On Saturday 01 October 2022 18:00:25 Lukas Wunner wrote:I did that a few years ago (rejected), but can attest to the robustness of the pcie hotplug code on non-hotplug slots.
Adding Marek, Pali & Jon to cc as they've worked on somewhat similar
On Sat, Oct 01, 2022 at 12:57:43AM +0530, Vidya Sagar wrote:
To support the Hot-plug feature, PCIe spec has a well-defined model for
hardware implementation and software programming interface. There are also
some architectures/platforms where the Hot-plug feature is implemented in a
non-standard way and software support for the respective implementations is
available with the kernel. This patch series attempts to add support for one
such non-standard way of supporting the Hot-plug feature where a single GPIO
is used to detect and report the Hot-Plug and Unplug events to the SW.
The platforms that can use this piece of software need to have GPIO routed
from the slot to the controller which can indicate the presence/absence of
the downstream device through its state. This GPIO should also have the
capability to interrupt the system when the connection/disconnection event
A GPIO Hot-plug framework is written which looks for a "hotplug-gpios" named
GPIO entry in the corresponding device-tree entry of the controller and
registers a hot-pluggable slot with the Hot-plug framework.
The platform drivers of the PCIe host bridges/root ports can register with the
aforementioned GPIO Hot-Plug framework along with ops to perform any platform
specific tasks during Hot-Plug/Unplug events.
Oza Pawandeep made an attempt to upstream support for a similar Hot-plug
feature implementation at a platform level, but the implementation as such
was very specific to that platform (at least the way I understood it).
This current series also attempts to address that by extracting out all the
common code to do with GPIO and Hot-plug core framework and expecting the
platform drivers to only register/unregister with the GPIO framework. So,
@Oza, could you try using the GPIO framework from this series and enable
Hot-plug support for your platform if it still makes sense?
Would not it better to rather synthesise PCIe Slot Capabilities support
in your PCIe Root Port device (e.g. via pci-bridge-emul.c) and then let
existing PCI hotplug code to take care for hotplugging? Because it
already implements all required stuff for re-scanning, registering and
unregistering PCIe devices for Root Ports with Slot Capabilities. And I
think that there is no need to have just another (GPIO based)
implementation of PCI hotplug.
Similar thing Marek and me have implemented for PCIe link state events
in patch series with Lukas pointed.
Regarding the DT documentation change to add about 'hotplug-gpios, I'm not
sure if pci.txt is the right place or the dt-schema repository
But, in the interest of keeping all the changes related to this feature in the
the same repository, I made the changes to the pci.txt file in this repo itself.
Please let me know if the documentation change needs to be moved to the other
The Changes have been tested on the Tegra234 platform.
Vidya Sagar (4):
dt-bindings: Add "hotplug-gpios" PCIe property
PCI/hotplug: Add GPIO PCIe hotplug driver
PCI: tegra194: Add support to configure a pluggable slot
PCI: tegra194: Enable GPIO based Hot-Plug support
Documentation/devicetree/bindings/pci/pci.txt | 4 +
drivers/pci/controller/dwc/pcie-tegra194.c | 85 +++++++-
drivers/pci/hotplug/Kconfig | 11 +
drivers/pci/hotplug/Makefile | 1 +
drivers/pci/hotplug/gpio_php.c | 200 ++++++++++++++++++
drivers/pci/hotplug/gpiophp.h | 40 ++++
6 files changed, 334 insertions(+), 7 deletions(-)
create mode 100644 drivers/pci/hotplug/gpio_php.c
create mode 100644 drivers/pci/hotplug/gpiophp.h