• Apple
• Store
• Mac
• iPod + iTunes
• iPhone
• Downloads
• Support

Please consider the security implications of this design before
proceeding.  The purpose of a subclass of IOUserClient is to vend a
new service from the kernel, and along with that comes the
responsibility of defending the kernel from (buggy and malicious)
clients.  In general, this means validating all parameters, never
granting a client access to memory that they are not specifically
allowed to see, and not allowing a client to interfere with the
operation of the system.

Mapping device registers into user space is almost always a violation
of that responsibility.  If your device registers provide access to a
DMA engine, then the client could use that to scrape physical memory
that is mapped into other processes.  If the registers can cause the
device to generate spurious interrupts, or interfere with the bus it
is attached to (say, alter PCI timings), it could potentially crash
or lock-up the system.

There are a few cases where mapping registers into user-space can be
OK:  If you are sure the registers can do no harm - say they provide
read-only access to a random-number seed generator and no combination
of reads or writes to those registers can alter any other memory or
interfere with any other device on the system.  Or, if you are just
writing an experimenter kext for helping with your bringup or for
private use and you don't plan to deploy this solution, then mapping
registers is obviously fine.

Sorry I didn't answer your question, but I thought this was important
to bring up,
Thanks,
- Dean


On May 23, 2007, at 2:17 AM, rohit dhamija wrote:
> Dear All,
>
> I am developing a KEXT for PCI device for Mac OS X 10.4.9 (ppc and
> intel) and stuck up while mapping device memory at user level. I
> searched all the older archives for this but could no where get the
> hint. Also, couldnot find any relevant sample for this. I hope you
> would please help me out in this matter.
>
> Below is the piece of code that I wrote at kernel level. To allow
> user client to map hardware registers into my application, I am
> using IOConnectMapMemory. This API returns success, and I get an
> address corresponding to the same. But when i see the memory
> contents corresponding to it, it shows all values as AAAAAAAAA,
> AAAAAAA, i.e. it seems nothing is mapped.
> /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
> * * * */
> #define super IOUserClient
> OSDefineMetaClassAndStructors(MyDriver_drv_user, IOUserClient)
>
> /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
> * * * */
> extern  IODeviceMemory *physicalMem;
>
> IOReturn MyDriver_drv_user::clientMemoryForType(UInt32 type,
> IOOptionBits * options, IOMemoryDescriptor ** memory)
> {
>  physicalMem->map( fTask, kIOMapAnywhere, kIOMapAnywhere, 0, 0); if
> (!physicalMem) return -1;
>  physicalMem->retain();
>  *memory = physicalMem;
>  return kIOReturnSuccess;
> }
>
> /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
> * * * */
> #define super IOService
> OSDefineMetaClassAndStructors(MyDriver_drv, IOService)
>
>  IOMemoryMap * fRegisterMap = NULL;
>  IODeviceMemory *physicalMem = NULL;
>  IOPCIDevice * fPCIDevice = NULL;
>  /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
> * * * */
>
> bool MyDriver_drv::start(IOService * provider)
> {
>     if (!super::start(provider))
>         return false;
>  fPCIDevice = (IOPCIDevice *) provider;
>   if(!fPCIDevice)
>   return false;
>
>    fPCIDevice->setMemoryEnable( true );
>    fPCIDevice->setIOEnable( true );
>    fRegisterMap = fPCIDevice->mapDeviceMemoryWithRegister
> (kIOPCIConfigBaseAddress2);
>
>    if(fRegisterMap)
>     {
>      physicalMem = IODeviceMemory::withRange(fRegisterMap-
> >getPhysicalAddress(),fRegisterMap->getLength());
>      physicalMem->initWithPhysicalAddress(fRegisterMap-
> >getPhysicalAddress(),fRegisterMap->getLength(), kIODirectionOutIn));
>     }
>
>     registerService();
>     return true;
> }
>
> Please let me know the fault with above piece of code and if
> possible, let me know the link of any sample application (if at all
> it exists)
>
> Thanks and Regards,
>
> Rohit Dhamija
>
>  _______________________________________________
> Do not post admin requests to the list. They will be ignored.
> Darwin-drivers mailing list      (email@hidden)
> Help/Unsubscribe/Update your Subscription:
> http://lists.apple.com/mailman/options/darwin-drivers/dean.reece%
> 40apple.com
>
> This email sent to email@hidden