[Xenomai] Question about rtdm_mmap_to user

[Xenomai] Question about rtdm_mmap_to user

[Marco Tessore]

Good morning,
I would like to share a buffer between user space task and kernel space 
realtime driver.

I used the function rtdm_mmap_to user in the following way:

In the driver, in the function myDriver_ioctl_rt I have the code:

        spild->agg_write_frame_buffer allocated with kalloc

         ret = rtdm_mmap_to_user( user_info,
                 spild->agg_write_frame_buffer,
                 agg_fun_local.data_size,
                 PROT_READ|PROT_WRITE ,
                 &tmpPtr,
                 NULL,
                 NULL
         );
         agg_fun_local.data_buffer = (uint8_t *) tmpPtr;

         rtdm_printk("mmap = %X\n", (unsigned) 
agg_fun_local.data_buffer); // print the returned address

         if (ret)
         {
             rtdm_printk("mmap error1\n");

             kfree(spild->agg_write_frame_buffer);
             return ret;
         }

         // check data
         spild->agg_write_frame_buffer[0] = 0xBE;
         spild->agg_write_frame_buffer[1] = 0xEF;

         // then returned the agg_fun_local structure to userspace

In the user space process I have the following code:
     ret = rt_dev_ioctl( device, IOCTL_MYDRIVER, &agg_fun );
     if (ret)
     {
         printf("PREP_AGG_WRITE failed, ret = %X", (int) ret);
     }

     printf("USER = Address = %X\n", (unsigned) agg_fun.data_buffer);
     printf("USER check %X%X\n", pAggFun->data_buffer[0], 
agg_fun.data_buffer[1]);

I have verified that the address in userspace is the same of the one 
returned by the rtdm_mmap_to_user.
The problem is that in the process I cannot see the check data written 
in the driver.

Is there something wrong in my code?
Are there some constraints tha I haven't took into account?

Thank you very much
Best regards
Marco Tessore

Re: [Xenomai] Question about rtdm_mmap_to user

[Gilles Chanteperdrix]

On 07/15/2014 06:03 PM, Marco Tessore wrote:
> Good morning,
> I would like to share a buffer between user space task and kernel space 
> realtime driver.
> 
> I used the function rtdm_mmap_to user in the following way:
> 
> In the driver, in the function myDriver_ioctl_rt I have the code:
> 
>         spild->agg_write_frame_buffer allocated with kalloc
> 
>          ret = rtdm_mmap_to_user( user_info,
>                  spild->agg_write_frame_buffer,
>                  agg_fun_local.data_size,
>                  PROT_READ|PROT_WRITE ,
>                  &tmpPtr,
>                  NULL,
>                  NULL
>          );
>          agg_fun_local.data_buffer = (uint8_t *) tmpPtr;
> 
>          rtdm_printk("mmap = %X\n", (unsigned) 
> agg_fun_local.data_buffer); // print the returned address
> 
>          if (ret)
>          {
>              rtdm_printk("mmap error1\n");
> 
>              kfree(spild->agg_write_frame_buffer);
>              return ret;
>          }
> 
>          // check data
>          spild->agg_write_frame_buffer[0] = 0xBE;
>          spild->agg_write_frame_buffer[1] = 0xEF;
> 
>          // then returned the agg_fun_local structure to userspace
> 
> In the user space process I have the following code:
>      ret = rt_dev_ioctl( device, IOCTL_MYDRIVER, &agg_fun );
>      if (ret)
>      {
>          printf("PREP_AGG_WRITE failed, ret = %X", (int) ret);
>      }
> 
>      printf("USER = Address = %X\n", (unsigned) agg_fun.data_buffer);
>      printf("USER check %X%X\n", pAggFun->data_buffer[0], 
> agg_fun.data_buffer[1]);
> 
> I have verified that the address in userspace is the same of the one 
> returned by the rtdm_mmap_to_user.
> The problem is that in the process I cannot see the check data written 
> in the driver.
> 
> Is there something wrong in my code?
> Are there some constraints tha I haven't took into account?

On which platform do you do this? If on ARM with a VIVT cache, you have
to flush the cache after writing one side for the other side to see the
change, or map the region with writecombine.

-- 
                                                                Gilles.