parisc-linux] HP9000/L2000 + FC60 Fiber Support

[António Ribeiro]

Hi,

Can anyone let me know is this cards are supported and they can be attached
to the FC60 from HP.


  Bus 16, device   0, function  0:
    Fibre Channel: PCI device 103c:1028 (Hewlett-Packard Company) (rev 11).
      IRQ 256.
      Master Capable.  No bursts.  Min Gnt=32.
      I/O at 0x20000 [0x200ff].
      I/O at 0x20100 [0x201ff].
      Non-prefetchable 32 bit memory at 0xfffffffff9040000
[0xfffffffff90401ff].
      Non-prefetchable 32 bit memory at 0xfffffffff9000000
[0xfffffffff901ffff].
  Bus 24, device   0, function  0:
    Fibre Channel: PCI device 103c:1028 (Hewlett-Packard Company) (rev 11).
      IRQ 320.
      Master Capable.  No bursts.  Min Gnt=32.
      I/O at 0x30000 [0x300ff].
      I/O at 0x30100 [0x301ff].
      Non-prefetchable 32 bit memory at 0xfffffffff9840000
[0xfffffffff98401ff].
      Non-prefetchable 32 bit memory at 0xfffffffff9800000
[0xfffffffff981ffff].


Thanks.

Antonio

Re: [parisc-linux] HP9000/L2000 + FC60 Fiber Support

[Matthew Wilcox]

On Tue, Aug 13, 2002 at 02:23:49PM +0100, António Ribeiro wrote:
>     Fibre Channel: PCI device 103c:1028 (Hewlett-Packard Company) (rev 11).

according to the latest pci.ids, this is:
Tach TL Fibre Channel Host Adapter

hope this helps someone determine whether or not it's supported.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] HP9000/L2000 + FC60 Fiber Support

[Grant Grundler]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 338 bytes --]

Matthew Wilcox wrote:
> On Tue, Aug 13, 2002 at 02:23:49PM +0100, António Ribeiro wrote:
> >     Fibre Channel: PCI device 103c:1028 (Hewlett-Packard Company) (rev 11).
> 
> according to the latest pci.ids, this is:
> Tach TL Fibre Channel Host Adapter

http://lists.parisc-linux.org/pipermail/parisc-linux/2002-August/017288.html

grant

[parisc-linux] cvs [login aborted]?

[Joel Soete]

Hi pa,

Is it an actual cvs server pb or a local pb I could have:
those last two days cvs -d :pserver:anonymous@cvs.parisc-linux.org:/var/cvs
"failed: Connection timed out"?

Any idea?

Thanks,
    Joel


------------------------------------------------------
Soldes Tiscali ADSL : 27,50 euros/mois jusque fin 2003.
On s'habitue vite à payer son ADSL moins cher!
Plus d'info? Cliquez ici... http://reg.tiscali.be/default.asp?lg=fr 

Re: [parisc-linux] cvs [login aborted]?

[Matthew Wilcox]

On Fri, Jul 18, 2003 at 07:27:43AM +0200, Joel Soete wrote:
> Hi pa,
> 
> Is it an actual cvs server pb or a local pb I could have:
> those last two days cvs -d :pserver:anonymous@cvs.parisc-linux.org:/var/cvs
> "failed: Connection timed out"?

Hm, it's not working for me either:

willy@honeydew:~/kernel/pci-2.5$ cvs up Makefile 
cvs [update aborted]: connect to cvs.parisc-linux.org(192.25.206.14):2401 failed: Connection timed out
willy@honeydew:~/kernel/pci-2.5$ cat CVS/Root 
:pserver:anonymous@cvs.parisc-linux.org:/var/cvs

I think someone must have changed the firewall config because I can
reach port 2401 from localhost, but not from any other machine.

-- 
"It's not Hollywood.  War is real, war is primarily not about defeat or
victory, it is about death.  I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk

Re: [parisc-linux] cvs [login aborted]?

[bame]

> I think someone must have changed the firewall config because I can
> reach port 2401 from localhost, but not from any other machine.

I don't know how that happened, but I reopened the port and anon
CVS should work now.

	-P