Date:   Tue, 15 Nov 2022 19:26:36 +0300
From:   Serge Semin <fancer.lancer@...il.com>
To:     Manivannan Sadhasivam <mani@...nel.org>
Cc:     Serge Semin <Sergey.Semin@...kalelectronics.ru>,
        Rob Herring <robh+dt@...nel.org>,
        Rob Herring <robh@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Bjorn Helgaas <bhelgaas@...gle.com>,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Cai Huoqing <cai.huoqing@...ux.dev>,
        Robin Murphy <robin.murphy@....com>,
        Jingoo Han <jingoohan1@...il.com>,
        Gustavo Pimentel <gustavo.pimentel@...opsys.com>,
        Lorenzo Pieralisi <lpieralisi@...nel.org>,
        Krzysztof Wilczyński <kw@...ux.com>,
        Alexey Malahov <Alexey.Malahov@...kalelectronics.ru>,
        Pavel Parkhomenko <Pavel.Parkhomenko@...kalelectronics.ru>,
        Frank Li <Frank.Li@....com>,
        Manivannan Sadhasivam <manivannan.sadhasivam@...aro.org>,
        caihuoqing <caihuoqing@...du.com>, Vinod Koul <vkoul@...nel.org>,
        linux-pci@...r.kernel.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v7 20/20] PCI: dwc: Add Baikal-T1 PCIe controller support

On Mon, Nov 14, 2022 at 11:41:55PM +0530, Manivannan Sadhasivam wrote:
> On Mon, Nov 14, 2022 at 02:20:59PM +0300, Serge Semin wrote:
> > On Mon, Nov 14, 2022 at 01:01:35PM +0530, Manivannan Sadhasivam wrote:
> > > On Sun, Nov 13, 2022 at 10:13:01PM +0300, Serge Semin wrote:
> > > > Baikal-T1 SoC is equipped with DWC PCIe v4.60a host controller. It can be
> > > > trained to work up to Gen.3 speed over up to x4 lanes. The host controller
> > > > is attached to the DW PCIe 3.0 PCS via the PIPE-4 interface, which in its
> > > > turn is connected to the DWC 10G PHY. The whole system is supposed to be
> > > > fed up with four clock sources: DBI peripheral clock, AXI application
> > > > clocks and external PHY/core reference clock generating the 100MHz signal.
> > > > In addition to that the platform provide a way to reset each part of the
> > > > controller: sticky/non-sticky bits, host controller core, PIPE interface,
> > > > PCS/PHY and Hot/Power reset signal. The driver also provides a way to
> > > > handle the GPIO-based PERST# signal.
> > > > 
> > > > Note due to the Baikal-T1 MMIO peculiarity we have to implement the DBI
> > > > interface accessors which make sure the IO operations are dword-aligned.
> > > > 
> > > > Signed-off-by: Serge Semin <Sergey.Semin@...kalelectronics.ru>
> > > > 
> > > > ---
> > > > 
> > > > Changelog v2:
> > > > - Rename 'syscon' property to 'baikal,bt1-syscon'.
> > > > 
> > > > Changelog v3:
> > > > - Use the clocks/resets handlers defined in the DW PCIe core descriptor.
> > > >   (@Rob)
> > > > - Redefine PCI host bridge config space accessors with the generic
> > > >   pci_generic_config_read32() and pci_generic_config_write32() methods.
> > > >   (@Rob)
> > > > 
> > > > Changelog v4:
> > > > - Drop PCIBIOS_* macros usage. (@Rob)
> > > > - Add "static const" to the dw_pcie_ops and dw_pcie_host_ops structure
> > > >   instances. (@Bjorn)
> > > > - Rename bt1_pcie_dw_ops to bt1_pcie_ops. (@Bjorn)
> > > > - Rename bt1_pcie_ops to bt1_pci_ops. (@Bjorn)
> > > > - Use start_link/stop_link suffixes in the corresponding callbacks.
> > > >   (@Bjorn)
> > > > - Change the get_res() method suffix to being get_resources(). (@Bjorn)
> > > > - Change *_{add,del}_dw_port() method to *_{add,del}_port(). (@Bjorn)
> > > > - Drop dma_coerce_mask_and_coherent() applied to the PCI host bridge
> > > >   kernel device instance. (@Bjorn)
> > > > - Add the comment above the dma_set_mask_and_coherent() method usage
> > > >   regarding the controller eDMA feature. (@Bjorn)
> > > > - Fix the comment above the core reset controls assertion. (@Bjorn)
> > > > - Replace delays and timeout numeric literals with macros. (@Bjorn)
> > > > 
> > > > Changelog v6:
> > > > - Move the DMA-mask setup to the eDMA driver. (@Robin)
> > > > 
> > > > Changelog v7:
> > > > - Replace if-then-dev_err_probe-return statement with just
> > > > return-dev_err_probe one.
> > > > ---
> > > >  drivers/pci/controller/dwc/Kconfig    |   9 +
> > > >  drivers/pci/controller/dwc/Makefile   |   1 +
> > > >  drivers/pci/controller/dwc/pcie-bt1.c | 643 ++++++++++++++++++++++++++
> > > >  3 files changed, 653 insertions(+)
> > > >  create mode 100644 drivers/pci/controller/dwc/pcie-bt1.c
> > > > 
> > > > diff --git a/drivers/pci/controller/dwc/Kconfig b/drivers/pci/controller/dwc/Kconfig
> > > > index 62ce3abf0f19..771b8b146623 100644
> > > > --- a/drivers/pci/controller/dwc/Kconfig
> > > > +++ b/drivers/pci/controller/dwc/Kconfig
> > > > @@ -222,6 +222,15 @@ config PCIE_ARTPEC6_EP
> > > >  	  Enables support for the PCIe controller in the ARTPEC-6 SoC to work in
> > > >  	  endpoint mode. This uses the DesignWare core.
> > > >  
> > > > +config PCIE_BT1
> > > > +	tristate "Baikal-T1 PCIe controller"
> > > 
> > 
> > > Wondering why cannot this be "PCIE_BAIKAL"? Are you sure that this same driver
> > > cannot be reused for other Baikal SoCs in future?
> > 
> > Well, there are at least two SoCs: Baikal-M1 and Baikal-S1, which
> > comprise the Synopsys DW PCIe Host IP-core on boards. But both of them
> > have different versions of the controller (4.70a and 5.40a, meanwhile
> > Baikal-T1 has 4.60a) and the clocks/reset/link
> > enable/disable/establish procedures are also different. So I have much
> > doubt we should be adding a support for all of them in a single driver
> > because the only common part for them most likely will be just the
> > probe and remove methods.) Thus having a generic driver name in the
> > kernel will cause a confusion (or will require so submit a pre-requisite
> > config/driver renaming patch) should we decide to submit the drivers
> > for the new controllers.
> > 
> 

