rmap and nvidia?

rmap and nvidia?

[mdew]

is there a nvidia patch available to make it work with rmap?

nirvana:~/NVIDIA_kernel-1.0-3123# make
echo \#define NV_COMPILER \"`cc -v 2>&1 | tail -1`\" > nv_compiler.h
cc -c -Wall -Wimplicit -Wreturn-type -Wswitch -Wformat -Wchar-subscripts
-Wparentheses -Wpointer-arith -Wcast-qual -Wno-multichar  -O -MD
-D__KERNEL__ -DMODULE -D_LOOSE_KERNEL_NAMES -DNTRM -D_GNU_SOURCE
-DRM_HEAPMGR -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE 
-DNV_MAJOR_VERSION=1 -DNV_MINOR_VERSION=0 -DNV_PATCHLEVEL=3123 
-DNV_UNIX   -DNV_LINUX   -DNVCPU_X86       -I.
-I/lib/modules/2.4.20-xfs-rmap15b/build/include -Wno-cast-qual nv.c
nv.c: In function `nv_get_phys_address':
nv.c:2182: warning: implicit declaration of function `pte_offset'
nv.c:2182: invalid type argument of `unary *'
make: *** [nv.o] Error 1

Re: rmap and nvidia?

[William Lee Irwin III]

On Sat, Dec 14, 2002 at 10:36:10PM +1300, mdew wrote:
> is there a nvidia patch available to make it work with rmap?
> 
> nirvana:~/NVIDIA_kernel-1.0-3123# make
> echo \#define NV_COMPILER \"`cc -v 2>&1 | tail -1`\" > nv_compiler.h
> cc -c -Wall -Wimplicit -Wreturn-type -Wswitch -Wformat -Wchar-subscripts
> -Wparentheses -Wpointer-arith -Wcast-qual -Wno-multichar  -O -MD
> -D__KERNEL__ -DMODULE -D_LOOSE_KERNEL_NAMES -DNTRM -D_GNU_SOURCE
> -DRM_HEAPMGR -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE 
> -DNV_MAJOR_VERSION=1 -DNV_MINOR_VERSION=0 -DNV_PATCHLEVEL=3123 
> -DNV_UNIX   -DNV_LINUX   -DNVCPU_X86       -I.
> -I/lib/modules/2.4.20-xfs-rmap15b/build/include -Wno-cast-qual nv.c
> nv.c: In function `nv_get_phys_address':
> nv.c:2182: warning: implicit declaration of function `pte_offset'
> nv.c:2182: invalid type argument of `unary *'
> make: *** [nv.o] Error 1

Use pte_offset_map() with a corresponding pte_unmap().


Bill

Re: rmap and nvidia?

[mdew]

On Sat, 2002-12-14 at 22:38, William Lee Irwin III wrote:
> On Sat, Dec 14, 2002 at 10:36:10PM +1300, mdew wrote:
> > is there a nvidia patch available to make it work with rmap?
> > 
> > nirvana:~/NVIDIA_kernel-1.0-3123# make
> > echo \#define NV_COMPILER \"`cc -v 2>&1 | tail -1`\" > nv_compiler.h
> > cc -c -Wall -Wimplicit -Wreturn-type -Wswitch -Wformat -Wchar-subscripts
> > -Wparentheses -Wpointer-arith -Wcast-qual -Wno-multichar  -O -MD
> > -D__KERNEL__ -DMODULE -D_LOOSE_KERNEL_NAMES -DNTRM -D_GNU_SOURCE
> > -DRM_HEAPMGR -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE 
> > -DNV_MAJOR_VERSION=1 -DNV_MINOR_VERSION=0 -DNV_PATCHLEVEL=3123 
> > -DNV_UNIX   -DNV_LINUX   -DNVCPU_X86       -I.
> > -I/lib/modules/2.4.20-xfs-rmap15b/build/include -Wno-cast-qual nv.c
> > nv.c: In function `nv_get_phys_address':
> > nv.c:2182: warning: implicit declaration of function `pte_offset'
> > nv.c:2182: invalid type argument of `unary *'
> > make: *** [nv.o] Error 1
> 
> Use pte_offset_map() with a corresponding pte_unmap().

err pardon?

Re: rmap and nvidia?

[Rik van Riel]

On Sat, 14 Dec 2002, mdew wrote:
> On Sat, 2002-12-14 at 22:38, William Lee Irwin III wrote:
> > On Sat, Dec 14, 2002 at 10:36:10PM +1300, mdew wrote:
> > > nv.c: In function `nv_get_phys_address':
> > > nv.c:2182: warning: implicit declaration of function `pte_offset'
> > > nv.c:2182: invalid type argument of `unary *'
> >
> > Use pte_offset_map() with a corresponding pte_unmap().
>
> err pardon?

wli just gave you the information you need to create a patch
for the nvidia driver.

