Skip to contentSkip to conversation

USB host support for Raspberry Pi 4 board (64-bit)

PATCH New Sylwester Nawrocki 6 participants 26 messages May 25, 2020 – Jul 10, 2020 1 message not in the archive

Subsystems (10)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): CGME20200525114005eucas1p1aa7efc1c0e27e0eb3e5b48690f96e68d@eucas1p1.samsung.com

#1 lore raw

v4 cover USB host support for Raspberry Pi 4 board (64-bit)

Hi all,

This patch series adds USB host support for Raspberry Pi 4 board. It includes the Broadcom STB PCIe controller driver ported from Linux kernel, a memory mapping update for the xHCI controller on PCIe bus for 64-bit builds and some related fixes and updates in the usb/xhci and the pci driver core code.

The 32-bit ARM part has be excluded from the series and will be posted separately.

The difference comparing to previous version is only in the config update patch "configs: Enable support for the XHCI controller on RPI4 board (ARM
64-bit)".

The patch series is based on v2020.07-rc1 tree.

Thanks,
Sylwester

Marek Szyprowski (3):
rpi4: shorten a mapping for the DRAM
rpi4: add a mapping for the PCIe XHCI controller MMIO registers (ARM

    64bit)
  configs: Enable support for the XHCI controller on RPI4 board (ARM
    64-bit)

Nicolas Saenz Julienne (1):
linux/bitfield.h: Add primitives for manipulating bitfields both in
host- and fixed-endian

Sylwester Nawrocki (5):
usb: xhci: Add missing cache flush in the scratchpad array
initialization
usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq
pci: Move some PCIe register offset definitions to a common header
pci: Add some PCI Express capability register offset definitions
pci: Add driver for Broadcom BCM2711 SoC PCIe controller

 arch/arm/mach-bcm283x/init.c  |  20 +-
 configs/rpi_4_defconfig       |   9 +
 configs/rpi_arm64_defconfig   |   8 +-
 drivers/pci/Kconfig           |   9 +
 drivers/pci/Makefile          |   1 +
 drivers/pci/pci-rcar-gen3.c   |   8 -
 drivers/pci/pcie_brcmstb.c    | 623 ++++++++++++++++++++++++++++++++++++++++++
 drivers/pci/pcie_intel_fpga.c |   3 -
 drivers/usb/host/xhci-mem.c   |   3 +
 include/linux/bitfield.h      |  52 ++++
 include/pci.h                 |  22 +-
 include/usb/xhci.h            |   8 -
 12 files changed, 740 insertions(+), 26 deletions(-)
 create mode 100644 drivers/pci/pcie_brcmstb.c
Signature · 1 line
2.7.4
#2 lore raw

PATCH v4 1/9 usb: xhci: Add missing cache flush in the scratchpad array initialization

In current code there is no cache flush after initializing the scratchpad buffer array with the scratchpad buffer pointers. This leads to a failure of the "slot enable" command on the rpi4 board (Broadcom STB PCIe controller + VL805 USB hub) - the very first TRB transfer on the command ring fails and there is a timeout while waiting for the command completion event. After adding the missing cache flush everything seems to be working as expected.

---
Changes since v1:
- none.

---
 drivers/usb/host/xhci-mem.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/usb/host/xhci-mem.c +3 -0
index 93450ee..729bdc3 100644
@@ -393,6 +393,9 @@ static int xhci_scratchpad_alloc(struct xhci_ctrl *ctrl)
 		scratchpad->sp_array[i] = cpu_to_le64(ptr);
 	}
 
+	xhci_flush_cache((uintptr_t)scratchpad->sp_array,
+			 sizeof(u64) * num_sp);
+
 	return 0;
 
 fail_sp3:
Signature · 1 line
2.7.4
#3 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

In reply to Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

---
Changes since v1:
- none.
Changes since RFC:
- dropped Kconfig option, switched to not using readq/writeq unconditionally.

---
 include/usb/xhci.h | 8 --------
 1 file changed, 8 deletions(-)
include/usb/xhci.h +0 -8
index 6017504..c16106a 100644
@@ -1111,28 +1111,20 @@ static inline void xhci_writel(uint32_t volatile *regs, const unsigned int val)
  */
 static inline u64 xhci_readq(__le64 volatile *regs)
 {
-#if BITS_PER_LONG == 64
-	return readq(regs);
-#else
 	__u32 *ptr = (__u32 *)regs;
 	u64 val_lo = readl(ptr);
 	u64 val_hi = readl(ptr + 1);
 	return val_lo + (val_hi << 32);
-#endif
 }
 
 static inline void xhci_writeq(__le64 volatile *regs, const u64 val)
 {
-#if BITS_PER_LONG == 64
-	writeq(val, regs);
-#else
 	__u32 *ptr = (__u32 *)regs;
 	u32 val_lo = lower_32_bits(val);
 	/* FIXME */
 	u32 val_hi = upper_32_bits(val);
 	writel(val_lo, ptr);
 	writel(val_hi, ptr + 1);
-#endif
 }
 
 int xhci_hcd_init(int index, struct xhci_hccr **ret_hccr,
Signature · 1 line
2.7.4
#4 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Sylwester,

Quoted 27 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

---
Changes since v1:
- none.
Changes since RFC:
- dropped Kconfig option, switched to not using readq/writeq unconditionally.

Quoted 3 lines
---
 include/usb/xhci.h | 8 --------
 1 file changed, 8 deletions(-)

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Regards,
Simon

#5 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Simon,

Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Signature · 2 lines
Regards,
Sylwester
#6 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Sylwester,

Quoted 26 lines · Sylwester Nawrocki

Hi Simon,

Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

OK then it should be modelled as such.

How is this done in Linux?

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

You can always add the device to the devicetree if needed, and then you get a compatible string.

Regards,
Simon

#7 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Simon,

Quoted 31 lines · Simon Glass
Quoted 23 lines · Sylwester Nawrocki
Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

Quoted 3 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

OK, I imagined something like that.

Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

Signature · 2 lines
Regards,
Sylwester
#8 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Sylwester,

Quoted 35 lines · Sylwester Nawrocki

Hi Simon,

Quoted 31 lines · Simon Glass
Quoted 29 lines · Sylwester Nawrocki
Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 6 lines
Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

Well the USB maintainer (Marek) might be OK with that, not sure.

Quoted 4 lines

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

In xhci_pci_probe() you can look at the PCI vendor/device with something like:

struct pci_child_platdata *plat = dev_get_parent_platdata(dev); //
see comments for that struct in pci.h

int quirks = 0;
if (plat->vendor == xxx && plat->device == xxx)
quirks |= SOMETHING
xhci_register(...., quirks); // add a new param

in xhci_register() you can store the quirk in ctrl.