> Most of the PCIe IPs out there have a single driver for a family/manufacturer.
> Unless the IP changes drastically (like a different core), we add a separate
> driver for that.

You say 'most', which means not all of them. In anyway I have much
doubt that the "single driver per family/manufacture" rule should be
followed all the time. The fact that it works in a lot of cases
doesn't mean it gets to be general. In fact it depends on the
platform/device implementation. Even if two devices have an interface
synthesized from the same IP-core the synthesize parameters can be
very much different, the platform-setting can differ too. In that case
there is no point in having those two devices handled by the same
driver, because that would just needlessly overcomplicate the code
with a sophisticated abstraction layer (like a full HAL). So it's
better to just split the code up following the KISS principal. It will
cause having simpler individual drivers, make them more readable and
better maintainable, will simplify the parallel development process.
IMO of course.

> 
> If you look at the Qcom driver, we have clubbed the support for dozens of SoCs
> that differ by clock/resets/link and each will be identified by a separate
> devicetree compatible. Here all the SoCs have synopsys based IP but only their
> resources are different, so grouping them together in a single driver makes
> sense.

Not only the resource, but the initialization procedures too.
Basically all DW PCIe LLDD contains: request resources,
platform-related inits, device-specific inits, and some of them even
IRQs handling. If all of that is absolutely different, then I don't
see much reason in having the combined driver even if the device is
produced by the same manufacturer.

Anyway I'd suggest to leave the driver in subject as is for now. If at
the moment of the other Baikal-* SoCs PCIe controller drivers
open-sourcing we discover that some part of this driver can be reused,
then we'll make sure this driver is converted to a more generic
version (Kconfig name, driver name, functions/data names) and submit
an update-patch with the new controller support. Currently I'd rather
have a more coherent driver in the kernel than a driver with generic
names but working with only one version of the controller.