Rik
-- 
Bravely reimplemented by the knights who say "NIH".
http://www.surriel.com/		http://guru.conectiva.com/
Current spamtrap:  <a href=mailto:"october@surriel.com">october@surriel.com</a>

Re: rmap and nvidia?

[Eyal Lebedinsky]

Rik van Riel wrote:
> 
> On Sat, 14 Dec 2002, mdew wrote:
> > On Sat, 2002-12-14 at 22:38, William Lee Irwin III wrote:
> > > On Sat, Dec 14, 2002 at 10:36:10PM +1300, mdew wrote:
> > > > nv.c: In function `nv_get_phys_address':
> > > > nv.c:2182: warning: implicit declaration of function `pte_offset'
> > > > nv.c:2182: invalid type argument of `unary *'
> > >
> > > Use pte_offset_map() with a corresponding pte_unmap().
> >
> > err pardon?
> 
> wli just gave you the information you need to create a patch
> for the nvidia driver.

The replies for people in the know (Rik, wli) give a clue but not an
answer. Use mere mortals want a proper patch in order to build and
use this kernel.

I will summarise my understanding so far; The original code says:

unsigned long
nv_get_phys_address(unsigned long address)
{
    pgd_t *pg_dir;
    pmd_t *pg_mid_dir;
    pte_t *pte__, pte;
.....
#if defined (pte_offset_atomic)
    pte__ = pte_offset_atomic(pg_mid_dir, address);
    pte = *pte__;
    pte_kunmap(pte__);
#else
    pte__ = NULL;
    pte = *pte_offset(pg_mid_dir, address);
#endif

    if (!pte_present(pte))
        goto failed;

    return ((pte_val(pte) & KERN_PAGE_MASK) | NV_MASK_OFFSET(address));
.....
}

The last line above is the problem. So far I could see two possible
changes that will compile, but I do not know which will function
correctly. The first replacement option:
    pte = *pte_offset(pg_mid_dir, address);

The second replacement option is more involved:
    pte__ = pte_offset_map(pg_mid_dir, address);
    pte = *pte__;

    if (!pte_present(pte))
        goto failed;

    pte_unmap(pte__);

Reading the patch itself I see places where the first approach is used,
while elsewhere the second is used. I do not know what pte_val(pte)
requires though. Can we do the pte_unmap(pte__) earlier or is the
mapping
necessary for pte_present(pte)? Will this work:
    pte__ = pte_offset_map(pg_mid_dir, address);
    pte = *pte__;
    pte_unmap(pte__);


In summary, you can see that for someone who is not intimately involved
the answers so far do not provide a working patch.

Thanks everybody.

--
Eyal Lebedinsky (eyal@eyal.emu.id.au) <http://samba.org/eyal/>

Re: rmap and nvidia?

[Philip Dodd]

Eyal Lebedinsky wrote:
8<
 > The replies for people in the know (Rik, wli) give a clue but not an
 > answer. Use mere mortals want a proper patch in order to build and
 > use this kernel.
8<

This driver, you mean ;-)

Re: rmap and nvidia?

[Eyal Lebedinsky]

Philip Dodd wrote:
> 
> Eyal Lebedinsky wrote:
> 8<
>  > The replies for people in the know (Rik, wli) give a clue but not an
>  > answer. Use mere mortals want a proper patch in order to build and
>  > use this kernel.
> 
> This driver, you mean ;-)

Well, yes and no. This kernel is of no use to me without this driver
right now.

--
Eyal Lebedinsky (eyal@eyal.emu.id.au) <http://samba.org/eyal/>

Re: rmap and nvidia?

[Ducrot Bruno]

On Sun, Dec 15, 2002 at 12:40:48PM +1100, Eyal Lebedinsky wrote:
> Philip Dodd wrote:
> > 
> > Eyal Lebedinsky wrote:
> > 8<
> >  > The replies for people in the know (Rik, wli) give a clue but not an
> >  > answer. Use mere mortals want a proper patch in order to build and
> >  > use this kernel.
> > 
> > This driver, you mean ;-)
> 
> Well, yes and no. This kernel is of no use to me without this driver
> right now.
> 

Ah, OK.  You need to play quake3 with this kernel, I guess.

-- 
Ducrot Bruno
http://www.poupinou.org        Page profaissionelle
http://toto.tu-me-saoules.com  Haume page

Re: rmap and nvidia?

[O.Sezer]

The 2.5 patches at http://www.minion.de/nvidia.html
may give some clue. They use pte_offset_map , yes,
but no corresponding pte_unmap ...

Re: rmap and nvidia?