Quoted 5 lines
Quoted 4 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

Quoted 7 lines

OK, I imagined something like that.

Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

So long as you get the 'reg' property correct (i.e. same bus, device, function) then you are OK. See pci-info.rst for docs

Regards,
Simon

#9 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Simon,

Quoted 29 lines · Simon Glass
Quoted 28 lines · Sylwester Nawrocki
Quoted 27 lines · Simon Glass
Quoted 26 lines · Sylwester Nawrocki
Quoted 11 lines · Simon Glass
Quoted 10 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 8 lines
Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 6 lines
Quoted 5 lines
Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

Well the USB maintainer (Marek) might be OK with that, not sure.

32bit access is always safe according to the XHCI specification and such one is always used by the Linux kernel. 64bit access might give a few CPU cycles of performance improvement, but in case of typical u-boot use case this is simply negligible. Do we really need to make the code over-engineered for no good reason?

Quoted 19 lines
Quoted 3 lines

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

In xhci_pci_probe() you can look at the PCI vendor/device with something like:

struct pci_child_platdata *plat = dev_get_parent_platdata(dev); //
see comments for that struct in pci.h

int quirks = 0;
if (plat->vendor == xxx && plat->device == xxx)

    quirks |= SOMETHING

xhci_register(...., quirks); // add a new param

in xhci_register() you can store the quirk in ctrl.

Quoted 3 lines
Quoted 3 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

Quoted 8 lines
Quoted 6 lines

OK, I imagined something like that.

Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

So long as you get the 'reg' property correct (i.e. same bus, device, function) then you are OK. See pci-info.rst for docs

Best regards

Signature · 2 lines
Marek Szyprowski, PhD
Samsung R&D Institute Poland
#10 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi,

Quoted 36 lines · Marek Szyprowski
Quoted 34 lines · Simon Glass
Quoted 33 lines · Sylwester Nawrocki
Quoted 27 lines · Simon Glass
Quoted 20 lines · Sylwester Nawrocki
Quoted 11 lines · Simon Glass
Quoted 10 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 8 lines
Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

Quoted 7 lines
Quoted 1 line

Well the USB maintainer (Marek) might be OK with that, not sure.

32bit access is always safe according to the XHCI specification and such one is always used by the Linux kernel. 64bit access might give a few CPU cycles of performance improvement, but in case of typical u-boot use case this is simply negligible. Do we really need to make the code over-engineered for no good reason?

I didn't have a chance to measure it but I believe in practice there would be no any significant performance improvement. There are only few xhci_readq, xhci_writeq calls in the code and most of them are in a one off initialization procedure. The only one that would matter is in xhci_acknowledge_event(), but if we add a new (ctrl) argument to xhci_{readq,writeq} and testing of the quirk it will likely be not much different as if we just used two 32-bit reads/writes unconditionally, as far as performance is concerned.

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

In xhci_pci_probe() you can look at the PCI vendor/device with something like:

struct pci_child_platdata *plat = dev_get_parent_platdata(dev); //
see comments for that struct in pci.h

We need to test the PCI bridge so we could use dev->parent rather than dev here. I'm not sure if the 64-bit data access issue is inherent to the PCI controller itself or rather to its integration logic within the SoC (the SCB bridge). In other words, if in other Broadcom SoC same PCI vendor/ device controller might not need the quirk. I guess such approach would be good enough.

The question is whether we want to introduce the quirks mechanism now for the readq/writeq issue. My feeling is that it's better to align with what Linux does and the quirks could be useful for other purposes.

Quoted 9 lines
Quoted 9 lines

int quirks = 0;
if (plat->vendor == xxx && plat->device == xxx)

    quirks |= SOMETHING

xhci_register(...., quirks); // add a new param

in xhci_register() you can store the quirk in ctrl.

Quoted 3 lines
Quoted 3 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

Quoted 8 lines
Quoted 8 lines
Quoted 6 lines

OK, I imagined something like that.

Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

So long as you get the 'reg' property correct (i.e. same bus, device, function) then you are OK. See pci-info.rst for docs

--
Regards,
Sylwester

#11 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

In reply to Simon Glass

Hi Simon,

Quoted 45 lines · Simon Glass

Hi Sylwester,

Quoted 41 lines · Sylwester Nawrocki

Hi Simon,

Quoted 31 lines · Simon Glass
Quoted 23 lines · Sylwester Nawrocki
Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

This was discussed during review of the previous version of this patch, and I think aligning to what Linux does is fine. Previously we discussed adding an Kconfig option to control this, but I feel that's not good. Having a quirk flag to detect this is a dynamic approach, compared to the static Kconfig option, but overall I don't see a need not to align with Linux xHCI driver.

Quoted 28 lines

Well the USB maintainer (Marek) might be OK with that, not sure.

Quoted 4 lines

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

In xhci_pci_probe() you can look at the PCI vendor/device with something like:

struct pci_child_platdata *plat = dev_get_parent_platdata(dev); //
see comments for that struct in pci.h

int quirks = 0;
if (plat->vendor == xxx && plat->device == xxx)
quirks |= SOMETHING
xhci_register(...., quirks); // add a new param

in xhci_register() you can store the quirk in ctrl.

Quoted 7 lines
Quoted 4 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

OK, I imagined something like that.

Quoted 9 lines
Quoted 5 lines
Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

So long as you get the 'reg' property correct (i.e. same bus, device, function) then you are OK. See pci-info.rst for docs

Regards,
Bin

#12 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Quoted 49 lines · Bin Meng

Hi Simon,

Quoted 45 lines · Simon Glass

Hi Sylwester,

Quoted 41 lines · Sylwester Nawrocki

Hi Simon,

Quoted 31 lines · Simon Glass
Quoted 23 lines · Sylwester Nawrocki
Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

Quoted 7 lines

This was discussed during review of the previous version of this patch, and I think aligning to what Linux does is fine. Previously we discussed adding an Kconfig option to control this, but I feel that's not good. Having a quirk flag to detect this is a dynamic approach, compared to the static Kconfig option, but overall I don't see a need not to align with Linux xHCI driver.

My understanding is, that we will keep the approach of 32-bit accesses. My plan was to take the whole series through my treen.

Bin is that fine with you?

Regards,
Matthias

Quoted 41 lines
Quoted 28 lines

Well the USB maintainer (Marek) might be OK with that, not sure.

Quoted 4 lines

And the quirks in the generic PCI XHCI driver are based on the PCI vendor and the PCI device ID, so it's not helpful. I couldn't find any reference to the parent PCI bridge there.

