Re: [PATCH v14 04/11] PCI: kirin: Add support for bridge slot DT schema

[Mauro Carvalho Chehab <mchehab@kernel.org>]

Hi Bjorn,

Em Tue, 24 May 2022 12:19:47 -0500
Bjorn Helgaas <helgaas@kernel.org> escreveu:

> On Tue, Oct 19, 2021 at 07:06:41AM +0100, Mauro Carvalho Chehab wrote:
> > On HiKey970, there's a PEX 8606 PCI bridge on its PHY with
> > 6 lanes. Only 4 lanes are connected:
> > 
> > 	lane 0 - connected to Kirin 970;
> > 	lane 4 - M.2 slot;
> > 	lane 5 - mini PCIe slot;
> > 	lane 6 - in-board Ethernet controller.
> > 
> > Each lane has its own PERST# gpio pin, and needs a clock
> > request.
> > 
> > Add support to parse a DT schema containing the above data.
> > 
> > Cc: Kishon Vijay Abraham I <kishon@ti.com>
> > Acked-by: Xiaowei Song <songxiaowei@hisilicon.com>
> > Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
> > ---
> > 
> > See [PATCH v14 00/11] at: https://lore.kernel.org/all/cover.1634622716.git.mchehab+huawei@kernel.org/
> > 
> >  drivers/pci/controller/dwc/pcie-kirin.c | 262 +++++++++++++++++++++---
> >  1 file changed, 231 insertions(+), 31 deletions(-)
> > 
> > diff --git a/drivers/pci/controller/dwc/pcie-kirin.c b/drivers/pci/controller/dwc/pcie-kirin.c
> > index 86c13661e02d..de375795a3b8 100644
> > --- a/drivers/pci/controller/dwc/pcie-kirin.c
> > +++ b/drivers/pci/controller/dwc/pcie-kirin.c
> > @@ -52,6 +52,19 @@
> >  #define PCIE_DEBOUNCE_PARAM	0xF0F400
> >  #define PCIE_OE_BYPASS		(0x3 << 28)
> >  
> > +/*
> > + * Max number of connected PCI slots at an external PCI bridge
> > + *
> > + * This is used on HiKey 970, which has a PEX 8606 bridge with has
> > + * 4 connected lanes (lane 0 upstream, and the other tree lanes,
> > + * one connected to an in-board Ethernet adapter and the other two
> > + * connected to M.2 and mini PCI slots.
> > + *
> > + * Each slot has a different clock source and uses a separate PERST#
> > + * pin.
> > ...  
> 
> > +static int kirin_pcie_add_bus(struct pci_bus *bus)
> > +{
> > +	struct dw_pcie *pci = to_dw_pcie_from_pp(bus->sysdata);
> > +	struct kirin_pcie *kirin_pcie = to_kirin_pcie(pci);
> > +	int i, ret;
> > +
> > +	if (!kirin_pcie->num_slots)
> > +		return 0;
> > +
> > +	/* Send PERST# to each slot */
> > +	for (i = 0; i < kirin_pcie->num_slots; i++) {
> > +		ret = gpio_direction_output(kirin_pcie->gpio_id_reset[i], 1);
> > +		if (ret) {
> > +			dev_err(pci->dev, "PERST# %s error: %d\n",
> > +				kirin_pcie->reset_names[i], ret);
> > +		}
> > +	}
> > +	usleep_range(PERST_2_ACCESS_MIN, PERST_2_ACCESS_MAX);
> > +
> > +	return 0;
> > +}
> > +
> > +
> >  static struct pci_ops kirin_pci_ops = {
> >  	.read = kirin_pcie_rd_own_conf,
> >  	.write = kirin_pcie_wr_own_conf,
> > +	.add_bus = kirin_pcie_add_bus,  
> 
> This seems a little weird.  Can you educate me?
> 
> From [1], it looks like the topology here is:
> 
>   00:00.0 Root Port
>   01:00.0 PEX 8606 Upstream Port
>   02:01.0 PEX 8606 Downstream Port
>   02:04.0 PEX 8606 Downstream Port
>   02:05.0 PEX 8606 Downstream Port
>   02:07.0 PEX 8606 Downstream Port
>   02:09.0 PEX 8606 Downstream Port
>   06:00.0 RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller 
> 
> The .add_bus() callback will be called for *every* child bus we want
> to enumerate.  So if any of those PEX 8606 Downstream Ports are
> connected to another switch, when we enumerate the secondary buses of
> that other switch, it looks like we'll send PERST# to all the slots
> again, which doesn't sound right.  Am I missing something?

The implementation made on Kirin 960/970 for PCI is to not have a
PERST# bus. Instead, it has one independent GPIO driving the PERST#
signal for each single individual port that is physically connected.

See the schematics at:

	- https://www.96boards.org/documentation/consumer/hikey/hikey970/hardware-docs/files/hikey970-schematics.pdf
	- https://github.com/96boards/documentation/blob/master/consumer/hikey/hikey960/hardware-docs/HiKey960_Schematics.pdf

Yet, my first proposal were to send all GPIOs altogether. See, for
instance:

	https://lore.kernel.org/all/3acf2c073e8ea67efaae91074dda0763bf7a2ab9.1626768323.git.mchehab+huawei@kernel.org/

There, the PERST# signals are initialized altogether, at the end
of hi3670_pcie_phy_power_on():

	/* perst assert Endpoints */
	usleep_range(21000, 23000);
	for (i = 0; i < phy->n_gpio_resets; i++) {
		ret = gpio_direction_output(phy->gpio_id_reset[i], 1);
		if (ret)
			return ret;
	}
	usleep_range(10000, 11000);

	ret = is_pipe_clk_stable(phy);
	if (!ret)
		goto disable_clks;

	hi3670_pcie_set_eyeparam(phy);

	ret = hi3670_pcie_noc_power(phy, false);
	if (ret)
		goto disable_clks;

During the review process, I was requested to change it in order to do it
via .add_bus. On my tests at the boards, I didn't see the same PERST#
signal to hit more than once, and the driver is working fine. 
So, this wasn't an actual issue, as far as I can tell. 

That's what it ended getting merged upstream.

I don't mind if somewone wants to return it to use the previous approach of
sending all per-port PERST# signals at the same time, just before calling 
is_pipe_clk_stable(), like the above. However, I don't have access to the 
hardware anymore, so I can't test any patches for it.

Thanks,
Mauro