Xilinx ZynqMP USB fixes
Subsystems (3)
v6 cover Xilinx ZynqMP USB fixes
Some fixes related to the Xilinx ZynqMP DWC3 wrapper driver to allow ZynqMP USB to work properly when the hardware is configured in USB 2.0-only mode.
Changes since v5:
-code formatting fixes, no functional change
Changes since v4:
-dropped DWC3 core patches as they are superseded by Sean Anderson's patchset "usb: dwc3: Calculate REFCLKPER et. al. from reference clock", ZynqMP-specific patches unchanged
Changes since v3:
-fixed DT schema dt-doc-validate error
Changes since v2:
-additional kerneldoc fixes
Changes since v1:
-added DT binding documentation for new attribute
-kerneldoc formatting and reworded comments
Robert Hancock (2):
usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
usb: dwc3: xilinx: Fix error handling when getting USB3 PHY
drivers/usb/dwc3/dwc3-xilinx.c | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)
Signature · 1 line
2.31.1
PATCH v6 1/2 usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
It appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
drivers/usb/dwc3/dwc3-xilinx.c +6 -2
@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/ writel (FPD_POWER_PRSNT_OPTION , priv_data->regs +XLNX_USB_FPD_POWER_PRSNT ); -/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ -writel (PIPE_CLK_SELECT , priv_data->regs +XLNX_USB_FPD_PIPE_CLK ); +/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3 +* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK ); ret =reset_control_deassert (crst);if (ret <0 ) {
Signature · 1 line
2.31.1
RE: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 29 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert Hancock <[email protected]>
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 modeIt appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/ writel (FPD_POWER_PRSNT_OPTION , priv_data->regs +XLNX_USB_FPD_POWER_PRSNT );
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx · line 178
Quoted 10 lines
drivers/usb/dwc3/dwc3-xilinx.c +6
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ -writel (PIPE_CLK_SELECT , priv_data->regs + XLNX_USB_FPD_PIPE_CLK); +/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3 +* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );
When USB3.0 is enabled in the design, FSBL will set this bit to PIPE_CLK_SELECT And it's state will be persistent till Linux stage. When this driver finds the usb3-phy property In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design has USB3.0 enabled, then this will clear this bit and ultimately it will fail.
It will be better to skip touching that bit in case the device tree does not have the usb3-phy property. This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.
---
int ret;
u32 reg;- usb3_phy = devm_phy_get(dev, "usb3-phy");
- if (PTR_ERR(usb3_phy) == -EPROBE_DEFER) {
- ret = -EPROBE_DEFER;
+ usb3_phy = devm_phy_optional_get(dev, "usb3-phy");
+ if (IS_ERR(usb3_phy)) {
+ ret = PTR_ERR(usb3_phy);
+ dev_err_probe(dev, ret, "failed to get USB3 PHY\n");
goto err;
- } else if (IS_ERR(usb3_phy)) {
- usb3_phy = NULL;
}+ if (!usb3_phy)
+ goto skip_usb3_phy;
+
crst = devm_reset_control_get_exclusive(dev, "usb_crst");
if (IS_ERR(crst)) {
ret = PTR_ERR(crst);
@@ -188,6 +190,7 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data)
goto err;
}+skip_usb3_phy:
/*
* This routes the USB DMA traffic to go through FPD path instead
* of reaching DDR directly. This traffic routing is needed to
---
Thanks,
Manish
Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
Quoted 54 lines · Manish Narani
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 31 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert Hancock <[email protected]>
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0
modeIt appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)drivers/usb/dwc3/dwc3- renamed from drivers/usb/dwc3/dwc3-xilinx.c +0
-0
xilinx.c
index9cc3ad701a29..dd6218d05159100644drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/ writel (FPD_POWER_PRSNT_OPTION , priv_data->regs +XLNX_USB_FPD_POWER_PRSNT );drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx · line 178
Quoted 10 lines
drivers/usb/dwc3/dwc3-xilinx.c +6
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ -writel (PIPE_CLK_SELECT , priv_data->regs + XLNX_USB_FPD_PIPE_CLK); +/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3 +* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );When USB3.0 is enabled in the design, FSBL will set this bit to
PIPE_CLK_SELECT
And it's state will be persistent till Linux stage. When this driver finds the usb3-phy property
In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design has USB3.0 enabled, then this will clear this bit and ultimately it will fail.
Quoted 5 lines
It will be better to skip touching that bit in case the device tree does not have the usb3-phy property.
This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.
So the original patch was tested against hardware that only had USB 2.0 support and seemed to work fine. However, we've since found an issue with some other hardware supporting USB 3.0 where either it doesn't detect devices at all, or they get detected but then seem to drop off the bus very quickly, and we get this repeatedly:
[ 99.858607] usb usb2-port1: Cannot enable. Maybe the USB cable is bad?
The same problem is reproducible on the Xilinx ZCU102 board with the same kernel build, where the USB works fine with the Xilinx kernel and a Petalinux 2021.2 pre-built ZCU102 image, so it's no hardware issue.