In xhci_pci_probe() you can look at the PCI vendor/device with something like:

struct pci_child_platdata *plat = dev_get_parent_platdata(dev); //
see comments for that struct in pci.h

int quirks = 0;
if (plat->vendor == xxx && plat->device == xxx)
quirks |= SOMETHING
xhci_register(...., quirks); // add a new param

in xhci_register() you can store the quirk in ctrl.

Quoted 7 lines
Quoted 4 lines

You can add a quirk in the PCI controller and then XHCI can check its parent's platdata to see the flag, perhaps, since the parent will always be UCLASS_PCI.

OK, I imagined something like that.

Quoted 9 lines
Quoted 5 lines
Quoted 2 lines

You can always add the device to the devicetree if needed, and then you get a compatible string.

Will have a look, I wasn't aware we could add a node just for such purpose without negative side effects.

So long as you get the 'reg' property correct (i.e. same bus, device, function) then you are OK. See pci-info.rst for docs

Regards,
Bin

#13 lore raw

PATCH v4 2/9 usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq

Hi Matthias,

Quoted 61 lines · Matthias Brugger
Quoted 49 lines · Bin Meng

Hi Simon,

Quoted 45 lines · Simon Glass

Hi Sylwester,

Quoted 41 lines · Sylwester Nawrocki

Hi Simon,

Quoted 31 lines · Simon Glass
Quoted 23 lines · Sylwester Nawrocki
Quoted 18 lines · Simon Glass
Quoted 11 lines · Sylwester Nawrocki

There might be hardware configurations where 64-bit data accesses to XHCI registers are not supported properly. This patch removes the readq/writeq so always two 32-bit accesses are used to read/write 64-bit XHCI registers, similarly as it is done in Linux kernel.

This patch fixes operation of the XHCI controller on RPI4 Broadcom BCM2711 SoC based board, where the VL805 USB XHCI controller is connected to the PCIe Root Complex, which is attached to the system through the SCB bridge.

Quoted 5 lines

Even though the architecture is 64-bit the PCIe BAR is 32-bit and likely the 64-bit wide register accesses initiated by the CPU are not properly translated to a sequence of 32-bit PCIe accesses. xhci_readq(), for example, always returns same value in upper and lower 32-bits, e.g. 0xabcd1234abcd1234 instead of 0x00000000abcd1234.

Quoted 3 lines

Then I think this should be done with a quirk flag, enabled for this particular device via the compatible string. It should not be an #if, but an if().

Quoted 6 lines

Thanks for your comments. I will check and see how this could be done. It might not be so straightforward since the XHCI controller is a PCI device matched by the pci_device_id so we would need to be looking at the compatible string of the PCI controller to set the quirk in the xhci layer. It's the PCI bridge that introduces the limitation, not the VL805 XHCI controller chip.

Quoted 3 lines

OK then it should be modelled as such.

How is this done in Linux?

In Linux simply always two 32-bit accesses are used for 64-bit registers read/write.

Quoted 7 lines

This was discussed during review of the previous version of this patch, and I think aligning to what Linux does is fine. Previously we discussed adding an Kconfig option to control this, but I feel that's not good. Having a quirk flag to detect this is a dynamic approach, compared to the static Kconfig option, but overall I don't see a need not to align with Linux xHCI driver.

Quoted 4 lines

My understanding is, that we will keep the approach of 32-bit accesses. My plan was to take the whole series through my treen.

Bin is that fine with you?

No problem. Thanks!

Regards,
Bin

#14 lore raw

PATCH v4 3/9 pci: Move some PCIe register offset definitions to a common header

In reply to Sylwester Nawrocki

Some PCI Express register offsets are currently defined in multiple drivers, move them to a common header to avoid re-definitions and as a pre-requisite for adding new PCIe driver.
While at it replace some spaces with tabs.

---
Changes since v1:
- none.
Changes since RFC:
- whitespace clean up.

---
 drivers/pci/pci-rcar-gen3.c   |  8 --------
 drivers/pci/pcie_intel_fpga.c |  3 ---
 include/pci.h                 | 13 +++++++++++--
 3 files changed, 11 insertions(+), 13 deletions(-)
drivers/pci/pci-rcar-gen3.c +0 -8
index 30eff67..393f1c9 100644
@@ -117,14 +117,6 @@
 #define RCAR_PCI_MAX_RESOURCES	4
 #define MAX_NR_INBOUND_MAPS	6
 
-#define PCI_EXP_FLAGS		2		/* Capabilities register */
-#define PCI_EXP_FLAGS_TYPE	0x00f0		/* Device/Port type */
-#define PCI_EXP_TYPE_ROOT_PORT	0x4		/* Root Port */
-#define PCI_EXP_LNKCAP		12		/* Link Capabilities */
-#define PCI_EXP_LNKCAP_DLLLARC	0x00100000	/* Data Link Layer Link Active Reporting Capable */
-#define PCI_EXP_SLTCAP		20		/* Slot Capabilities */
-#define PCI_EXP_SLTCAP_PSN	0xfff80000	/* Physical Slot Number */
-
 enum {
 	RCAR_PCI_ACCESS_READ,
 	RCAR_PCI_ACCESS_WRITE,
drivers/pci/pcie_intel_fpga.c +0 -3
index 6a9f29c..69363a0 100644
@@ -65,9 +65,6 @@
 #define IS_ROOT_PORT(pcie, bdf)				\
 		((PCI_BUS(bdf) == pcie->first_busno) ? true : false)
 
-#define PCI_EXP_LNKSTA		18	/* Link Status */
-#define PCI_EXP_LNKSTA_DLLLA	0x2000	/* Data Link Layer Link Active */
-
 /**
  * struct intel_fpga_pcie - Intel FPGA PCIe controller state
  * @bus: Pointer to the PCI bus
include/pci.h +11 -2
index aff56b2..dfdbb32 100644
@@ -471,10 +471,19 @@
 #define  PCI_EA_FIELD_MASK	0xfffffffc	/* For Base & Max Offset */
 