> 
> > > 
> > > > +	depends on MIPS_BAIKAL_T1 || COMPILE_TEST
> > > > +	depends on PCI_MSI_IRQ_DOMAIN
> > > > +	select PCIE_DW_HOST
> > > > +	help
> > > > +	  Enables support for the PCIe controller in the Baikal-T1 SoC to work
> > > > +	  in host mode. It's based on the Synopsys DWC PCIe v4.60a IP-core.
> > > > +
> > > >  config PCIE_ROCKCHIP_DW_HOST
> > > >  	bool "Rockchip DesignWare PCIe controller"
> > > >  	select PCIE_DW
> > > > diff --git a/drivers/pci/controller/dwc/Makefile b/drivers/pci/controller/dwc/Makefile
> > > > index 8ba7b67f5e50..bf5c311875a1 100644
> > > > --- a/drivers/pci/controller/dwc/Makefile
> > > > +++ b/drivers/pci/controller/dwc/Makefile
> > > > @@ -3,6 +3,7 @@ obj-$(CONFIG_PCIE_DW) += pcie-designware.o
> > > >  obj-$(CONFIG_PCIE_DW_HOST) += pcie-designware-host.o
> > > >  obj-$(CONFIG_PCIE_DW_EP) += pcie-designware-ep.o
> > > >  obj-$(CONFIG_PCIE_DW_PLAT) += pcie-designware-plat.o
> > > > +obj-$(CONFIG_PCIE_BT1) += pcie-bt1.o
> > > >  obj-$(CONFIG_PCI_DRA7XX) += pci-dra7xx.o
> > > >  obj-$(CONFIG_PCI_EXYNOS) += pci-exynos.o
> > > >  obj-$(CONFIG_PCIE_FU740) += pcie-fu740.o
> > > > diff --git a/drivers/pci/controller/dwc/pcie-bt1.c b/drivers/pci/controller/dwc/pcie-bt1.c
> > > > new file mode 100644
> > > > index 000000000000..3346770e6654
> > > > --- /dev/null
> > > > +++ b/drivers/pci/controller/dwc/pcie-bt1.c
> > > > @@ -0,0 +1,643 @@
> 
> [...]
> 
> > > > +/*
> > > > + * Baikal-T1 MMIO space must be read/written by the dword-aligned
> > > > + * instructions. Note the methods are optimized to have the dword operations
> > > > + * performed with minimum overhead as the most frequently used ones.
> > > > + */
> > > > +static int bt1_pcie_read_mmio(void __iomem *addr, int size, u32 *val)
> > > > +{
> > > > +	unsigned int ofs = (uintptr_t)addr & 0x3;
> > > > +
> > > > +	if (!IS_ALIGNED((uintptr_t)addr, size))
> > > > +		return -EINVAL;
> > > > +
> > > > +	*val = readl(addr - ofs) >> ofs * BITS_PER_BYTE;
> > > 
> > 
> > > Why can't you use the _relaxed variants?
> > 
> > As a part of a nitpick fix I could, but in this case I don't think
> > it's very much necessary and IMO it still can be dangerous, since the
> > IO-accessors utilization is hidden behind the wrapper, which then is
> > used not only in the LLDD, but in the generic driver too. So depending
> > on the DW PCIe core driver implementation the strong ordering might be
> > required if not at the current stage, but in future. So I'd rather be on
> > the safe side in this case especially seeing it won't give us much
> > performance gain at runtime since the method is mainly used during the
> > probe/initialization process.
> > 
> 

> Well, I don't see any danger in making this as the relaxed version and that's
> why asked. For the safe side of things, we could always use the non-relaxed
> version everywhere ;)

There are in general many nasty and hard to fix bugs might be caused
due to using the relaxed-version of the accessors. Let's take the DW
PCIe driver for instance. The DW PCIe host driver performs the
DBI-access protected by the spin-lock in the methods
dw_pci_bottom_mask()/dw_pci_bottom_unmask(). In accordance with
"Documentation/memory-barriers.txt" the "relaxed" method don't
guarantee the ordering with respect to locking. It doesn't guarantee
an ordering of the parallel accesses to the same IO-address either. It
means the actual IO may be completed after the lock is released, due
to which a concurrent code may see an outdated value if it
reads/writes from/to the same IO-address in the framework of the
critical section. Such code may work most of the time until at some
point you see that some IRQ line isn't masked/unmasked even after the
corresponding mask/unmask method is executed. Moreover absolutely the
same code using the relaxed-version can work on one platform and fail
on another due to many things: IO-bus implementation,
platform-specific locks design, relaxed IO-accessors implementation,
etc (though in this context we are talking about a single platform).

In this particular case you suggest to update the method which is
utilized exactly for the DBI-accesses. Thus if we did as you say at
some random point it might work not as expected depending on the
context like described above. So on the second thought answering to
your original question, no, I can't use the "relaxed"-version here at
the very least due to the reason described above and taking into
account that the driver core methods dw_pcie_read()/dw_pcie_write()
don't use them too.

Please also note all the DW PCIe LLDD (histb, exynos, kirin) defining
the custom DBI-accessors use the strictest IO-accessors version in the
same way as the DW PCIe core driver.

