Re: [PATCH 0/4] Redesign MSI-X support in PCIe Endpoint Core
From: Kishon Vijay Abraham I
Date: Thu Jan 09 2020 - 05:17:28 EST
Hi,
On 12/12/19 4:16 AM, Bjorn Helgaas wrote:
> On Wed, Dec 11, 2019 at 06:16:04PM +0530, Kishon Vijay Abraham I wrote:
>> Existing MSI-X support in Endpoint core has limitations:
>> 1) MSIX table (which is mapped to a BAR) is not allocated by
>> anyone. Ideally this should be allocated by endpoint
>> function driver.
>> 2) Endpoint controller can choose any random BARs for MSIX
>> table (irrespective of whether the endpoint function driver
>> has allocated memory for it or not)
>>
>> In order to avoid these limitations, pci_epc_set_msix() is
>> modified to include BAR Indicator register (BIR) configuration
>> and MSIX table offset as arguments. This series also fixed MSIX
>> support in dwc driver and add MSI-X support in Cadence PCIe driver.
>>
>> The previous version of Cadence EP MSI-X support is @ [1].
>> This series is created on top of [2]
>>
>> [1] -> https://patchwork.ozlabs.org/patch/971160/
>> [2] -> http://lore.kernel.org/r/20191209092147.22901-1-kishon@xxxxxx
>>
>> Alan Douglas (1):
>> PCI: cadence: Add MSI-X support to Endpoint driver
>>
>> Kishon Vijay Abraham I (3):
>> PCI: endpoint: Fix ->set_msix() to take BIR and offset as arguments
>> PCI: dwc: Fix dw_pcie_ep_raise_msix_irq() to get correct MSIX table
>> address
>> PCI: keystone: Add AM654 PCIe Endpoint to raise MSIX interrupt
>
> Trivial nits:
>
> - There's a mix of "MSI-X" and "MSIX" in the subjects, commit logs,
> and comments. I prefer "MSI-X" to match usage in the spec.
>
> - "Fixes:" tags need not include "commit". It doesn't *hurt*
> anything, but it takes up space that could be used for the
> subject.
>
> - Commit references typically use a 12-char SHA1. Again, doesn't
> hurt anything.
I'll fix all this in my next revision.
Xiaowei, Gustavo,
The issues we discussed in [1] should be fixed with this series. Can
you help test this in your platforms?
[1] -> https://lkml.org/lkml/2019/11/6/678
Thanks
Kishon
>