 /* PCI Express capabilities */
+#define PCI_EXP_FLAGS		2	/* Capabilities register */
+#define  PCI_EXP_FLAGS_TYPE	0x00f0	/* Device/Port type */
+#define  PCI_EXP_TYPE_ROOT_PORT 0x4	/* Root Port */
 #define PCI_EXP_DEVCAP		4	/* Device capabilities */
-#define  PCI_EXP_DEVCAP_FLR     0x10000000 /* Function Level Reset */
+#define  PCI_EXP_DEVCAP_FLR	0x10000000 /* Function Level Reset */
 #define PCI_EXP_DEVCTL		8	/* Device Control */
-#define  PCI_EXP_DEVCTL_BCR_FLR 0x8000  /* Bridge Configuration Retry / FLR */
+#define  PCI_EXP_DEVCTL_BCR_FLR	0x8000  /* Bridge Configuration Retry / FLR */
+#define PCI_EXP_LNKCAP		12	/* Link Capabilities */
+#define  PCI_EXP_LNKCAP_DLLLARC	0x00100000 /* Data Link Layer Link Active Reporting Capable */
+#define PCI_EXP_LNKSTA		18	/* Link Status */
+#define  PCI_EXP_LNKSTA_DLLLA	0x2000	/* Data Link Layer Link Active */
+#define PCI_EXP_SLTCAP		20	/* Slot Capabilities */
+#define  PCI_EXP_SLTCAP_PSN	0xfff80000 /* Physical Slot Number */
 
 /* Include the ID list */
 
Signature · 1 line
2.7.4
#15 lore raw

PATCH v4 4/9 rpi4: shorten a mapping for the DRAM

In reply to Sylwester Nawrocki

From: Marek Szyprowski <[email protected]>

Remove the overlap between DRAM and device's IO area.

---
Changes since v1:
- none.

---
 arch/arm/mach-bcm283x/init.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-bcm283x/init.c +1 -1
index 9966d6c..4295356 100644
@@ -38,7 +38,7 @@ static struct mm_region bcm2711_mem_map[] = {
 	{
 		.virt = 0x00000000UL,
 		.phys = 0x00000000UL,
-		.size = 0xfe000000UL,
+		.size = 0xfc000000UL,
 		.attrs = PTE_BLOCK_MEMTYPE(MT_NORMAL) |
 			 PTE_BLOCK_INNER_SHARE
 	}, {
Signature · 1 line
2.7.4
#16 lore raw

PATCH v4 5/9 rpi4: add a mapping for the PCIe XHCI controller MMIO registers (ARM 64bit)

In reply to Sylwester Nawrocki

From: Marek Szyprowski <[email protected]>

Create a non-cacheable mapping for the 0x600000000 physical memory region, where MMIO registers for the PCIe XHCI controller are instantiated by the PCIe bridge.

---
Changes since v3:
- none.
Changes since v2:
- fixed typo MAX_MAP_MAX_ENTRIES -> MEM_MAP_MAX_ENTRIES
Changes since v1:
- none.

---
 arch/arm/mach-bcm283x/init.c | 18 +++++++++++++++---
 1 file changed, 15 insertions(+), 3 deletions(-)
arch/arm/mach-bcm283x/init.c +15 -3
index 4295356..9f5bca3 100644
@@ -11,10 +11,15 @@
 #include <dm/device.h>
 #include <fdt_support.h>
 
+#define BCM2711_RPI4_PCIE_XHCI_MMIO_PHYS	0x600000000UL
+#define BCM2711_RPI4_PCIE_XHCI_MMIO_SIZE	0x800000UL
+
 #ifdef CONFIG_ARM64
 #include <asm/armv8/mmu.h>
 
-static struct mm_region bcm283x_mem_map[] = {
+#define MEM_MAP_MAX_ENTRIES (4)
+
+static struct mm_region bcm283x_mem_map[MEM_MAP_MAX_ENTRIES] = {
 	{
 		.virt = 0x00000000UL,
 		.phys = 0x00000000UL,
@@ -34,7 +39,7 @@ static struct mm_region bcm283x_mem_map[] = {
 	}
 };
 
-static struct mm_region bcm2711_mem_map[] = {
+static struct mm_region bcm2711_mem_map[MEM_MAP_MAX_ENTRIES] = {
 	{
 		.virt = 0x00000000UL,
 		.phys = 0x00000000UL,
@@ -49,6 +54,13 @@ static struct mm_region bcm2711_mem_map[] = {
 			 PTE_BLOCK_NON_SHARE |
 			 PTE_BLOCK_PXN | PTE_BLOCK_UXN
 	}, {
+		.virt = BCM2711_RPI4_PCIE_XHCI_MMIO_PHYS,
+		.phys = BCM2711_RPI4_PCIE_XHCI_MMIO_PHYS,
+		.size = BCM2711_RPI4_PCIE_XHCI_MMIO_SIZE,
+		.attrs = PTE_BLOCK_MEMTYPE(MT_DEVICE_NGNRNE) |
+			 PTE_BLOCK_NON_SHARE |
+			 PTE_BLOCK_PXN | PTE_BLOCK_UXN
+	}, {
 		/* List terminator */
 		0,
 	}
@@ -71,7 +83,7 @@ static void _rpi_update_mem_map(struct mm_region *pd)
 {
 	int i;
 
-	for (i = 0; i < 2; i++) {
+	for (i = 0; i < MEM_MAP_MAX_ENTRIES; i++) {
 		mem_map[i].virt = pd[i].virt;
 		mem_map[i].phys = pd[i].phys;
 		mem_map[i].size = pd[i].size;
Signature · 1 line
2.7.4
#17 lore raw

PATCH v4 6/9 linux/bitfield.h: Add primitives for manipulating bitfields both in host- and fixed-endian

In reply to Sylwester Nawrocki

From: Nicolas Saenz Julienne <[email protected]>

Imports Al Viro's original Linux commit 00b0c9b82663a, which contains an in depth explanation and two fixes from Johannes Berg:
e7d4a95da86e0 "bitfield: fix *_encode_bits()", 37a3862e12382 "bitfield: add u8 helpers".

[s.nawrocki: added empty lines between functions and macros]

---
Changes since v3:
- none.
Changes since v2:
- added some more missing whitespaces as pointed out by checkpatch.pl.
Changes since v1:
- added empty lines between functions and macros.
Changes since RFC:
- new patch.