[William Lee Irwin III]

On Mon, Dec 16, 2002 at 04:25:11PM +0200, O. Sezer wrote:
> The 2.5 patches at http://www.minion.de/nvidia.html
> may give some clue. They use pte_offset_map , yes,
> but no corresponding pte_unmap ...

That will BUG() with highmem debugging on. Put the pte_unmap() in the
appropriate place.


Bill

Re: rmap and nvidia?

[Andrew McGregor]

Come on, think laptops.  XFree86 doesn't support all the NVidias without 
that driver.  Well, except in vesa framebuffer mode, which is horrible.

--On Monday, December 16, 2002 14:32:23 +0100 Ducrot Bruno 
<poup@poupinou.org> wrote:

> On Sun, Dec 15, 2002 at 12:40:48PM +1100, Eyal Lebedinsky wrote:
>> Philip Dodd wrote:
>> >
>> > Eyal Lebedinsky wrote:
>> > 8<
>> >  > The replies for people in the know (Rik, wli) give a clue but not an
>> >  > answer. Use mere mortals want a proper patch in order to build and
>> >  > use this kernel.
>> >
>> > This driver, you mean ;-)
>>
>> Well, yes and no. This kernel is of no use to me without this driver
>> right now.
>>
>
> Ah, OK.  You need to play quake3 with this kernel, I guess.
>
> --
> Ducrot Bruno
> http://www.poupinou.org        Page profaissionelle
> http://toto.tu-me-saoules.com  Haume page
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
>

Re: rmap and nvidia?

[O.Sezer]

[-- Attachment #1: Type: text/plain, Size: 93 bytes --]

Is this patch correct in any way?
(Ripped out of the 2.5 patch and modified some).

Thanks.


[-- Attachment #2: NVIDIA_kernel_rmap.patch --]
[-- Type: text/plain, Size: 2173 bytes --]

diff -urN NVIDIA_kernel-1.0-4191/nv.c NVIDIA_kernel.rmap15b/nv.c
--- NVIDIA_kernel-1.0-4191/nv.c	2002-12-09 22:27:15.000000000 +0200
+++ NVIDIA_kernel.rmap15b/nv.c	2002-12-17 00:03:12.000000000 +0200
@@ -2217,7 +2217,7 @@
 {
     pgd_t *pg_dir;
     pmd_t *pg_mid_dir;
-    pte_t *pte__, pte;
+    pte_t pte;
 
 #if defined(NVCPU_IA64)
     if (address > __IA64_UNCACHED_OFFSET)
@@ -2241,14 +2241,7 @@
     if (pmd_none(*pg_mid_dir))
         goto failed;
 
-#if defined (pte_offset_atomic)
-    pte__ = pte_offset_atomic(pg_mid_dir, address);
-    pte = *pte__;
-    pte_kunmap(pte__);
-#else
-    pte__ = NULL;
-    pte = *pte_offset(pg_mid_dir, address);
-#endif
+    PTE_OFFSET(pg_mid_dir, address, pte);
 
     if (!pte_present(pte))
         goto failed;
diff -urN NVIDIA_kernel-1.0-4191/nv-linux.h NVIDIA_kernel.rmap15b/nv-linux.h
--- NVIDIA_kernel-1.0-4191/nv-linux.h	2002-12-09 22:27:15.000000000 +0200
+++ NVIDIA_kernel.rmap15b/nv-linux.h	2002-12-17 00:02:19.000000000 +0200
@@ -146,6 +146,28 @@
 #  define VMA_PRIVATE(vma)              ((void*)((vma)->vm_pte))
 #endif
 
+#ifdef pte_offset_map		/* rmap-vm or 2.5  */
+#define PTE_OFFSET(pmd, address, pte)               \
+ {                                                  \
+     pte_t *pPTE;                                   \
+     pPTE = pte_offset_map(pmd, address);           \
+     pte = *pPTE;                                   \
+     pte_unmap(pPTE);                               \
+ }
+#else
+#ifdef pte_offset_atomic		/* aa-vm   */
+#define PTE_OFFSET(pmd, address, pte)               \
+  {                                                 \
+     pte_t *pPTE;                                   \
+     pPTE = pte_offset_atomic(pmd, address);        \
+     pte = *pPTE;                                   \
+     pte_kunmap(pPTE);                              \
+  }
+#else	/* !pte_offset_atomic */
+#define PTE_OFFSET(pmd, address, pte)   (pte = *pte_offset(pmd, address))
+#endif	/* pte_offset_atomic  */
+#endif	/* pte_offset_map     */
+
 #define NV_PAGE_ALIGN(addr)             ( ((addr) + PAGE_SIZE - 1) / PAGE_SIZE)
 #define NV_MASK_OFFSET(addr)            ( (addr) & (PAGE_SIZE - 1) )
 

Re: rmap and nvidia?

