Re: Re: [PATCH V2 3/5] PCI: tegra: Add support for PCIe endpoint mode in Tegra194

From: Vidya Sagar
Date: Mon Jan 13 2020 - 13:12:47 EST




On 1/6/20 6:32 PM, Thierry Reding wrote:
On Fri, Jan 03, 2020 at 06:14:02PM +0530, Vidya Sagar wrote:
Add support for the endpoint mode of Synopsys DesignWare core based
dual mode PCIe controllers present in Tegra194 SoC.

Signed-off-by: Vidya Sagar <vidyas@xxxxxxxxxx>
---
V2:
* Addressed Bjorn's review comments
* Made changes as part of addressing review comments for other patches

drivers/pci/controller/dwc/Kconfig | 30 +-
drivers/pci/controller/dwc/pcie-tegra194.c | 782 ++++++++++++++++++++-
2 files changed, 796 insertions(+), 16 deletions(-)

diff --git a/drivers/pci/controller/dwc/pcie-tegra194.c b/drivers/pci/controller/dwc/pcie-tegra194.c
[...]
@@ -411,11 +466,66 @@ static irqreturn_t tegra_pcie_rp_irq_handler(struct tegra_pcie_dw *pcie)
return IRQ_HANDLED;
}
+static irqreturn_t tegra_pcie_ep_irq_handler(struct tegra_pcie_dw *pcie)
+{
+ struct dw_pcie_ep *ep = &pcie->pci.ep;
+ int spurious = 1;
+ u32 val, tmp;
+
+ val = appl_readl(pcie, APPL_INTR_STATUS_L0);
+ if (val & APPL_INTR_STATUS_L0_LINK_STATE_INT) {
+ val = appl_readl(pcie, APPL_INTR_STATUS_L1_0_0);
+ appl_writel(pcie, val, APPL_INTR_STATUS_L1_0_0);
+ if (val & APPL_INTR_STATUS_L1_0_0_HOT_RESET_DONE) {
+ /* clear any stale PEX_RST interrupt */
+ if (!kfifo_put(&pcie->event_fifo, EP_HOT_RST_DONE)) {
+ dev_err(pcie->dev, "EVENT FIFO is full\n");
+ return IRQ_HANDLED;
+ }
+ wake_up(&pcie->wq);
+ }

Overall this patch looks a little cluttered. A few blank lines before
and after (or between) block statements would help make this more
readable, in my opinion.
Done.


+ if (val & APPL_INTR_STATUS_L1_0_0_RDLH_LINK_UP_CHGED) {
+ tmp = appl_readl(pcie, APPL_LINK_STATUS);
+ if (tmp & APPL_LINK_STATUS_RDLH_LINK_UP) {
+ dev_info(pcie->dev, "Link is up with Host\n");

Do we want this to be an info message? Looks to me like this is mostly
useful for debug purposes, as a quick way to check if the link is up.
For production use, this would perhaps be better to expose as a sysfs
attribute so that userspace can query it at runtime rather than search
through kernel logs.
I changed it to dev_dbg message.


+ dw_pcie_ep_linkup(ep);
+ }
+ }
+ spurious = 0;
+ }
+
+ if (val & APPL_INTR_STATUS_L0_PCI_CMD_EN_INT) {
+ val = appl_readl(pcie, APPL_INTR_STATUS_L1_15);
+ appl_writel(pcie, val, APPL_INTR_STATUS_L1_15);
+ if (val & APPL_INTR_STATUS_L1_15_CFG_BME_CHGED) {
+ if (!kfifo_put(&pcie->event_fifo, EP_BME_CHANGE)) {
+ dev_err(pcie->dev, "EVENT FIFO is full\n");
+ return IRQ_HANDLED;
+ }
+ wake_up(&pcie->wq);
+ }
+ spurious = 0;
+ }
+
+ if (spurious) {
+ dev_warn(pcie->dev, "Random interrupt (STATUS = 0x%08X)\n",
+ val);
+ appl_writel(pcie, val, APPL_INTR_STATUS_L0);
+ }
+
+ return IRQ_HANDLED;
+}
+
static irqreturn_t tegra_pcie_irq_handler(int irq, void *arg)
{
struct tegra_pcie_dw *pcie = arg;
- return tegra_pcie_rp_irq_handler(pcie);
+ if (pcie->mode == DW_PCIE_RC_TYPE)
+ return tegra_pcie_rp_irq_handler(pcie);
+ else if (pcie->mode == DW_PCIE_EP_TYPE)
+ return tegra_pcie_ep_irq_handler(pcie);
+
+ return IRQ_NONE;
}

We already know at probe time whether the controller is in root complex
or endpoint mode, right? Couldn't we just install the correct handler
rather than multiplex here? It's not a very big deal, but given that
these are interrupts, avoiding the additional indirection might be a
good idea.
Done.