---
 include/linux/bitfield.h | 52 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 52 insertions(+)
include/linux/bitfield.h +52 -0
index 8b9d6ff..7ad8b08 100644
@@ -103,4 +103,56 @@
 		(typeof(_mask))(((_reg) & (_mask)) >> __bf_shf(_mask));	\
 	})
 
+extern void __compiletime_error("value doesn't fit into mask")
+__field_overflow(void);
+extern void __compiletime_error("bad bitfield mask")
+__bad_mask(void);
+
+static __always_inline u64 field_multiplier(u64 field)
+{
+	if ((field | (field - 1)) & ((field | (field - 1)) + 1))
+		__bad_mask();
+	return field & -field;
+}
+
+static __always_inline u64 field_mask(u64 field)
+{
+	return field / field_multiplier(field);
+}
+
+#define ____MAKE_OP(type, base, to, from)				\
+static __always_inline __##type type##_encode_bits(base v, base field)	\
+{									\
+	if (__builtin_constant_p(v) && (v & ~field_mask(field)))	\
+		__field_overflow();					\
+	return to((v & field_mask(field)) * field_multiplier(field));	\
+}									\
+static __always_inline __##type type##_replace_bits(__##type old,	\
+					base val, base field)		\
+{									\
+	return (old & ~to(field)) | type##_encode_bits(val, field);	\
+}									\
+static __always_inline void type##p_replace_bits(__##type * p,		\
+					base val, base field)		\
+{									\
+	*p = (*p & ~to(field)) | type##_encode_bits(val, field);	\
+}									\
+static __always_inline base type##_get_bits(__##type v, base field)	\
+{									\
+	return (from(v) & field) / field_multiplier(field);		\
+}
+
+#define __MAKE_OP(size)							\
+	____MAKE_OP(le##size, u##size, cpu_to_le##size, le##size##_to_cpu) \
+	____MAKE_OP(be##size, u##size, cpu_to_be##size, be##size##_to_cpu) \
+	____MAKE_OP(u##size, u##size, ,)
+
+____MAKE_OP(u8, u8, ,)
+__MAKE_OP(16)
+__MAKE_OP(32)
+__MAKE_OP(64)
+
+#undef __MAKE_OP
+#undef ____MAKE_OP
+
 #endif
Signature · 1 line
2.7.4
#18 lore raw

PATCH v4 6/9 linux/bitfield.h: Add primitives for manipulating bitfields both in host- and fixed-endian

Quoted 10 lines · Sylwester Nawrocki

From: Nicolas Saenz Julienne <[email protected]>

Imports Al Viro's original Linux commit 00b0c9b82663a, which contains an in depth explanation and two fixes from Johannes Berg:
e7d4a95da86e0 "bitfield: fix *_encode_bits()", 37a3862e12382 "bitfield: add u8 helpers".

[s.nawrocki: added empty lines between functions and macros]

This seems to break compilation of
arch/arm/mach-uniphier/clk/pll-base-ld20.c

https://gitlab.denx.de/u-boot/custodians/u-boot-raspberrypi/-/jobs/121248

Nicolas, Sylwester, can you please have a look.

Regards,
Matthias

Quoted 75 lines

---
Changes since v3:
- none.
Changes since v2:
- added some more missing whitespaces as pointed out by checkpatch.pl.
Changes since v1:
- added empty lines between functions and macros.
Changes since RFC:
- new patch.

---
  include/linux/bitfield.h | 52 ++++++++++++++++++++++++++++++++++++++++++++++++
  1 file changed, 52 insertions(+)
include/linux/bitfield.h +52 -0
index 8b9d6ff..7ad8b08 100644
@@ -103,4 +103,56 @@
  		(typeof(_mask))(((_reg) & (_mask)) >> __bf_shf(_mask));	\
  	})
 
+extern void __compiletime_error("value doesn't fit into mask")
+__field_overflow(void);
+extern void __compiletime_error("bad bitfield mask")
+__bad_mask(void);
+
+static __always_inline u64 field_multiplier(u64 field)
+{
+	if ((field | (field - 1)) & ((field | (field - 1)) + 1))
+		__bad_mask();
+	return field & -field;
+}
+
+static __always_inline u64 field_mask(u64 field)
+{
+	return field / field_multiplier(field);
+}
+
+#define ____MAKE_OP(type, base, to, from)				\
+static __always_inline __##type type##_encode_bits(base v, base field)	\
+{									\
+	if (__builtin_constant_p(v) && (v & ~field_mask(field)))	\
+		__field_overflow();					\
+	return to((v & field_mask(field)) * field_multiplier(field));	\
+}									\
+static __always_inline __##type type##_replace_bits(__##type old,	\
+					base val, base field)		\
+{									\
+	return (old & ~to(field)) | type##_encode_bits(val, field);	\
+}									\
+static __always_inline void type##p_replace_bits(__##type * p,		\
+					base val, base field)		\
+{									\
+	*p = (*p & ~to(field)) | type##_encode_bits(val, field);	\
+}									\
+static __always_inline base type##_get_bits(__##type v, base field)	\
+{									\
+	return (from(v) & field) / field_multiplier(field);		\
+}
+
+#define __MAKE_OP(size)							\
+	____MAKE_OP(le##size, u##size, cpu_to_le##size, le##size##_to_cpu) \
+	____MAKE_OP(be##size, u##size, cpu_to_be##size, be##size##_to_cpu) \
+	____MAKE_OP(u##size, u##size, ,)
+
+____MAKE_OP(u8, u8, ,)
+__MAKE_OP(16)
+__MAKE_OP(32)
+__MAKE_OP(64)
+
+#undef __MAKE_OP
+#undef ____MAKE_OP
+
  #endif
#19 lore raw

PATCH v4 6/9 linux/bitfield.h: Add primitives for manipulating bitfields both in host- and fixed-endian

Quoted 14 lines · Matthias Brugger
Quoted 10 lines · Sylwester Nawrocki

From: Nicolas Saenz Julienne <[email protected]>

Imports Al Viro's original Linux commit 00b0c9b82663a, which contains an in depth explanation and two fixes from Johannes Berg:
? e7d4a95da86e0 "bitfield: fix *_encode_bits()",
? 37a3862e12382 "bitfield: add u8 helpers".

[s.nawrocki: added empty lines between functions and macros]
Quoted 7 lines

This seems to break compilation of
arch/arm/mach-uniphier/clk/pll-base-ld20.c

https://gitlab.denx.de/u-boot/custodians/u-boot-raspberrypi/-/jobs/121248

Nicolas, Sylwester, can you please have a look.

Just in case you can't access the URL:
https://pastebin.com/KhDHf8e1

Quoted 77 lines

Regards,
Matthias

Quoted 74 lines

---
Changes since v3:
? - none.
Changes since v2:
? - added some more missing whitespaces as pointed out by checkpatch.pl. Changes since v1:
? - added empty lines between functions and macros.
Changes since RFC:
? - new patch.
---
? include/linux/bitfield.h | 52 ++++++++++++++++++++++++++++++++++++++++++++++++ ? 1 file changed, 52 insertions(+)

include/linux/bitfield.h +52 -0
index 8b9d6ff..7ad8b08 100644
@@ -103,4 +103,56 @@
 ????????? (typeof(_mask))(((_reg) & (_mask)) >> __bf_shf(_mask));??? \
 ????? })
+extern void __compiletime_error("value doesn't fit into mask")
+__field_overflow(void);
+extern void __compiletime_error("bad bitfield mask")
+__bad_mask(void);
+
+static __always_inline u64 field_multiplier(u64 field)
+{
+??? if ((field | (field - 1)) & ((field | (field - 1)) + 1))
+??????? __bad_mask();
+??? return field & -field;
+}
+
+static __always_inline u64 field_mask(u64 field)
+{
+??? return field / field_multiplier(field);
+}
+
+#define ____MAKE_OP(type, base, to, from)??????????????? \
+static __always_inline __##type type##_encode_bits(base v, base field)??? \
+{??????????????????????????????????? \
+??? if (__builtin_constant_p(v) && (v & ~field_mask(field)))??? \
+??????? __field_overflow();??????????????????? \
+??? return to((v & field_mask(field)) * field_multiplier(field));??? \
+}??????????????????????????????????? \
+static __always_inline __##type type##_replace_bits(__##type old,??? \
+??????????????????? base val, base field)??????? \
+{??????????????????????????????????? \
+??? return (old & ~to(field)) | type##_encode_bits(val, field);??? \
+}??????????????????????????????????? \
+static __always_inline void type##p_replace_bits(__##type * p,??????? \
+??????????????????? base val, base field)??????? \
+{??????????????????????????????????? \
+??? *p = (*p & ~to(field)) | type##_encode_bits(val, field);??? \
+}??????????????????????????????????? \
+static __always_inline base type##_get_bits(__##type v, base field)??? \
+{??????????????????????????????????? \
+??? return (from(v) & field) / field_multiplier(field);??????? \
+}
+
+#define __MAKE_OP(size)??????????????????????????? \
+??? ____MAKE_OP(le##size, u##size, cpu_to_le##size, le##size##_to_cpu) \
+??? ____MAKE_OP(be##size, u##size, cpu_to_be##size, be##size##_to_cpu) \
+??? ____MAKE_OP(u##size, u##size, ,)
+
+____MAKE_OP(u8, u8, ,)
+__MAKE_OP(16)
+__MAKE_OP(32)
+__MAKE_OP(64)
+
+#undef __MAKE_OP
+#undef ____MAKE_OP
+

? #endif

#20 lore raw

PATCH 1/1 include/bitfield.h: include byteorder.h

This is needed in order to use the cpu_to_*()/*_to_cpu() family of functions.

---

IMO the ideal thing would be to squash this into the original patch.

 include/linux/bitfield.h | 1 +
 1 file changed, 1 insertion(+)
include/linux/bitfield.h +1 -0
index 7ad8b088ed..7ad02f8cbb 100644
@@ -16,6 +16,7 @@
 #define _LINUX_BITFIELD_H
 
 #include <linux/bug.h>
+#include <asm/byteorder.h>
 
 /*
  * Bitfield access macros
Signature · 11 lines
2.27.0



-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: This is a digitally signed message part
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20200709/19ad489b/attachment.sig>
#21 lore raw

PATCH 1/1 include/bitfield.h: include byteorder.h

Quoted 7 lines · Nicolas Saenz Julienne

This is needed in order to use the cpu_to_*()/*_to_cpu() family of functions.

---

IMO the ideal thing would be to squash this into the original patch.

I'll test and squash, thanks!

Quoted 17 lines
  include/linux/bitfield.h | 1 +
  1 file changed, 1 insertion(+)
include/linux/bitfield.h +1 -0
index 7ad8b088ed..7ad02f8cbb 100644
@@ -16,6 +16,7 @@
  #define _LINUX_BITFIELD_H
 
  #include <linux/bug.h>
+#include <asm/byteorder.h>
 
  /*
   * Bitfield access macros
#22 lore raw

PATCH v4 7/9 pci: Add some PCI Express capability register offset definitions

In reply to Sylwester Nawrocki

Add PCI Express capability definitions required by the Broadcom STB PCIe controller driver.

---
Changes since v3:
- none.
Changes since v2:
- added Current Link Speed defines.
Changes since v1:
- none.
Changes since RFC:
- ensure the entries are added in order, sorted by ascending address values.

---
 include/pci.h | 9 +++++++++
 1 file changed, 9 insertions(+)
include/pci.h +9 -0
index dfdbb32..ff5f620 100644
@@ -479,11 +479,20 @@
 #define PCI_EXP_DEVCTL		8	/* Device Control */
 #define  PCI_EXP_DEVCTL_BCR_FLR	0x8000  /* Bridge Configuration Retry / FLR */
 #define PCI_EXP_LNKCAP		12	/* Link Capabilities */
+#define  PCI_EXP_LNKCAP_SLS	0x0000000f /* Supported Link Speeds */
+#define  PCI_EXP_LNKCAP_MLW	0x000003f0 /* Maximum Link Width */
 #define  PCI_EXP_LNKCAP_DLLLARC	0x00100000 /* Data Link Layer Link Active Reporting Capable */
 #define PCI_EXP_LNKSTA		18	/* Link Status */
+#define  PCI_EXP_LNKSTA_CLS	0x000f	/* Current Link Speed */
+#define  PCI_EXP_LNKSTA_CLS_2_5GB 0x0001 /* Current Link Speed 2.5GT/s */
+#define  PCI_EXP_LNKSTA_CLS_5_0GB 0x0002 /* Current Link Speed 5.0GT/s */
+#define  PCI_EXP_LNKSTA_CLS_8_0GB 0x0003 /* Current Link Speed 8.0GT/s */
+#define  PCI_EXP_LNKSTA_NLW	0x03f0	/* Negotiated Link Width */
+#define  PCI_EXP_LNKSTA_NLW_SHIFT 4	/* start of NLW mask in link status */
 #define  PCI_EXP_LNKSTA_DLLLA	0x2000	/* Data Link Layer Link Active */
 #define PCI_EXP_SLTCAP		20	/* Slot Capabilities */
 #define  PCI_EXP_SLTCAP_PSN	0xfff80000 /* Physical Slot Number */
+#define PCI_EXP_LNKCTL2		48	/* Link Control 2 */
 
 /* Include the ID list */
 
Signature · 1 line
2.7.4
#23 lore raw

PATCH v4 8/9 pci: Add driver for Broadcom BCM2711 SoC PCIe controller

In reply to Sylwester Nawrocki

This patch adds basic driver PCI Express controller found on Broadcom set-top-box SoCs, e.g. BCM2711.
The code is based on Linux upstream driver (pcie-brcmstb.c) with MSI handling removed. The inbound access memory region is not currently parsed from dma-ranges DT property and a fixed 3GB region is used.

The patch has been tested on RPI4 board, i.e. on BCM2711 SoC with VL805 USB Host Controller.

---
Changes since v3:
- none.
Changes since v2 (mostly addressing Simon's review comments):
- removed MDO_RD_DONE, MDIO_WT_DONE macro definitions,
- updated the Kconfig entry help text,
- reordered #include entries to match the coding style,
- s/udev/dev,
- s/ENODEV/EINVAL in brcm_pcie_probe() and brcm_pcie_config_address() functions,
- Simplified brcm_pcie_mdio_{read, write} functions (readl_poll_timeout),
- shortened register bit fields macro definitions,
- dropped brcm_pcie_perst_set() and brcm_pcie_bridge_sw_init_set() function in favour of direct clrbits_le32/setbits_le32 calls,
- use setbits_le32/clrbits_le32/clrsetbits_le32 instead of readl(), u32p_replace_bits(), writel() sequence
- simplified brcm_pcie_config_address(), brcm_pcie_set_gen() functions,
- changed reset pulse delay to 100 us,
- Replaced FIELD_GET() with open coded bitwise operations,
- brcm_cpie_cfg_index() function merged into brcm_pcie_config_address(),
- use standard PCI PCI_EXP_LNKSTA_CLS_* link speed defines
- added kernel-doc function comments.

Changes since v1:
- fixed argument in brcm_pcie_set_ssc() function call,
- changed rc_bar2_size assignment to value 0xC0000000, as in upstream devicetree.
Changes since RFC:
- reworked to align with current Linux mainline version and u-boot driver by Nicolas Saenz Julienne.

---
 drivers/pci/Kconfig        |   9 +
 drivers/pci/Makefile       |   1 +
 drivers/pci/pcie_brcmstb.c | 623 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 633 insertions(+)
 create mode 100644 drivers/pci/pcie_brcmstb.c
drivers/pci/Kconfig +9 -0
index 437cd9a..543bd46 100644
@@ -197,4 +197,13 @@ config PCIE_MEDIATEK
 	  Say Y here if you want to enable Gen2 PCIe controller,
 	  which could be found on MT7623 SoC family.
 
+config PCI_BRCMSTB
+	bool "Broadcom STB PCIe controller"
+	depends on DM_PCI
+	depends on ARCH_BCM283X
+	help
+	  Say Y here if you want to enable support for PCIe controller
+	  on Broadcom set-top-box (STB) SoCs.
+	  This driver currently supports only BCM2711 SoC and RC mode
+	  of the controller.
 endif
drivers/pci/Makefile +1 -0
index c051ecc..3e53b1f 100644
@@ -43,3 +43,4 @@ obj-$(CONFIG_PCI_PHYTIUM) += pcie_phytium.o
 obj-$(CONFIG_PCIE_INTEL_FPGA) += pcie_intel_fpga.o
 obj-$(CONFIG_PCI_KEYSTONE) += pcie_dw_ti.o
 obj-$(CONFIG_PCIE_MEDIATEK) += pcie_mediatek.o
+obj-$(CONFIG_PCI_BRCMSTB) += pcie_brcmstb.o
drivers/pci/pcie_brcmstb.c new file +623 -0

623 lines not on this page · raw message

Signature · 1 line
2.7.4
#24 lore raw

PATCH v4 9/9 configs: Enable support for the XHCI controller on RPI4 board (ARM 64-bit)

In reply to Sylwester Nawrocki

From: Marek Szyprowski <[email protected]>

This requires enabling BRCMSTB PCIe and XHCI_PCI drivers as well as PCI and USB commands. To get it working one has to call the following commands:
"pci enum; usb start;", thus such commands have been added to the default "preboot" environment variable. One has to update their environment if it is already configured to get this feature working out of the box.

---
Changes since v3:
- added missing rpi_4_defconfig changes
Changes since v2:
- rpi_4_32b_defconfig, rpi_4_defconfig changes moved to separate patch
Changes since v1:
- removed unneeded CONFIG_XHCI_64BIT_DWORD_ACCESS_ONLY entry.
Changes since RFC:
- none.

---
 configs/rpi_4_defconfig     | 9 +++++++++
 configs/rpi_arm64_defconfig | 8 +++++++-
 2 files changed, 16 insertions(+), 1 deletion(-)
configs/rpi_4_defconfig +9 -0
index 6d148da..3f1a1b5 100644
@@ -8,6 +8,8 @@ CONFIG_NR_DRAM_BANKS=2
 CONFIG_DISTRO_DEFAULTS=y
 CONFIG_OF_BOARD_SETUP=y
 # CONFIG_ARCH_FIXUP_FDT_MEMORY is not set
+CONFIG_USE_PREBOOT=y
+CONFIG_PREBOOT="pci enum; usb start;"
 CONFIG_MISC_INIT_R=y
 # CONFIG_DISPLAY_CPUINFO is not set
 # CONFIG_DISPLAY_BOARDINFO is not set
@@ -15,6 +17,8 @@ CONFIG_SYS_PROMPT="U-Boot> "
 CONFIG_CMD_DFU=y
 CONFIG_CMD_GPIO=y
 CONFIG_CMD_MMC=y
+CONFIG_CMD_PCI=y
+CONFIG_CMD_USB=y
 CONFIG_CMD_FS_UUID=y
 CONFIG_OF_BOARD=y
 CONFIG_ENV_FAT_INTERFACE="mmc"
@@ -28,12 +32,17 @@ CONFIG_MMC_SDHCI=y
 CONFIG_MMC_SDHCI_BCM2835=y
 CONFIG_DM_ETH=y
 CONFIG_BCMGENET=y
+CONFIG_PCI=y
+CONFIG_DM_PCI=y
+CONFIG_PCI_BRCMSTB=y
 CONFIG_PINCTRL=y
 # CONFIG_PINCTRL_GENERIC is not set
 # CONFIG_REQUIRE_SERIAL_CONSOLE is not set
 CONFIG_USB=y
 CONFIG_DM_USB=y
 CONFIG_DM_USB_GADGET=y
+CONFIG_USB_XHCI_HCD=y
+CONFIG_USB_XHCI_PCI=y
 CONFIG_USB_GADGET=y
 CONFIG_USB_GADGET_MANUFACTURER="FSL"
 CONFIG_USB_GADGET_VENDOR_NUM=0x0525
configs/rpi_arm64_defconfig +7 -1
index fea86be..f12d1e3 100644
@@ -7,13 +7,14 @@ CONFIG_NR_DRAM_BANKS=2
 CONFIG_DISTRO_DEFAULTS=y
 CONFIG_OF_BOARD_SETUP=y
 CONFIG_USE_PREBOOT=y
-CONFIG_PREBOOT="usb start"
+CONFIG_PREBOOT="pci enum; usb start;"
 CONFIG_MISC_INIT_R=y
 # CONFIG_DISPLAY_CPUINFO is not set
 # CONFIG_DISPLAY_BOARDINFO is not set
 CONFIG_SYS_PROMPT="U-Boot> "
 CONFIG_CMD_GPIO=y
 CONFIG_CMD_MMC=y
+CONFIG_CMD_PCI=y
 CONFIG_CMD_USB=y
 CONFIG_CMD_FS_UUID=y
 CONFIG_OF_BOARD=y
@@ -26,11 +27,16 @@ CONFIG_MMC_SDHCI=y
 CONFIG_MMC_SDHCI_BCM2835=y
 CONFIG_DM_ETH=y
 CONFIG_BCMGENET=y
+CONFIG_PCI=y
+CONFIG_DM_PCI=y
+CONFIG_PCI_BRCMSTB=y
 CONFIG_PINCTRL=y
 # CONFIG_PINCTRL_GENERIC is not set
 # CONFIG_REQUIRE_SERIAL_CONSOLE is not set
 CONFIG_USB=y
 CONFIG_DM_USB=y
+CONFIG_USB_XHCI_HCD=y
+CONFIG_USB_XHCI_PCI=y
 CONFIG_USB_DWC2=y
 CONFIG_USB_KEYBOARD=y
 CONFIG_USB_HOST_ETHER=y
Signature · 1 line
2.7.4
#25 lore raw