[Eyal Lebedinsky]

"O.Sezer" wrote:
> 
> Is this patch correct in any way?
> (Ripped out of the 2.5 patch and modified some).

The equivalent minimal patch for nv.c in 2.4 will then be:

--- modules/nvidia/NVIDIA_kernel/nv.c.orig      Tue Dec 10 07:27:15 2002
+++ modules/nvidia/NVIDIA_kernel/nv.c   Tue Dec 17 10:07:37 2002
@@ -2247,7 +2247,13 @@
     pte_kunmap(pte__);
 #else
     pte__ = NULL;
+#ifdef pte_offset
     pte = *pte_offset(pg_mid_dir, address);
+#else	/* rmap-vm */
+    pte__ = pte_offset_map(pg_mid_dir, address);
+    pte = *pte__;
+    pte_unmap(pte__);
+#endif
 #endif

--
Eyal Lebedinsky (eyal@eyal.emu.id.au) <http://samba.org/eyal/>

Re: rmap and nvidia?

[Andrew McGregor]

The bit calling pte_unmap in nv-linux.h is the only difference from the 
current minion.de patch, but it works, doesn't oops any more, nor does it 
leak memory any more :-)

Great!  X on 2.5 on my laptop!

Andrew

--On Tuesday, December 17, 2002 00:22:34 +0200 "O.Sezer" 
<sezero@superonline.com> wrote:

> Is this patch correct in any way?
> (Ripped out of the 2.5 patch and modified some).
>
> Thanks.
>

Re: rmap and nvidia?

[Andrew McGregor]

[-- Attachment #1: Type: text/plain, Size: 375 bytes --]

So, first apply the patch for 4191 from www.minion.de, then the attached 
one based on yours.  Been running overnight and beaten on by XScreesaver, 
no memory leak anymore.

Andrew

--On Tuesday, December 17, 2002 00:22:34 +0200 "O.Sezer" 
<sezero@superonline.com> wrote:

> Is this patch correct in any way?
> (Ripped out of the 2.5 patch and modified some).
>
> Thanks.
>


[-- Attachment #2: NVIDIA_kernel-1.0-4191-2.5-pte_unmap.diff --]
[-- Type: application/octet-stream, Size: 1607 bytes --]

--- NVIDIA_kernel-1.0-4191-2.5-minion/nv-linux.h	2002-12-18 08:25:50.000000000 +1300
+++ NVIDIA_kernel-1.0-4191-2.5/nv-linux.h	2002-12-17 22:03:31.000000000 +1300
@@ -151,21 +151,27 @@
 #  define SMP_NUM_CPUS                  smp_num_cpus
 #endif
 
-#ifdef KERNEL_2_5
-#define PTE_OFFSET(pmd, address, pte)   (pte = *pte_offset_map(pmd, address))
+#ifdef pte_offset_map		/* rmap-vm or 2.5  */
+#define PTE_OFFSET(pmd, address, pte)               \
+ {                                                  \
+     pte_t *pPTE;                                   \
+     pPTE = pte_offset_map(pmd, address);           \
+     pte = *pPTE;                                   \
+     pte_unmap(pPTE);                               \
+ }
 #else
-#ifdef pte_offset_atomic
+#ifdef pte_offset_atomic		/* aa-vm   */
 #define PTE_OFFSET(pmd, address, pte)               \
-  {                                                  \
+  {                                                 \
      pte_t *pPTE;                                   \
      pPTE = pte_offset_atomic(pmd, address);        \
      pte = *pPTE;                                   \
      pte_kunmap(pPTE);                              \
   }
-#else /* !pte_offset_atomic */
+#else	/* !pte_offset_atomic */
 #define PTE_OFFSET(pmd, address, pte)   (pte = *pte_offset(pmd, address))
-#endif /* pte_offset_atomic */
-#endif /* KERNEL_2_5 */
+#endif	/* pte_offset_atomic  */
+#endif	/* pte_offset_map     */
 
 #define NV_PAGE_ALIGN(addr)             ( ((addr) + PAGE_SIZE - 1) / PAGE_SIZE)
 #define NV_MASK_OFFSET(addr)            ( (addr) & (PAGE_SIZE - 1) )

Re: rmap and nvidia?

[O.Sezer]

So, we should thank to wli for the information.



Andrew McGregor wrote:
> So, first apply the patch for 4191 from www.minion.de, then the attached 
> one based on yours.  Been running overnight and beaten on by 
> XScreesaver, no memory leak anymore.
> 
> Andrew
> 
> --On Tuesday, December 17, 2002 00:22:34 +0200 "O.Sezer" 
> <sezero@superonline.com> wrote:
> 
>> Is this patch correct in any way?
>> (Ripped out of the 2.5 patch and modified some).
>>
>> Thanks.
>>
>