Hi greg k-h,
在 2023/6/9 下午5:25, Greg Kroah-Hartman 写道:
On Fri, Jun 09, 2023 at 05:07:03PM +0800, zhuyinbo wrote:
在 2023/6/9 下午2:13, Greg Kroah-Hartman 写道:
On Fri, Jun 09, 2023 at 10:50:47AM +0800, Yinbo Zhu wrote:
The dwc2 driver has everything we need to run in PCI mode except
for pci_device_id driver_data parse. With that to set Loongson
dwc2 element and added identified as PCI_VENDOR_ID_LOONGSON
and PCI_DEVICE_ID_LOONGSON_DWC2 in dwc2_pci_ids, the Loongson
dwc2 controller will work.
Signed-off-by: Yinbo Zhu <zhuyinbo@xxxxxxxxxxx>
---
Change in v2:
1. Move the dwc2 pci ID from pci_ids.h to params.c.
2. Add some code logic to ensure that the current device is
a PCI device.
3. Fix the compile issue when dwc2 pci driver as module.
drivers/usb/dwc2/core.h | 1 +
drivers/usb/dwc2/params.c | 39 ++++++++++++++++++++++++++++++++++++++-
drivers/usb/dwc2/pci.c | 14 +-------------
3 files changed, 40 insertions(+), 14 deletions(-)
diff --git a/drivers/usb/dwc2/core.h b/drivers/usb/dwc2/core.h
index 0bb4c0c845bf..c92a1da46a01 100644
--- a/drivers/usb/dwc2/core.h
+++ b/drivers/usb/dwc2/core.h
@@ -1330,6 +1330,7 @@ irqreturn_t dwc2_handle_common_intr(int irq, void *dev);
/* The device ID match table */
extern const struct of_device_id dwc2_of_match_table[];
extern const struct acpi_device_id dwc2_acpi_match[];
+extern const struct pci_device_id dwc2_pci_ids[];
int dwc2_lowlevel_hw_enable(struct dwc2_hsotg *hsotg);
int dwc2_lowlevel_hw_disable(struct dwc2_hsotg *hsotg);
diff --git a/drivers/usb/dwc2/params.c b/drivers/usb/dwc2/params.c
index 21d16533bd2f..6b68a8830781 100644
--- a/drivers/usb/dwc2/params.c
+++ b/drivers/usb/dwc2/params.c
@@ -7,9 +7,14 @@
#include <linux/module.h>
#include <linux/of_device.h>
#include <linux/usb/of.h>
+#include <linux/pci_ids.h>
+#include <linux/pci.h>
#include "core.h"
+#define PCI_PRODUCT_ID_HAPS_HSOTG 0xabc0
+#define PCI_DEVICE_ID_LOONGSON_DWC2 0x7a04
+
static void dwc2_set_bcm_params(struct dwc2_hsotg *hsotg)
{
struct dwc2_core_params *p = &hsotg->params;
@@ -55,6 +60,14 @@ static void dwc2_set_jz4775_params(struct dwc2_hsotg *hsotg)
!device_property_read_bool(hsotg->dev, "disable-over-current");
}
+static void dwc2_set_loongson_params(struct dwc2_hsotg *hsotg)
+{
+ struct dwc2_core_params *p = &hsotg->params;
+
+ p->phy_utmi_width = 8;
+ p->power_down = DWC2_POWER_DOWN_PARAM_PARTIAL;
+}
+
static void dwc2_set_x1600_params(struct dwc2_hsotg *hsotg)
{
struct dwc2_core_params *p = &hsotg->params;
@@ -281,6 +294,23 @@ const struct acpi_device_id dwc2_acpi_match[] = {
};
MODULE_DEVICE_TABLE(acpi, dwc2_acpi_match);
+const struct pci_device_id dwc2_pci_ids[] = {
+ {
+ PCI_DEVICE(PCI_VENDOR_ID_SYNOPSYS, PCI_PRODUCT_ID_HAPS_HSOTG),
+ },
+ {
+ PCI_DEVICE(PCI_VENDOR_ID_STMICRO,
+ PCI_DEVICE_ID_STMICRO_USB_OTG),
+ },
+ {
+ PCI_DEVICE(PCI_VENDOR_ID_LOONGSON, PCI_DEVICE_ID_LOONGSON_DWC2),
+ .driver_data = (unsigned long)dwc2_set_loongson_params,
+ },
+ { /* end: all zeroes */ }
+};
+MODULE_DEVICE_TABLE(pci, dwc2_pci_ids);
+EXPORT_SYMBOL_GPL(dwc2_pci_ids);
+
static void dwc2_set_param_otg_cap(struct dwc2_hsotg *hsotg)
{
switch (hsotg->hw_params.op_mode) {
@@ -927,13 +957,20 @@ int dwc2_init_params(struct dwc2_hsotg *hsotg)
if (match && match->data) {
set_params = match->data;
set_params(hsotg);
- } else {
+ } else if (!match) {
const struct acpi_device_id *amatch;
+ const struct pci_device_id *pmatch = NULL;
amatch = acpi_match_device(dwc2_acpi_match, hsotg->dev);
if (amatch && amatch->driver_data) {
set_params = (set_params_cb)amatch->driver_data;
set_params(hsotg);
+ } else if (!amatch)
+ pmatch = pci_match_id(dwc2_pci_ids, to_pci_dev(hsotg->dev->parent));
At this point in time, how can you guarantee that the parent device
really is a PCI one? This function is being called from a platform
device callback, and platform devices should NEVER be a child of a PCI
device, as that's not how PCI or platform devices work.
So how is this even possible?
confused,
Hi greg k-h,
My current considerations are based on that the dwc2 drivers support
three types of devices, they are Platform device / ACPI device / PCI
device, and The dwc2/platform.c can all cover the three type dwc2
device, no matter how it is registered.
So, when a dwc2 device wasn't platform device and acpi device and It
will be a PCI device, and the the dwc2/pci.c will register a dwc2
platform device that dwc2 device is a device that in dwc2/platform.c's
dwc2_driver_probe. the "&dwc2->dev->parent" is the "hsotg->dev->parent"
and it was also the &pci->dev.
That's wrong, a PCI device should NEVER register a platform device under
it, as obviously it's not a platform device at all.
It doesn't seems like that for current dwc2 code. The dwc2/pci.c driver
register a platform device called "dwc2" was to match dwc2/platform.c's
dwc2 driver called "dwc2" and ensure dwc2/platform's dwc2_driver_probe
function can be called. After all, the dwc2/platform's dwc2_driver_probe
is the one that truly initializes the dwc2 controller. In addition, the
same applies to dwc3 code.
Do we need to refactor all the dwc (dwc2/dwc3) code ?
So please work to fix that layering violation here, as that's not
correct and will cause problems as you are showing here.
So, We can use "to_pci_dev(hsotg->dev->parent)" to gain a dwc2 pci
device.
No, not if the parent is something else as you have no way of knowing
this at this point in time.
Do we need to support legacy dwc2 device drivers in the future ? which
legacy device driver was that dwc2 as a platform device but not support
dts. If that's not the case, I think my logic seems to guarantee it.
1) DWC2 PCI device driver (drivers/usb/dwc2/pci.c) :
static int dwc2_pci_probe(struct pci_dev *pci,
const struct pci_device_id *id)
{
...
struct platform_device *dwc2;
...
struct device *dev = &pci->dev;
...
dwc2 = platform_device_alloc("dwc2", PLATFORM_DEVID_AUTO);
//"dwc2" was used to match dwc2 platform driver
This needs to be fixed, it's not ok at all. Just use the real PCI
device here please.
...
dwc2->dev.parent = dev;
...
}
2) DWC2 PLATFROM driver (drivers/usb/dwc2/platform.c) :
// This driver cover pci type device.
static const char dwc2_driver_name[] = "dwc2";
static struct platform_driver dwc2_platform_driver = {
.driver = {
.name = dwc2_driver_name,
//"dwc2" was used to match pci type dwc2 device
...
},
.probe = dwc2_driver_probe,
.remove = dwc2_driver_remove,
.shutdown = dwc2_driver_shutdown,
};
static int dwc2_driver_probe(struct platform_device *dev)
{
// The dev is the "dwc2" platform device in dwc2/pci.c driver.
struct dwc2_hsotg *hsotg;
...
hsotg->dev = &dev->dev;
Yes, and if this was a real platform device, the parent would NOT be a
PCI device, and so your check will cause all sorts of problems, right?
When the "match" and "amatch" was all NULL and considering that the
current driver does not support legacy dwc2 device and the current
platform device seems that it must be created by dwc2/pci.c. Is there
anything else I don't know about ?
Thanks,
Yinbo
