Using mini-os ioremap()

Using mini-os ioremap()

[Matthew Fioravante]

[-- Attachment #1.1: Type: text/plain, Size: 307 bytes --]

  I'm trying to map some mmio pages into my mini-os domain to use a 
hardware device. The device is at address fed40000. I'm calling 
ioremap() on this address but its failing, telling me that mfn fed40 is 
in ram space. (Its failing the mfn_is_ram() check.)

What am I doing wrong here?

Thanks!


[-- Attachment #1.2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 2518 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: Using mini-os ioremap()

[Samuel Thibault]

Matthew Fioravante, le Wed 05 Jan 2011 14:24:55 -0500, a écrit :
>  I'm trying to map some mmio pages into my mini-os domain to use a 
> hardware device. The device is at address fed40000. I'm calling 
> ioremap() on this address but its failing, telling me that mfn fed40 is 
> in ram space. (Its failing the mfn_is_ram() check.)
> 
> What am I doing wrong here?

Mmm, I don't think anything you're doing is wrong. Notice the comment:

static long system_ram_end_mfn;
int mfn_is_ram(unsigned long mfn)
{
    /* very crude check if a given MFN is memory or not. Probably should
     * make this a little more sophisticated ;) */
    return (mfn <= system_ram_end_mfn) ? 1 : 0;
}

Do you have more than 4GiB memory, i.e. 0xfed40000 is actually between
two RAM areas?  In such case mfn_is_ram has to really get more
sophisticated.

(In the interim you could just comment the mfn_is_ram() test from
do_ioremap to make things work)

Samuel

Re: Using mini-os ioremap()

[Matthew Fioravante]

[-- Attachment #1.1: Type: text/plain, Size: 1482 bytes --]

  My box doesn't have more than 4GB. I've tried this on 32 bit and 64 bit.

I tried nuking the mfn_is_ram check. When I do that the call returns a 
virtual memory address but when I try to read and write to it, it 
appears to act just like a normal page of memory and not the device.

Does anyone know offhand of a mini-os project that either tested the 
ioremap() call or actually does some memory mapped io operations?

On 01/05/2011 05:55 PM, Samuel Thibault wrote:
> Matthew Fioravante, le Wed 05 Jan 2011 14:24:55 -0500, a écrit :
>>   I'm trying to map some mmio pages into my mini-os domain to use a
>> hardware device. The device is at address fed40000. I'm calling
>> ioremap() on this address but its failing, telling me that mfn fed40 is
>> in ram space. (Its failing the mfn_is_ram() check.)
>>
>> What am I doing wrong here?
> Mmm, I don't think anything you're doing is wrong. Notice the comment:
>
> static long system_ram_end_mfn;
> int mfn_is_ram(unsigned long mfn)
> {
>      /* very crude check if a given MFN is memory or not. Probably should
>       * make this a little more sophisticated ;) */
>      return (mfn<= system_ram_end_mfn) ? 1 : 0;
> }
>
> Do you have more than 4GiB memory, i.e. 0xfed40000 is actually between
> two RAM areas?  In such case mfn_is_ram has to really get more
> sophisticated.
>
> (In the interim you could just comment the mfn_is_ram() test from
> do_ioremap to make things work)
>
> Samuel



[-- Attachment #1.2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 2518 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: Using mini-os ioremap()

[Samuel Thibault]

Matthew Fioravante, le Thu 06 Jan 2011 12:00:04 -0500, a écrit :
>  My box doesn't have more than 4GB. I've tried this on 32 bit and 64 bit.

How much does it have? Could you post the memory map recorded in xm
dmesg?

> I tried nuking the mfn_is_ram check. When I do that the call returns a 
> virtual memory address but when I try to read and write to it, it 
> appears to act just like a normal page of memory and not the device.

Mmm... I wonder how that can be possible if your address is really an IO
area.

> Does anyone know offhand of a mini-os project that either tested the 
> ioremap() call or actually does some memory mapped io operations?

It was implemeted by Rolf Neugebauer <rolf.neugebauer@netronome.com>,
Cc-ed.

Samuel