[lc-devel] 2.4.16-0.20pre2 NOT working under UML

[Rodrigo S. de C. <rc...@im...>]

Hi,

The 2.4.16-0.20pre2 version is not currently working under User Mode
Linux (at least not up to 2.4.16-2um). There's a problem regarding
ptes and the virtual swap addresses used by our code.  We don't know
why some bits are not correctly updated, what messes up everything
when a pte faults a not updated address. Check the email I sent to
user-mode-linux-devel mailing list below (no reply so far).

If somebody feels like digging deep into this problem, it would be
very helpful. I don't know very much of UML internals, so I don't know
exactly what's going on there with respect to pte, tlb, etc.

----- Forwarded message from Rodrigo Souza de Castro <rc...@im...> -----

Date: Wed, 12 Dec 2001 12:07:15 -0200
From: Rodrigo Souza de Castro <rc...@im...>
Subject: [uml-devel] swap entry and newpage bit
To: use...@li...
Delivered-To: rc...@im...

Hi,

[Jeff, I talked to you yesterday about this problem]

In my code, I use a virtual swap address, which is nothing but an
ordinary swap entry, but its type is MAX_SWAPFILES. This address is
assigned to a pte in try_to_swap_out(), like an usual swap entry and
that's _all_ (there some other stuff wrt to these ptes, but I
removed). It should work as usual swap addresses do work, right? But
it does not, at least under UML (on a real machine it does).

The problem I noticed is that, when setting a pte to a swap entry, the
NEWPAGE bit is turned on to tell fix_range() that this address should
be unmapped. It turns out that, for swap adresses that are shared by
many ptes, the address is not unmapped and the NEWPAGE bit is not
turned off soon enough before any pte faults. Then, in fault handlers
(handle_pte_fault() and do_swap_page()), the pte has a wrong value, so
I can't find the page in swap cache. For example, if the swap cache
page is set to 0x3307c and the wrong pte is set to 0x3307e (due to
newpage bit), I won't be able to find it.

In _switch_to, everytime we swith to a new task, you flush its tlb
(flush_tlb_all()), so there shouldn't have this problem. An important
question before bother you more: do you think that it's UML-related or
it's unlikely?

I added a printk to fix_range when fixing my virtual swap entries and
the problem was gone. Will it be a race condition? I am not doing
anything but setting the pte in try_to_swap_out and handling faults in
do_swap_page(), so I have no clue. :-(

The same code was working in the old 2.4.10, when the newpage bit was
not set for swap entries. 

Regards,
-- 
Rodrigo S. de Castro <rc...@im...>


_______________________________________________
User-mode-linux-devel mailing list
Use...@li...
https://lists.sourceforge.net/lists/listinfo/user-mode-linux-devel



----- End forwarded message -----

-- 
Rodrigo S. de Castro <rc...@im...>