Re: [Query] : PCIe - Endpoint Function

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Oct 29, 2019 at 10:54 AM Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
>
> Hi,
>
> On 29/10/19 3:54 PM, Lad, Prabhakar wrote:
> > Hi Kishon,
> >
> > On Tue, Oct 22, 2019 at 7:06 AM Lad, Prabhakar
> > <prabhakar.csengg@xxxxxxxxx> wrote:
> >>
> >> Hi Kishon,
> >>
> >>
> >> On Tue, Oct 15, 2019 at 8:53 AM Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
> >>>
> >>> Hi Prabhakar,
> >>>
> >>> On 11/10/19 7:07 PM, Lad, Prabhakar wrote:
> >>>> Hi Kishon
> >>>>
> >>>> On Fri, Oct 11, 2019 at 8:35 AM Lad, Prabhakar
> >>>> <prabhakar.csengg@xxxxxxxxx> wrote:
> >>>>>
> >>>>> Hi Kishon,
> >>>>>
> >>>>> On Thu, Oct 10, 2019 at 12:32 PM Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
> >>>>>>
> >>>>>> Hi Prabhakar,
> >>>>>>
> >>>>>> On 10/10/19 4:57 PM, Lad, Prabhakar wrote:
> >>>>>>> Hello,
> >>>>>>>
> >>>>>>> I am currently working on adding pcie-endpoint support for a
> >>>>>>> controller, this controller doesn't support outbound- inbound address
> >>>>>>> translations, it has 1-1 mapping between the CPU and PCI addresses,
> >>>>>>> the current endpoint framework is based on  outbound-inbound
> >>>>>>> translations, what is the best approach to add this support, or is
> >>>>>>> there any WIP already for it ?
> >>>>>>
> >>>>>> How will the endpoint access host buffer without outbound ATU? I assume the PCI
> >>>>>> address reserved for endpoint is not the full 32-bit or 64-bit address space?
> >>>>>> In that case, the endpoint cannot directly access the host buffer (unless the
> >>>>>> host already knows the address space of the endpoint and gives the endpoint an
> >>>>>> address in its OB address space).
> >>>>>>
> >>>> I lied in my previous mail.
> >>>>
> >>>> a] The controller needs the cpu_address before starting the link, ie
> >>>> with the current implementation,the bars physical address in endpoint
> >>>> are assigned
> >>>> using dma_alloc_coherent(), but I what I actually want here is the
> >>>> phys_addr returned by pci_epc_mem_alloc_addr().
> >>>>
> >>>> b] In the pci_endpoint_test driver, the pci_address sent to the
> >>>> endpoint driver is again dma_alloc_coherent(), but the address which I
> >>>> actually want to
> >>>> send to endpoint is the BAR's assigned regions in the RC.
> >>>
> >>> The BAR assigned regions are usually used by RC to access EP memory.
> >>> dma_alloc_coherent() is used in pci_endpoint_test to allocate buffer in host
> >>> memory to be accessed by EP. Can you again check if statement 'b' is accurate?
> >>>
> >> yes you were correct, I misread the manual I have a rough driver
> >> working now, will post as
> >> soon as I tidy it up.
> >>
> > after several runs of pcitest I hit the following issue any pointers
> > on would this be the RC/endpoint ?
>
> It's difficult to tell without seeing your EP controller driver. It could be a
> ordering issue. Can you add mb() after memcpy_fromio() in pcitest (for the
> error below)?
>
adding a memory barrier to doesn't help either. Ill debug further.

Cheers,
--Prabhakar Lad



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux