Re: [PATCH v8 3/9] usb: dwc3: core: Access XHCI address space temporarily to read port info

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 5/16/2023 8:32 PM, Krishna Kurapati PSSNV wrote:


On 5/16/2023 5:41 PM, Johan Hovold wrote:
On Sun, May 14, 2023 at 11:19:11AM +0530, Krishna Kurapati wrote:
Currently host-only capable DWC3 controllers support Multiport.
Temporarily map XHCI address space for host-only controllers and parse
XHCI Extended Capabilities registers to read number of usb2 ports and
usb3 ports present on multiport controller. Each USB Port is at least HS
capable.

The port info for usb2 and usb3 phy are identified as num_usb2_ports
and num_usb3_ports. The intention is as follows:

Wherever we need to perform phy operations like:

LOOP_OVER_NUMBER_OF_AVAILABLE_PORTS()
{
    phy_set_mode(dwc->usb2_generic_phy[i], PHY_MODE_USB_HOST);
    phy_set_mode(dwc->usb3_generic_phy[i], PHY_MODE_USB_HOST);
}

If number of usb2 ports is 3, loop can go from index 0-2 for
usb2_generic_phy. If number of usb3-ports is 2, we don't know for sure,
if the first 2 ports are SS capable or some other ports like (2 and 3)
are SS capable. So instead, num_usb2_ports is used to loop around all
phy's (both hs and ss) for performing phy operations. If any
usb3_generic_phy turns out to be NULL, phy operation just bails out.

num_usb3_ports is used to modify GUSB3PIPECTL registers while setting up
phy's as we need to know how many SS capable ports are there for this.

Signed-off-by: Krishna Kurapati <quic_kriskura@xxxxxxxxxxx>
---
  drivers/usb/dwc3/core.c | 113 ++++++++++++++++++++++++++++++++++++++++
  drivers/usb/dwc3/core.h |  17 +++++-
  2 files changed, 129 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
index 0beaab932e7d..e983aef1fb93 100644
--- a/drivers/usb/dwc3/core.c
+++ b/drivers/usb/dwc3/core.c
@@ -1767,6 +1767,104 @@ static int dwc3_get_clocks(struct dwc3 *dwc)
      return 0;
  }
+/**
+ * dwc3_xhci_find_next_ext_cap - Find the offset of the extended capabilities
+ *                    with capability ID id.
+ *
+ * @base:    PCI MMIO registers base address.
+ * @start:    address at which to start looking, (0 or HCC_PARAMS to start at
+ *        beginning of list)
+ * @id:        Extended capability ID to search for, or 0 for the next
+ *        capability
+ *
+ * Returns the offset of the next matching extended capability structure. + * Some capabilities can occur several times, e.g., the XHCI_EXT_CAPS_PROTOCOL,
+ * and this provides a way to find them all.
+ */
+static int dwc3_xhci_find_next_ext_cap(void __iomem *base, u32 start, int id)
+{
+    u32 val;
+    u32 next;
+    u32 offset;
+
+    offset = start;
+    if (!start || start == XHCI_HCC_PARAMS_OFFSET) {
+        val = readl(base + XHCI_HCC_PARAMS_OFFSET);
+        if (val == ~0)
+            return 0;
+        offset = XHCI_HCC_EXT_CAPS(val) << 2;
+        if (!offset)
+            return 0;
+    }
+    do {
+        val = readl(base + offset);
+        if (val == ~0)
+            return 0;
+        if (offset != start && (id == 0 || XHCI_EXT_CAPS_ID(val) == id))
+            return offset;
+
+        next = XHCI_EXT_CAPS_NEXT(val);
+        offset += next << 2;
+    } while (next);
+
+    return 0;
+}

You should not make another copy of xhci_find_next_ext_cap(), but rather
use it directly.

We already have drivers outside of usb/host using this function so it
should be fine to do the same for now:

    #include "../host/xhci-ext-caps.h"

Hi Johan,

  This was the approach which we followed when we first introduced the patch [1]. But Thinh suggested to duplicate code so that we can avoid any dependency on xhci (which seems to be right). So since its just one function, I duplicated it here.

Hi Thinh,

  Would like to know your opinion here on how to proceed further.

Regards,
Krishna,

+static int dwc3_read_port_info(struct dwc3 *dwc)
+{
+    void __iomem        *regs;

Call this one 'base' instead.

+    u32            offset;
+    u32            temp;

I see that the xhci driver use 'temp' for this, but I'd prefer 'val'.

+    u8            major_revision;
+    int            ret = 0;
+
+    /*
+     * Remap xHCI address space to access XHCI ext cap regs,
+     * since it is needed to get port info.
+     */
+    regs = ioremap(dwc->xhci_resources[0].start,
+                resource_size(&dwc->xhci_resources[0]));
+    if (IS_ERR(regs))
+        return PTR_ERR(regs);
+
+    offset = dwc3_xhci_find_next_ext_cap(regs, 0,
+                    XHCI_EXT_CAPS_PROTOCOL);
+    while (offset) {

This would be better implemented as a do-while loop (cf.
xdbc_reset_debug_port()).

+        temp = readl(regs + offset);
+        major_revision = XHCI_EXT_PORT_MAJOR(temp);
+
+        temp = readl(regs + offset + 0x08);

We should try to avoid magic constants, but I see that we already have
cases accessing these fields like this.

+        if (major_revision == 0x03) {
+            dwc->num_usb3_ports += XHCI_EXT_PORT_COUNT(temp);
+        } else if (major_revision <= 0x02) {
+            dwc->num_usb2_ports += XHCI_EXT_PORT_COUNT(temp);
+        } else {
+            dev_err(dwc->dev,
+                "Unrecognized port major revision %d\n", major_revision);

Please add a line break after the string.

Perhaps this should be handles as in xhci core by simply warning and
continuing instead.

I broke the loop and went to unmap as we are not sure what values would be read. Any use of continuing ?

+            ret = -EINVAL;
+            goto unmap_reg;
+        }
+
+        offset = dwc3_xhci_find_next_ext_cap(regs, offset,
+                        XHCI_EXT_CAPS_PROTOCOL);
+    }
+
+    temp = readl(regs + DWC3_XHCI_HCSPARAMS1);
+    if (HCS_MAX_PORTS(temp) != (dwc->num_usb3_ports + dwc->num_usb2_ports)) {
+        dev_err(dwc->dev,
+            "Mismatched reported MAXPORTS (%d)\n", HCS_MAX_PORTS(temp));
+        ret = -EINVAL;
+        goto unmap_reg;
+    }

Not sure this is needed either.

Could this risk regressing platforms which does not have currently have
all PHYs described in DT?

No, it doesn't. AFAIK, this only tells how many ports are present as per the core consultant configuration of the device. I tried to explain what would happen incase phy's are not present in DT in [2] & [3].

You do however need to make sure that both num_usb<n>_ports is no larger
than MAX_PORTS_SUPPORTED to avoid memory corruption when you're adding
fixed sized arrays for the PHYs later in the series.

+
+    dev_dbg(dwc->dev,
+        "hs-ports: %d ss-ports: %d\n", dwc->num_usb2_ports, dwc->num_usb3_ports);

Use %u for unsigned values.

And please try to stay within 80 columns.

Thanks for catching this potential bug. Will add that if check as well in v9.

+
+unmap_reg:
+    iounmap(regs);
+    return ret;
+}
+
  static int dwc3_probe(struct platform_device *pdev)
  {
      struct device        *dev = &pdev->dev;
@@ -1774,6 +1872,7 @@ static int dwc3_probe(struct platform_device *pdev)
      void __iomem        *regs;
      struct dwc3        *dwc;
      int            ret;
+    unsigned int        hw_mode;
      dwc = devm_kzalloc(dev, sizeof(*dwc), GFP_KERNEL);
      if (!dwc)
@@ -1843,6 +1942,20 @@ static int dwc3_probe(struct platform_device *pdev)
              goto err_disable_clks;
      }
+    /*
+     * Currently DWC3 controllers that are host-only capable
+     * support Multiport

Are you missing an "only" after "Currently" above?
> Please add a full stop.