[...]
@@ -986,6 +1115,42 @@ static int tegra_pcie_dw_parse_dt(struct tegra_pcie_dw *pcie)
pcie->enable_cdm_check =
of_property_read_bool(np, "snps,enable-cdm-check");
+ if (pcie->mode == DW_PCIE_RC_TYPE)
+ return 0;
+
+ /* Endpoint mode specific DT entries */
+ name = devm_kasprintf(pcie->dev, GFP_KERNEL,
+ "tegra_pcie_%u_pex_rst_gpio", pcie->cid);
+ if (!name) {
+ dev_err(pcie->dev, "Failed to create PERST GPIO string\n");
+ return -ENOMEM;
+ }
+ pcie->pex_rst_gpiod = devm_gpiod_get_from_of_node(pcie->dev, np,

If np == pcie->dev.of_node, you can simply use devm_gpiod_get() here,
can't you?
Done.


[...]
+static void pex_ep_event_hot_rst_done(struct tegra_pcie_dw *pcie)
+{
+ u32 val = 0;

The initialization here seems unnecessary.
Done.


[...]
+static int tegra_pcie_ep_raise_legacy_irq(struct tegra_pcie_dw *pcie, u16 irq)
+{
+ /* Tegra194 supports only INTA */
+ if (irq > 1)
+ return -EINVAL;
+
+ appl_writel(pcie, 1, APPL_LEGACY_INTX);
+ mdelay(1);

Spinning for 1 ms these days is quite a lot. Does this have to be a busy
loop or could you use something like usleep_range(1000, 2000) to allow
the CPU to do something else in the meantime?

Also, does the legacy INTX pulse have to be a whole millisecond wide? Or
could this be shorter? A one millisecond pulse implies a maximum of 1000
interrupts per second, which seems a bit low.
1 ms is what all the other implementations also are using and they are using mdelay also :( . But, I think this doesn't have to be a busy loop and I'll change it to usleep_range(1000, 2000).


@@ -1440,6 +2178,12 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev)
int ret;
u32 i;
+ match = of_match_device(of_match_ptr(tegra_pcie_dw_of_match), dev);
+ if (!match)
+ return -EINVAL;
+
+ data = (struct tegra_pcie_dw_of_data *)match->data;

of_device_get_match_data()?
Done.


+
pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
if (!pcie)
return -ENOMEM;
@@ -1449,6 +2193,7 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev)
pci->ops = &tegra_dw_pcie_ops;
pp = &pci->pp;
pcie->dev = &pdev->dev;
+ pcie->mode = (enum dw_pcie_device_mode)data->mode;
ret = tegra_pcie_dw_parse_dt(pcie);
if (ret < 0) {
@@ -1462,6 +2207,9 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev)
return ret;
}
+ if (pcie->pex_refclk_sel_gpiod)
+ gpiod_set_value(pcie->pex_refclk_sel_gpiod, 1);
+
pcie->pex_ctl_supply = devm_regulator_get(dev, "vddio-pex-ctl");
if (IS_ERR(pcie->pex_ctl_supply)) {
ret = PTR_ERR(pcie->pex_ctl_supply);
@@ -1570,11 +2318,24 @@ static int tegra_pcie_dw_probe(struct platform_device *pdev)
platform_set_drvdata(pdev, pcie);
- ret = tegra_pcie_config_rp(pcie);
- if (ret && ret != -ENOMEDIUM)
- goto fail;
- else
- return 0;
+ switch (pcie->mode) {
+ case DW_PCIE_RC_TYPE:
+ ret = tegra_pcie_config_rp(pcie);
+ if (ret && ret != -ENOMEDIUM)
+ goto fail;
+ else
+ return 0;
+ break;
+
+ case DW_PCIE_EP_TYPE:
+ ret = tegra_pcie_config_ep(pcie, pdev);
+ if (ret < 0)
+ goto fail;
+ break;
+
+ default:
+ dev_err(dev, "Invalid PCIe device type %d\n", pcie->mode);
+ }
fail:
tegra_bpmp_put(pcie->bpmp);
@@ -1593,6 +2354,8 @@ static int tegra_pcie_dw_remove(struct platform_device *pdev)
pm_runtime_put_sync(pcie->dev);
pm_runtime_disable(pcie->dev);
tegra_bpmp_put(pcie->bpmp);
+ if (pcie->pex_refclk_sel_gpiod)
+ gpiod_set_value(pcie->pex_refclk_sel_gpiod, 0);
return 0;
}
@@ -1697,13 +2460,6 @@ static void tegra_pcie_dw_shutdown(struct platform_device *pdev)
__deinit_controller(pcie);
}
-static const struct of_device_id tegra_pcie_dw_of_match[] = {
- {
- .compatible = "nvidia,tegra194-pcie",
- },
- {},
-};
-

No need to move this around if you use of_device_get_match_data().
Yes.

Thanks for the review.
Vidya Sagar

Thierry

static const struct dev_pm_ops tegra_pcie_dw_pm_ops = {
.suspend_late = tegra_pcie_dw_suspend_late,
.suspend_noirq = tegra_pcie_dw_suspend_noirq,
--
2.17.1