v4 cover USB host support for Raspberry Pi 4 board (64-bit)

In reply to Sylwester Nawrocki

Hi Sylwester,

Quoted 31 lines · Sylwester Nawrocki

Hi all,

This patch series adds USB host support for Raspberry Pi 4 board. It includes the Broadcom STB PCIe controller driver ported from Linux kernel, a memory mapping update for the xHCI controller on PCIe bus for 64-bit builds and some related fixes and updates in the usb/xhci and the pci driver core code.

The 32-bit ARM part has be excluded from the series and will be posted separately.

The difference comparing to previous version is only in the config update patch "configs: Enable support for the XHCI controller on RPI4 board (ARM
64-bit)".

The patch series is based on v2020.07-rc1 tree.

Thanks,
Sylwester

Marek Szyprowski (3):
rpi4: shorten a mapping for the DRAM
rpi4: add a mapping for the PCIe XHCI controller MMIO registers (ARM

    64bit)
  configs: Enable support for the XHCI controller on RPI4 board (ARM
    64-bit)

Nicolas Saenz Julienne (1):
linux/bitfield.h: Add primitives for manipulating bitfields both in
host- and fixed-endian

Quoted 8 lines

Sylwester Nawrocki (5):
usb: xhci: Add missing cache flush in the scratchpad array
initialization
usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq
pci: Move some PCIe register offset definitions to a common header
pci: Add some PCI Express capability register offset definitions
pci: Add driver for Broadcom BCM2711 SoC PCIe controller

Series looks good to me. The problem is that on RPi4 with 8 GB the USB FW get's loaded via the RPi FW. Nicolas provided a series which is under review right now. Without this series the 8 GB RPi4 won't boot. As soon as the series is ready I'll take all the RPi4 PCI/USB related patches into the next branch.

Regards,
Matthias

Quoted 15 lines
 arch/arm/mach-bcm283x/init.c  |  20 +-
 configs/rpi_4_defconfig       |   9 +
 configs/rpi_arm64_defconfig   |   8 +-
 drivers/pci/Kconfig           |   9 +
 drivers/pci/Makefile          |   1 +
 drivers/pci/pci-rcar-gen3.c   |   8 -
 drivers/pci/pcie_brcmstb.c    | 623 ++++++++++++++++++++++++++++++++++++++++++
 drivers/pci/pcie_intel_fpga.c |   3 -
 drivers/usb/host/xhci-mem.c   |   3 +
 include/linux/bitfield.h      |  52 ++++
 include/pci.h                 |  22 +-
 include/usb/xhci.h            |   8 -
 12 files changed, 740 insertions(+), 26 deletions(-)
 create mode 100644 drivers/pci/pcie_brcmstb.c
#26 lore raw

v4 cover USB host support for Raspberry Pi 4 board (64-bit)

In reply to Sylwester Nawrocki
Quoted 31 lines · Sylwester Nawrocki

Hi all,

This patch series adds USB host support for Raspberry Pi 4 board. It includes the Broadcom STB PCIe controller driver ported from Linux kernel, a memory mapping update for the xHCI controller on PCIe bus for 64-bit builds and some related fixes and updates in the usb/xhci and the pci driver core code.

The 32-bit ARM part has be excluded from the series and will be posted separately.

The difference comparing to previous version is only in the config update patch "configs: Enable support for the XHCI controller on RPI4 board (ARM
64-bit)".

The patch series is based on v2020.07-rc1 tree.

Thanks,
Sylwester

Marek Szyprowski (3):
rpi4: shorten a mapping for the DRAM
rpi4: add a mapping for the PCIe XHCI controller MMIO registers (ARM

     64bit)
   configs: Enable support for the XHCI controller on RPI4 board (ARM
     64-bit)

Nicolas Saenz Julienne (1):
linux/bitfield.h: Add primitives for manipulating bitfields both in
host- and fixed-endian

Quoted 7 lines

Sylwester Nawrocki (5):
usb: xhci: Add missing cache flush in the scratchpad array
initialization
usb: xhci: Use only 32-bit accesses in xhci_writeq/xhci_readq
pci: Move some PCIe register offset definitions to a common header
pci: Add some PCI Express capability register offset definitions
pci: Add driver for Broadcom BCM2711 SoC PCIe controller

Applied now to rpi-next

Thanks!

Quoted 16 lines
  arch/arm/mach-bcm283x/init.c  |  20 +-
  configs/rpi_4_defconfig       |   9 +
  configs/rpi_arm64_defconfig   |   8 +-
  drivers/pci/Kconfig           |   9 +
  drivers/pci/Makefile          |   1 +
  drivers/pci/pci-rcar-gen3.c   |   8 -
  drivers/pci/pcie_brcmstb.c    | 623 ++++++++++++++++++++++++++++++++++++++++++
  drivers/pci/pcie_intel_fpga.c |   3 -
  drivers/usb/host/xhci-mem.c   |   3 +
  include/linux/bitfield.h      |  52 ++++
  include/pci.h                 |  22 +-
  include/usb/xhci.h            |   8 -
  12 files changed, 740 insertions(+), 26 deletions(-)

create mode 100644 drivers/pci/pcie_brcmstb.c