-Sergey

> 
> > > 
> > > > +	if (size == 4) {
> > > > +		return 0;
> > > > +	} else if (size == 2) {
> > > > +		*val &= 0xffff;
> > > > +		return 0;
> > > > +	} else if (size == 1) {
> > > > +		*val &= 0xff;
> > > > +		return 0;
> > > > +	}
> > > > +
> > > > +	return -EINVAL;
> > > > +}
> > > > +
> > > 
> > > [...]
> > > 
> > > > +/*
> > > > + * Implements the cold reset procedure in accordance with the reference manual
> > > > + * and available PM signals.
> > > > + */
> > > > +static int bt1_pcie_cold_start_bus(struct bt1_pcie *btpci)
> > > > +{
> > > > +	struct device *dev = btpci->dw.dev;
> > > > +	struct dw_pcie *pci = &btpci->dw;
> > > > +	u32 val;
> > > > +	int ret;
> > > > +
> > > > +	/* First get out of the Power/Hot reset state */
> > > > +	ret = reset_control_deassert(pci->core_rsts[DW_PCIE_PWR_RST].rstc);
> > > > +	if (ret) {
> > > > +		dev_err(dev, "Failed to deassert PHY reset\n");
> > > > +		return ret;
> > > > +	}
> > > > +
> > > > +	ret = reset_control_deassert(pci->core_rsts[DW_PCIE_HOT_RST].rstc);
> > > > +	if (ret) {
> > > > +		dev_err(dev, "Failed to deassert hot reset\n");
> > > > +		goto err_assert_pwr_rst;
> > > > +	}
> > > > +
> > > > +	/* Wait for the PM-core to stop requesting the PHY reset */
> > > 
> > 
> > > What is PM core here? By first look I thought you are referring to Linux PM
> > > core framework.
> > 
> > See the DW PCIe HW-manual. The IP-core has it's own PM-controller.
> > 
> 
> Oh, I was not aware of that...
> 
> > > 
> > > > +	ret = regmap_read_poll_timeout(btpci->sys_regs, BT1_CCU_PCIE_RSTC, val,
> > > > +				       !(val & BT1_CCU_PCIE_REQ_PHY_RST),
> > > > +				       BT1_PCIE_REQ_DELAY_US, BT1_PCIE_REQ_TIMEOUT_US);
> > > > +	if (ret) {
> > > > +		dev_err(dev, "Timed out waiting for PM to stop PHY resetting\n");
> > > 
> > > With relation to my above comment, this log might be confusing.
> > 
> > See above.
> 
> [...]
> 
> > > > +static const struct of_device_id bt1_pcie_of_match[] = {
> > > > +	{ .compatible = "baikal,bt1-pcie" },
> > > > +	{},
> > > > +};
> > > > +MODULE_DEVICE_TABLE(of, bt1_pcie_of_match);
> > > > +
> > > > +static struct platform_driver bt1_pcie_driver = {
> > > > +	.probe = bt1_pcie_probe,
> > > > +	.remove = bt1_pcie_remove,
> > > > +	.driver = {
> > > > +		.name	= "bt1-pcie",
> > > > +		.of_match_table = bt1_pcie_of_match,
> > > 
> > 
> > > You might also want to add PROBE_ASYNCHRONOUS flag to allow parallel probing of
> > > drivers while the dwc core waits for PHY link to be up in dw_pcie_wait_for_link().
> > 
> > Thanks for reminding me about that flag (though it's
> > PROBE_PREFER_ASYNCHRONOUS).
> 
> Ah, yes!
> 
> > I was thinking to add it after getting
> > read the Rob' comment here
> > https://patchwork.kernel.org/project/linux-pci/patch/20220913101237.4337-1-vidyas@nvidia.com/#25035943
> > But then successfully forgot about it. It works well on our platform
> > and even saves us of 0.5 seconds of the bootup time if no device is
> > attached to the PCIe controller. No kidding, it's indeed good
> > suggestion since the whole bootup time is of about 3 seconds. So we'll
> > be able to reduce it for about 13%. I'll provide this update on v8.
> > 
> 
> Cool!
> 
> Thanks,
> Mani
> 
> > -Sergey
> > 
> > > 
> > > Thanks,
> > > Mani 
> > > 
> > > > +	},
> > > > +};
> > > > +module_platform_driver(bt1_pcie_driver);
> > > > +
> > > > +MODULE_AUTHOR("Serge Semin <Sergey.Semin@...kalelectronics.ru>");
> > > > +MODULE_DESCRIPTION("Baikal-T1 PCIe driver");
> > > > +MODULE_LICENSE("GPL");
> > > > -- 
> > > > 2.38.1
> > > > 
> > > > 
> > > 
> > > -- 
> > > மணிவண்ணன் சதாசிவம்
> 
> -- 
> மணிவண்ணன் சதாசிவம்

Powered by blists - more mailing lists