Hi, On Mon, May 18, 2009 at 6:50 AM, Grant Grundler <[email protected]> wrote: > Hi Arun, Sandeep, > > On Sat, May 16, 2009 at 06:33:44PM +0530, SandeepKsinha wrote: >> Hi Arun, >> >> On Sat, May 16, 2009 at 5:48 PM, arun c <[email protected]> wrote: >> > Hi all, >> > >> > I want to copy data from user space to PCI memory. >> > >> > I mapped the PCI memory of the card by, >> > >> > 1)pci_request_regions(pci_dev, DEVICE_NAME); >> > 2)buffer_addr = pci_iomap(pci_dev, 1, 1024); >> > >> > Now I want to write data supplied by the >> > user to buffer_addr(BAR1 pre-fetch-able memory >> > of the device) >> > >> > I did like this: >> > >> > copy_from_user(buffer_addr, usr_addr, 1024) >> > >> > This method is working fine on an X86 running >> > linux 2.6.27 kernel. I can see the data clearly copied. >> > >> > I want to know is this legal for all the platforms? and for >> > older kernels starting from 2.6.16? >> > >> > If the above method is totally legal then can I use >> > copy_to_user(usr_addr, buffer_addr, 1024) also? >> >> AFAIK, yes. > > Sorry, I don't think so. > > Copying stuff to non-cacheable addresses can have other side effects.
My destination memory is a PCI pre-fetchable memory. I am not sure this is cacheable or not. > So the optimizations available when copying TO cacheable address space > are generally not available (or have risks). > > That's why we have copy_from_user_toio() I greped for copy_from_user_toio, only some files in the sound/pci folder seems to be using it. I make my scenario more clear here. I want to transfer data in and out from user space to PCI memory. I know I can do a copy_from_user to get data to kernel and do any iowrite* flavours to eventually write to PCI memory, and do the exact reverse to transfer data to user. I just wanted to eliminate the unnecessary buffering of data in kernel for which kernel has no interest. copy_from_user_toio and copy_to_user_fromio are designed for this purpose? I can test my driver on PPC and X86 machines, rest of the platforms are not available with me. That is why I wanted a bullet proof implementation. Regards, Arun C -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to [email protected] Please read the FAQ at http://kernelnewbies.org/FAQ
