Re: pte_val(*pte) as lvalue

Re: pte_val(*pte) as lvalue

[Andi Kleen]

Timur Tabi <ttabi@interactivesi.com> writes:

> 
> What is the accepted way to assign an integer to a pte that works in 2.2 and
> 2.4?

set_pte(pte, mk_pte( ... )) 


-Andi

eCos

[Matthew M. DeLoera]

Does anyone on the list use eCos (RedHat's embedded Linux)?

I'm having some build and other general problems, and I'd like to ask
someone about them off-list.

Thanks...
- Matt

Re: pte_val(*pte) as lvalue

[Timur Tabi]

** Reply to message from Andi Kleen <ak@suse.de> on 03 Jul 2001 01:33:42 +0200

> Timur Tabi <ttabi@interactivesi.com> writes:
> 
> > 
> > What is the accepted way to assign an integer to a pte that works in 2.2 and
> > 2.4?
> 
> set_pte(pte, mk_pte( ... )) 

I'm not sure how to use mk_pte.  The first parameter is a struct page *, which
I don't have.  All I'm doing is modifying the PTE value.  I don't want to "make"
another one.


-- 
Timur Tabi - ttabi@interactivesi.com
Interactive Silicon - http://www.interactivesi.com

Re: pte_val(*pte) as lvalue

[Russell King]

On Tue, Jul 03, 2001 at 01:32:36PM -0500, Timur Tabi wrote:
> ** Reply to message from Andi Kleen <ak@suse.de> on 03 Jul 2001 01:33:42 +0200
> > set_pte(pte, mk_pte( ... )) 
> 
> I'm not sure how to use mk_pte.  The first parameter is a struct page *,
> which I don't have.  All I'm doing is modifying the PTE value.  I don't
> want to "make" another one.

set_pte is the only way you can guarantee that the architecture
implementation gets to do what it needs to do with the PTE value
before stuffing it into the PTE tables.

Can I ask what the nature of the PTE modification is, and where you
are making this modification?

--
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: pte_val(*pte) as lvalue

[Timur Tabi]

Russell King wrote:

>Can I ask what the nature of the PTE modification is, and where you
>are making this modification?
>
I've written a hack which enables PAT (Page Address Translation) for a 
particular page:

void set_pte_pat(pte_t *pte, unsigned long pat_index)
{
    unsigned long p = pte_val(*pte);

    p &= ~(_PAGE_PROTNONE | _PAGE_PCD | _PAGE_PWT);    // zero-out the 
relevant bits

    if (pat_index & 4)
    p |= _PAGE_PROTNONE;

    if (pat_index & 2)
    p |= _PAGE_PCD;

    if (pat_index & 1)
    p |= _PAGE_PWT;

#if CONFIG_X86_PAE
    pte->pte_high = 0;
    pte->pte_low = p;
#else
    pte_val(*pte) = p;
#endif

-- 
Timur Tabi
Interactive Silicon

Re: pte_val(*pte) as lvalue

[Russell King]

On Tue, Jul 03, 2001 at 01:50:05PM -0500, Timur Tabi wrote:
> Russell King wrote:
> >Can I ask what the nature of the PTE modification is, and where you
> >are making this modification?
> >
> I've written a hack which enables PAT (Page Address Translation) for a 
> particular page:

Firstly, I'll say I'm no x86 expert by any means.  However, it may be
better to extend the pte bit handling functions in
include/asm-i386/pgtable.h to include the bits you need to handle.
(eg, see how pte_mkwrite and pte_wrprotect are implemented.)  This is
probably the cleanest way of handling these bits.

--
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

pte_val(*pte) as lvalue

[Timur Tabi]

In my driver, I have this code:

    unsigned long p = pte_val(*pte);

    [set some bits in "p"]

    pte_val(*pte) = p;

This works fine in 2.2, and 2.4 when PAE support is disabled.  When PAE support
is enabled, it doesn't work, because the pte_val macro can no longer be an
lvalue.

What is the accepted way to assign an integer to a pte that works in 2.2 and
2.4?


-- 
Timur Tabi - ttabi@interactivesi.com
Interactive Silicon - http://www.interactivesi.com