[okl4-developer] Request virt to phys mapping in Iguana application

[Nelson Tam]

Hi Kashin,

OK, I've taken a good look at your problem now.

On 28/01/2008, at 5:18 PM, kashin Lin wrote:

> i have tried "memsection_virt_to_phys" as following:
>
> uintptr_t  virt, mem_sect;
> uintptr_t  scm_phys;
> size_t      size;
>
> mem_sect = memsection_create_user(0x1000, &virt);
> result = hardware_back_memsection(mem_sect,0x01c40000, L4_IOMemory);
>
> scm_phys = memsection_virt_to_phys(virt, &size);
>
> but i always got "scm:0xfffffffe, size:0x0"  not "scm:0x01c40000,  
> size:0x1000".
> is the way i used wrong or there are something i should take care?


First I'd like to point out that the "preferred" method of granting  
physical memory mappings to programs, is using the OKL4 object  
environment.

The OKL4 object environment is configured using an XML file in build/ 
images/weaver.xml.  This file is read during link time by the  
Elfweaver tool.  During system startup, the specified mappings will be  
processed before any program starts execution.

The advantages are numerous, but here are two good ones: i) you avoid  
taking the pagefault for mappings that you know will happen anyway;  
and more importantly ii) programs are only given physical mappings for  
those specified by the XML file.

In any case, here is how to configure the physical mapping:

a) next to other <physical_memory> tags, add:

   <physical_memory name="kashin_mem">
     <region base="0x01c40000" size="0x1000" />
   </physical_memory>

b) next to other <physical_pool> tags, add:

   <physical_pool name="kashin_pool" direct="true">
     <memory src="kashin_mem" />
   </physical_pool>

c) inside the <program> tag that you are running your test on, add:

   <memsection name="kashin_memsect" size="0x1000"  
physpool="kashin_pool" />

Now, in your program, you can do the following:

#include <l4/misc.h>
#include <l4/kdebug.h>

#include <iguana/env.h>
#include <iguana/memsection.h>

void
mapping_test()
{
     memsection_ref_t    ms;

     int                 result;

     uint32_t            * ptr;
     uintptr_t           ms_base, ms_size;
     size_t              size;

     ms      = env_memsection(iguana_getenv("KASHIN_MEMSECT"));

     ms_base = (uintptr_t)memsection_base(ms);
     ms_size = memsection_size(ms);
     result = memsection_virt_to_phys(ms_base, &size);

     printf("base:    0x%lx, size:    0x%lx\n", (unsigned long)result,  
(unsigned long)size);
     printf("ms_base: 0x%lx, ms_size: 0x%lx\n", ms_base, ms_size);

     ptr = (uint32_t *)ms_base;
     *ptr = 0xdeadbeef;

     printf("ptr: 0x%lx, *ptr: 0x%lx\n", (unsigned long)ptr, *ptr);

     L4_KDB_Enter("Success!");
}

Now compile.  But before you run it, you need to relink using the new  
XML file.

  $ rm -rf build/images/image.*
  $ vim build/images/weaver.xml (apply the changes now)
  $ ./tools/build.py <your_args> simulate


If you get an assertion like so:

Assertion failed: item, function env_memsection, file libs/iguana/src/ 
env.c, line 170.

then you haven't configured your XML properly.


You should see something like:

base:    0xfffffffe, size:    0x0
ms_base: 0x80090000, ms_size: 0x1000
ptr: 0x80090000, *ptr: 0xdeadbeef
--- KD# User: Success! ---

These are the same results as you have before.  But now you're in KDB,  
do the following:

 > ptab (you should just press 'p')
Space [current]: current (you should just press 'return')
80000000 [f010b70c]: tree=f010b70c
80000000 [a0116c01]:   tree=f4116c00
80040000 [00000000]:     tree=f4116d00
8004c000 [a00d5ffe]:       phys=a00d5000 pg=f4116d30   4KB rwx (RWX)  
user   WB domain = 9
8004d000 [a00d6ffe]:       phys=a00d6000 pg=f4116d34   4KB rwx (RWX)  
user   WB domain = 9
8004e000 [a00d7ffe]:       phys=a00d7000 pg=f4116d38   4KB rwx (RWX)  
user   WB domain = 9
80090000 [01c40ffe]:     tree=f4116e40
80090000 [01c40ffe]:       phys=01c40000 pg=f4116e40   4KB rwx (RWX)  
user   WB domain = 9
800a0000 [a0500ffd]:     phys=a0500000 pg=f4116e80    64KB rwx (RWX)  
user   WB domain = 9
80600000 [a0118001]:   tree=f4118000
80600000 [a0090aae]:     tree=f4118000
80600000 [a0090aae]:       phys=a0090000 pg=f4118000   4KB r~x (RWX)  
user   WB domain = 9
80601000 [a0091aae]:       phys=a0091000 pg=f4118004   4KB r~x (RWX)  
user   WB domain = 9
80602000 [a0092aae]:       phys=a0092000 pg=f4118008   4KB r~x (RWX)  
user   WB domain = 9
80603000 [a0093aae]:       phys=a0093000 pg=f411800c   4KB r~x (RWX)  
user   WB domain = 9
80604000 [a0094aae]:       phys=a0094000 pg=f4118010   4KB r~x (RWX)  
user   WB domain = 9
80605000 [a0095aae]:       phys=a0095000 pg=f4118014   4KB r~x (RWX)  
user   WB domain = 9
80606000 [a0096aae]:       phys=a0096000 pg=f4118018   4KB r~x (RWX)  
user   WB domain = 9
80607000 [a0097aae]:       phys=a0097000 pg=f411801c   4KB r~x (RWX)  
user   WB domain = 9
80608000 [a0098aae]:       phys=a0098000 pg=f4118020   4KB r~x (RWX)  
user   WB domain = 9
80610000 [a0072ffe]:     tree=f4118040
80610000 [a0072ffe]:       phys=a0072000 pg=f4118040   4KB rwx (RWX)  
user   WB domain = 9

What you're seeing is the page table.  Notice how the entry for  
80090000 actually maps to 01c40000 now.

Moreover, I can confirm that on our latest internal build, we're  
seeing correct results like so:

base:    0x1c40000, size:    0x1000
ms_base: 0x80090000, ms_size: 0x1000
ptr: 0x80090000, *ptr: 0xdeadbeef
--- KD# User: Success! ---

What this means is that we have already fixed this bug recently, and  
in the next public release you will be able to call  
memsection_virt_to_phys() and get correct results.

I hope this helps!  Keep the questions coming.
--
(nt)

Nelson Tam
nelson at ok-labs.com