Grant Tables & mm.c

Grant Tables & mm.c

[Koripella Srinivas]

[-- Attachment #1.1: Type: text/plain, Size: 529 bytes --]

Hi,
I am using grant tables to set up a shared memory between two guests.
One problem i am facing is that as soon as the shared memory is set up one of the domain crashes with the log message 
 "Attempt to implicitly unmap a granted PTE".
the message is from function "put_page_from_l1e" in mm.c(arch/x86).

what could be the reason? Any pointers would be of great help






		
__________________________________________________________
Yahoo! India Answers: Share what you know. Learn something new
http://in.answers.yahoo.com/

[-- Attachment #1.2: Type: text/html, Size: 778 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Grant Tables & mm.c

[Keir Fraser]

[-- Attachment #1.1: Type: text/plain, Size: 977 bytes --]

You must explicitly unmap grants. So the domain which uses
GNTTABOP_map_grant_ref must subsequently unmap it with
GNTTABOP_unmap_grant_ref. You cannot just zap the pte!

 -- Keir

On 4/7/07 11:16, "Koripella Srinivas" <talkwithsrinivas@yahoo.co.in> wrote:

> Hi,
> I am using grant tables to set up a shared memory between two guests.
> One problem i am facing is that as soon as the shared memory is set up one of
> the domain crashes with the log message
>  "Attempt to implicitly unmap a granted PTE".
> the message is from function "put_page_from_l1e" in mm.c(arch/x86).
> 
> what could be the reason? Any pointers would be of great help
> 
> 
> 
> 
>  Here¹s a new way to find what you're looking for - Yahoo! Answers
> <http://us.rd.yahoo.com/mail/in/yanswers/*http://in.answers.yahoo.com/>
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel



[-- Attachment #1.2: Type: text/html, Size: 1805 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Grant Tables & mm.c

[Mark Williamson]

> I am using grant tables to set up a shared memory between two guests.
> One problem i am facing is that as soon as the shared memory is set up one
> of the domain crashes with the log message "Attempt to implicitly unmap a
> granted PTE".
> the message is from function "put_page_from_l1e" in mm.c(arch/x86).
>
> what could be the reason? Any pointers would be of great help

This means that something is attempting to zero the PTE pointing to the 
granted page without doing an explicit grant unmap hypercall; this is not 
allowed by Xen.

Do you have any idea what might be causing this?  Are you mapping pages to 
userspace?  Linux tends to zap userspace page mappings quite cheerfully 
unless you do something to stop it, could this be your problem?

Cheers,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!

Re: Grant Tables & mm.c

[Jeremy Fitzhardinge]

Keir Fraser wrote:
> You must explicitly unmap grants. So the domain which uses 
> GNTTABOP_map_grant_ref must subsequently unmap it with 
> GNTTABOP_unmap_grant_ref. You cannot just zap the pte!

Does that just apply to the last pte, or any pte?

    J

Re: Grant Tables & mm.c

[Keir Fraser]

On 4/7/07 19:42, "Jeremy Fitzhardinge" <jeremy@goop.org> wrote:

>> You must explicitly unmap grants. So the domain which uses
>> GNTTABOP_map_grant_ref must subsequently unmap it with
>> GNTTABOP_unmap_grant_ref. You cannot just zap the pte!
> 
> Does that just apply to the last pte, or any pte?

Any pte that was created via GNTTABOP_map_grant_ref.

 -- Keir

Re: Grant Tables & mm.c

[Mark Williamson]

> " Linux tends to zap userspace page mappings quite cheerfully
> unless you do something to stop it, could this be your problem?"
>
> What does this mean exactly??

Linux "zaps" mappings when it removes them from pagetables.  Really this 
corresponds to zeroing a pte corresponding to that mapping.

Unless you do something to stop the Linux mm system from treating the grant 
table mappins like "normal" RAM, it will eventually want to simply zero them.  
This will break for grant mappings, since it is not a permitted means of 
unmapping granted memory.

> Any way i can prevent it from happening?

It depends on what you're trying to do...

Have you looked at the userspace grantables driver in our Linux tree?  This 
provides a device for mapping granted memory safely into userspace, handling 
cleanup correctly.  Even if that's not directly useful, maybe you can find 
some code in there that'll help you.

Cheers,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!

Re: Grant Tables & mm.c

[Koripella Srinivas]

[-- Attachment #1.1: Type: text/plain, Size: 1788 bytes --]

I was looking at the user space grant tables but i did not find any code calling the function
"remap_pfn_range" to map the kmalloc ed allocated memory into user space.
What could be the reason?? Or am i missing something

Thanks Mark.


----- Original Message ----
From: Mark Williamson <mark.williamson@cl.cam.ac.uk>
To: Koripella Srinivas <talkwithsrinivas@yahoo.co.in>
Cc: xen-devel@lists.xensource.com
Sent: Friday, 6 July, 2007 9:55:48 AM
Subject: Re: [Xen-devel] Grant Tables  & mm.c

> " Linux tends to zap userspace page mappings quite cheerfully
> unless you do something to stop it, could this be your problem?"
>
> What does this mean exactly??

Linux "zaps" mappings when it removes them from pagetables.  Really this 
corresponds to zeroing a pte corresponding to that mapping.

Unless you do something to stop the Linux mm system from treating the grant 
table mappins like "normal" RAM, it will eventually want to simply zero them.  
This will break for grant mappings, since it is not a permitted means of 
unmapping granted memory.

> Any way i can prevent it from happening?

It depends on what you're trying to do...

Have you looked at the userspace grantables driver in our Linux tree?  This 
provides a device for mapping granted memory safely into userspace, handling 
cleanup correctly.  Even if that's not directly useful, maybe you can find 
some code in there that'll help you.

Cheers,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!







		
__________________________________________________________
Yahoo! India Answers: Share what you know. Learn something new
http://in.answers.yahoo.com/

[-- Attachment #1.2: Type: text/html, Size: 2429 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Grant Tables & mm.c

[Derek Murray]

On 7 Jul 2007, at 14:44, Koripella Srinivas wrote:

> I was looking at the user space grant tables but i did not find any  
> code calling the function
> "remap_pfn_range" to map the kmalloc ed allocated memory into user  
> space.
> What could be the reason?? Or am i missing something

The mapping takes place in drivers/xen/gntdev/gntdev.c, in the  
function gntdev_mmap(). In the normal case (no shadow page tables),  
the page is mapped into user space by calling  
HYPERVISOR_grant_table_op, with the GNTTABOP_map_grant_ref operation,  
and (amongst other things) a pointer to the page table entry to be  
rewritten. Therefore, it is not necessary to call remap_pfn_range().

I'm not sure what you mean by "the kmalloc ed allocated memory". The  
memory is allocated by another domain, and inserted in the grant  
table. In the kernel that carries out the mapping, kmalloc is used  
only for allocating kernel-private data structures.

Regards,

Derek Murray.