On 2018-07-31 11:59 AM, Russell King - ARM Linux wrote:
For the thread associated with this patch set, a review of a previous
patch for ARM posted last Tuesday on this subject asked a series of
questions about the PCI-nature of this. The review has not been
responded to.
We are dealing with this now. More appropriate to do it this time than before having reworked set.
If it is inappropriate to offer RapidIO for any architecture that
happens to has PCI, then it is inappropriate to offer it for any
ARM machine that happens to have PCI.
It is completely appropriate to use RapidIO on any architecture that has PCI/PCIe using existing PCIe-to-SRIO host bridges. Works well with Marvell and NVIDIA boards.
Confusion here is caused by the fact that there are ARM and non-ARM devices that offer on-chip RapidIO host controllers as well as PCIe. E.g. TI Keystone/KeystoneII, FSL 85xx/86xx, Xilinx and Altera FPGAs with ARM cores, Cavium on MIPS. In most cases external buses are configurable and we have to address possible combinations.
I already posted some explanation in response to your earlier comment.
In light of the lack of explanation on this point so far, I'm naking
the ARM part of this series for now.
Explanations posted.
I also think that the HAS_RAPIDIO thing is misleading and needs
sorting out (as I've mentioned in other emails, including the one
I refer to above) before rapidio becomes available more widely.
Highly likely it is used right now in a base station of mobile operator near you :)
On Tue, Jul 31, 2018 at 10:29:48AM -0400, Alexei Colin wrote:
quoted
Resending the patchset from prior submission:
https://lkml.org/lkml/2018/7/30/911
The only change are the Cc tags in all patches now include the mailing
lists for all affected architectures, and patch 1/6 (which adds the menu
item to RapdidIO subsystem Kconfig) is CCed to all maintainers who are
getting this cover letter. The cover letter has been updated with
explanations to points raised in the feedback.
The top-level Kconfig entry for RapidIO subsystem is currently
duplicated in several architecture-specific Kconfig files. This set of
patches does two things:
1. Move the Kconfig menu definition into the RapidIO subsystem and
remove the duplicate definitions from arch Kconfig files.
2. Enable RapidIO Kconfig menu entry for arm and arm64 architectures,
where it was not enabled before. I tested that subsystem and drivers
build successfully for both architectures, and tested that the modules
load on a custom arm64 Qemu model.
For all architectures, RapidIO menu should be offered when either:
(1) The platform has a PCI bus (which host a RapidIO module on the bus).
(2) The platform has a RapidIO IP block (connected to a system bus, e.g.
AXI on ARM). In this case, 'select HAS_RAPIDIO' should be added to the
'config ARCH_*' menu entry for the SoCs that offer the IP block.
Prior to this patchset, different architectures used different criteria:
* powerpc: (1) and (2)
* mips: (1) and (2) after recent commit into next that added (2):
https://www.linux-mips.org/archives/linux-mips/2018-07/msg00596.html
fc5d988878942e9b42a4de5204bdd452f3f1ce47
491ec1553e0075f345fbe476a93775eabcbc40b6
* x86: (1)
* arm,arm64: none (RapidIO menus never offered)
This set of architectures are the ones that implement support for
59 more quoted lines