+     */
+    hw_mode = DWC3_GHWPARAMS0_MODE(dwc->hwparams.hwparams0);
+    if (hw_mode == DWC3_GHWPARAMS0_MODE_HOST) {
+        ret = dwc3_read_port_info(dwc);
+        if (ret)
+            goto err_disable_clks;
+    } else {
+        dwc->num_usb2_ports = 1;
+        dwc->num_usb3_ports = 1;
+    }
+
      spin_lock_init(&dwc->lock);
      mutex_init(&dwc->mutex);
diff --git a/drivers/usb/dwc3/core.h b/drivers/usb/dwc3/core.h
index d56457c02996..d3401963bc27 100644
--- a/drivers/usb/dwc3/core.h
+++ b/drivers/usb/dwc3/core.h
@@ -35,6 +35,17 @@
  #define DWC3_MSG_MAX    500
+/* Define XHCI Extcap register offsets for getting multiport info */
+#define XHCI_HCC_PARAMS_OFFSET    0x10
+#define DWC3_XHCI_HCSPARAMS1    0x04
+#define XHCI_EXT_CAPS_PROTOCOL    2
+#define XHCI_HCC_EXT_CAPS(x)    (((x) >> 16) & 0xffff)
+#define XHCI_EXT_CAPS_ID(x)     (((x) >> 0) & 0xff)
+#define XHCI_EXT_CAPS_NEXT(x)   (((x) >> 8) & 0xff)
+#define XHCI_EXT_PORT_MAJOR(x)  (((x) >> 24) & 0xff)
+#define XHCI_EXT_PORT_COUNT(x)  (((x) >> 8) & 0xff)
+#define HCS_MAX_PORTS(x)        (((x) >> 24) & 0x7f)
+

You should use the xhci defines instead of these copies too.

  /* Global constants */
  #define DWC3_PULL_UP_TIMEOUT    500    /* ms */
  #define DWC3_BOUNCE_SIZE    1024    /* size of a superspeed bulk */
@@ -1025,6 +1036,8 @@ struct dwc3_scratchpad_array {
   * @usb_psy: pointer to power supply interface.
   * @usb2_phy: pointer to USB2 PHY
   * @usb3_phy: pointer to USB3 PHY
+ * @num_usb2_ports: number of usb2 ports.
+ * @num_usb3_ports: number of usb3 ports.

Use upper case "USBn" and drop the full stops for consistency.

Please move these after the PHY structures.

   * @usb2_generic_phy: pointer to USB2 PHY
   * @usb3_generic_phy: pointer to USB3 PHY
   * @phys_ready: flag to indicate that PHYs are ready
@@ -1162,6 +1175,9 @@ struct dwc3 {
      struct usb_phy        *usb2_phy;
      struct usb_phy        *usb3_phy;
+    u8            num_usb2_ports;
+    u8            num_usb3_ports;
+
      struct phy        *usb2_generic_phy;
      struct phy        *usb3_generic_phy;
@@ -1649,5 +1665,4 @@ static inline int dwc3_ulpi_init(struct dwc3 *dwc)
  static inline void dwc3_ulpi_exit(struct dwc3 *dwc)
  { }
  #endif
-

This is an unrelated change that should be dropped.

  #endif /* __DRIVERS_USB_DWC3_CORE_H */

Johan

[1]: https://lore.kernel.org/all/20230310163420.7582-3-quic_kriskura@xxxxxxxxxxx/

[2]: https://lore.kernel.org/all/4eb26a54-148b-942f-01c6-64e66541de8b@xxxxxxxxxxx/

[3]: https://lore.kernel.org/all/966c1001-6d64-9163-0c07-96595156fc8c@xxxxxxxxxxx/

Thanks for the review and comments 🙂. Will make sure to fix them in v9.

Regards,
Krishna,



[Index of Archives]     [Linux Media]     [Linux Input]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Old Linux USB Devel Archive]

  Powered by Linux