On 01/07/2015 04:18 PM, Arnd Bergmann wrote:
On Wednesday 07 January 2015 13:49:50 Murali Karicheri wrote:
PCI devices on Keystone doesn't have correct dma_pfn_offset set. This patch
add capability to set the dma configuration such as dma-mask, dma_pfn_offset,
and dma ops etc using the information from DT. The prior RFCs and discussions
are available at [1] and [2] below.
[2] : https://www.mail-archive.com/linux-kernel@xxxxxxxxxxxxxxx/msg790244.html
[1] : http://www.gossamer-threads.com/lists/linux/kernel/2024591
Looks all good to me in this version, I'm just unsure about one thing:
- Limit the of_iommu_configure to non pci devices
My last recommendation was to pass the b/d/f number into
of_pci_dma_configure to handle this correctly. What was your
reason for not doing it in the end?
Arnd,
I had responded to this already on the list and reproduced below your
remark and my response below from that thread.
---------------cut-------------------------------------------------------
> Actually regarding the bit I wrote above, it might be helpful to pass
> the PCI_DEVID() into both of_iommu_configure and of_dma_configure.
>
> While this may or may not be sufficient, I think there is no question
> about it being needed for the ARM SMMU with PCI, so we may as well add
> it at the point when you touch the same lines already. In the platform
> bus case, just pass zero here.
Arnd,
PCI_DEVID() is defined as
#define PCI_DEVID(bus, devfn) ((((u16)(bus)) << 8) | (devfn))
So PCI_DEVID of 0 is a valid PCI DEVID.
How about checking if the device is PCI in of_iommu_configure() using
dev_is_pci(dev) macro and return immediately for PCI? Need to include
pci.h in this file though.
Some of the iommu drivers already include this.
a0868495@ares-ubuntu:~/projects/linux-keystone$ grep -r pci.h drivers/iommu/
drivers/iommu/amd_iommu_v2.c:#include <linux/pci.h>
drivers/iommu/dmar.c:#include <linux/pci.h>
drivers/iommu/amd_iommu_types.h:#include <linux/pci.h>
drivers/iommu/amd_iommu.c:#include <linux/pci.h>
drivers/iommu/fsl_pamu_domain.c:#include <sysdev/fsl_pci.h>
drivers/iommu/iommu.c:#include <linux/pci.h>
drivers/iommu/intel-iommu.c:#include <linux/pci.h>
drivers/iommu/intel_irq_remapping.c:#include <linux/pci.h>
drivers/iommu/arm-smmu.c:#include <linux/pci.h>
drivers/iommu/amd_iommu_init.c:#include <linux/pci.h>
drivers/iommu/irq_remapping.c:#include <linux/pci.h>
This will allow us to re-visit this later for IOMMU support for PCI
without polluting the API.
-----------------------cut-end---------------------------------------
I assumed you want to use value of zero for b/d/f to indicate it is for
platform. Also use the non zero value to skip the DT attribute parsing
for IOMMU DT attribute in of_iommu_configure() for PCI.
I did dev_is_pci() for skipping the processing for PCI. I thought it is
better to add the b/d/f argument later when this is re-visited later.
BTW,
I still haven't received any comment from Bjorn w.r.t PCI part.
BJorn, could you review and comment on 3/4 and 4/4?
Regards,
Murali
We will likely have to change this again if we take your current version,
but that can be a follow-up patch. I believe AMD requires it to use
PCI on their their Seattle platform.
Arnd
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Murali Karicheri
Linux Kernel, Texas Instruments
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
