[Top] [All Lists]

Re: SGI XFS on ppc

To: Daniel Moore <dxm@xxxxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: SGI XFS on ppc
From: Steve Lord <lord@xxxxxxx>
Date: Wed, 02 Aug 2000 20:43:54 -0500
Cc: thomas.graichen@xxxxxxxxxxxxx, linux-xfs@xxxxxxxxxxx
In-reply-to: Your message of "Thu, 03 Aug 2000 09:49:01 +1000
Sender: owner-linux-xfs-announce@xxxxxxxxxxx
> Steve Lord writes:
>  => > ok - i have found the place where it hangs (by the help of lots of
>  => > printk's and at the end also delays :-) - it loops in xfs_xlatesb
>  => > through the while loop taking the "if (arch == ARCH_NOCONVERT" and
>  => > the "if (dir>0) {" (not the "else" one in both cases) ... does this
>  => > help you in any way ? - should i print out any variables there ?
>  => My cheap answer is we wait for the Australians who did the endian
>  => conversion
>  => work to come in to work.  A big endian version of this code has booted bu
>  => any changes this required have not been put back into the tree, I suspect
>  => something in this area, since this code is dealing with byte swapping
>  => and should be a noop in your case.
> We're awake now but soon to be in meetings.
> Stubbing out xfs_xlatesb is fine in the mount, but later on it gets
> used to copy selected fields between two superblocks, so you'll need
> it to work.
> When you say it loops in xfs_xlatesb, do you mean it loops _forever_?
> Because it should loop quite a few times (once for each field) but
> obviously should terminate.

This was a loops forever - i.e. get bored waiting for it, there is no
kdb on this architecture, so debugging is more interesting! But yes,
this needs to work in the long run (tomorrow?).


> If it loops forever, I suspect there must be an endian issue with
> xfs_lowbit64. Could you print out the value "f" returned by this
> each time around and let me know what you get? You should get
> "0", "1", "2"... "42" and then the function should exit
> The only way I can see it getting stuck is if xfs_lowbit64 has issues
> and ends up returning the wrong number, never letting the while terminate.
> You'll need xfs_lowbit64 to be correct for everything to work, so it
> would be worth checking it out...
> Let me know how you get on.
> -----------------------------------------------------
>  Daniel Moore                  dxm@xxxxxxx
>  R&D Software Engineer         Phone: +61-3-98348209
>  SGI Performance Tools Group   Fax:   +61-3-98132378
> -----------------------------------------------------

<Prev in Thread] Current Thread [Next in Thread>