I've been trying to isolate any relevant differences between the Xilinx kernel and mainline in this respect but haven't had much success. One difference in this particular dwc3-xilinx code is that the Xilinx kernel has code to reset the ULPI PHY which is not in the mainline version yet. However, adding that in doesn't seem to fix the problem.
Have you (or anyone else on the CC list) done any testing of USB 3.0 devices and USB 3.0 capable hardware on mainline with ZynqMP to know if this is a more general issue?
Quoted 35 lines
---
int ret; u32 reg;(quoted diff) +8
-5- usb3_phy = devm_phy_get(dev, "usb3-phy"); - if (PTR_ERR(usb3_phy) == -EPROBE_DEFER) { - ret = -EPROBE_DEFER; + usb3_phy = devm_phy_optional_get(dev, "usb3-phy"); + if (IS_ERR(usb3_phy)) { + ret = PTR_ERR(usb3_phy); + dev_err_probe(dev, ret, "failed to get USB3 PHY\n"); goto err; - } else if (IS_ERR(usb3_phy)) { - usb3_phy = NULL; } + if (!usb3_phy) + goto skip_usb3_phy; + crst = devm_reset_control_get_exclusive(dev, "usb_crst"); if (IS_ERR(crst)) { ret = PTR_ERR(crst);@@ -188,6 +190,7 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) goto err; } +skip_usb3_phy: /* * This routes the USB DMA traffic to go through FPD path instead* of reaching DDR directly. This traffic routing is needed to
---Thanks,
Manish
RE: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
Hi Robert,
Quoted 93 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Tuesday, January 25, 2022 12:31 AM
To: Manish Narani <[email protected]>; [email protected]
Cc: Michal Simek <[email protected]>; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 modeQuoted 11 lines · Manish Narani
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 7 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert HancockQuoted 14 lines
Quoted 14 lines
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0
modeIt appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)Quoted 6 lines
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct
Quoted 6 lines
drivers/usb/dwc3/dwc3- renamed from drivers/usb/dwc3/dwc3-xilinx.c +0
-0
xilinx.c
index9cc3ad701a29..dd6218d05159100644drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct
dwc3_xlnxQuoted 23 lines
Quoted 4 lines
drivers/usb/dwc3/dwc3-xilinx.c +0
-0*priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/ writel (FPD_POWER_PRSNT_OPTION , priv_data->regs +XLNX_USB_FPD_POWER_PRSNT );drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) · line 175
Quoted 10 lines
drivers/usb/dwc3/dwc3-xilinx.c +6
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ -writel (PIPE_CLK_SELECT , priv_data->regs + XLNX_USB_FPD_PIPE_CLK); +/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3 +* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );When USB3.0 is enabled in the design, FSBL will set this bit to
PIPE_CLK_SELECT
And it's state will be persistent till Linux stage. When this driver finds the usb3-phy property
In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design has USB3.0 enabled, then this will clear this bit and ultimately it will fail.Quoted 5 lines
It will be better to skip touching that bit in case the device tree does not have the usb3-phy property.
This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.So the original patch was tested against hardware that only had USB 2.0 support
and seemed to work fine. However, we've since found an issue with some other
hardware supporting USB 3.0 where either it doesn't detect devices at all, or they get detected but then seem to drop off the bus very quickly, and we get this repeatedly:[ 99.858607] usb usb2-port1: Cannot enable. Maybe the USB cable is bad?The same problem is reproducible on the Xilinx ZCU102 board with the same kernel build, where the USB works fine with the Xilinx kernel and a Petalinux 2021.2 pre-built ZCU102 image, so it's no hardware issue.
I've been trying to isolate any relevant differences between the Xilinx kernel and mainline in this respect but haven't had much success. One difference in this particular dwc3-xilinx code is that the Xilinx kernel has code to reset the ULPI PHY which is not in the mainline version yet. However, adding that in
doesn't seem to fix the problem.
Quoted 4 lines
Have you (or anyone else on the CC list) done any testing of USB 3.0 devices and USB 3.0 capable hardware on mainline with ZynqMP to know if this is a more
general issue?
Yes, The USB fixes are work in progress and will be sent to mainline in near future. This fix is one of them. Your patch is solving the problem in case of USB 2.0 But not the USB 3.0 entirely. There is this corner case which I mentioned, breaks USB 3.0 functionality with your patch.
Thanks,
Manish
Quoted 36 lines
Quoted 36 lines
---
int ret; u32 reg;(quoted diff) +8
-5- usb3_phy = devm_phy_get(dev, "usb3-phy"); - if (PTR_ERR(usb3_phy) == -EPROBE_DEFER) { - ret = -EPROBE_DEFER; + usb3_phy = devm_phy_optional_get(dev, "usb3-phy"); + if (IS_ERR(usb3_phy)) { + ret = PTR_ERR(usb3_phy); + dev_err_probe(dev, ret, "failed to get USB3 PHY\n"); goto err; - } else if (IS_ERR(usb3_phy)) { - usb3_phy = NULL; } + if (!usb3_phy) + goto skip_usb3_phy; + crst = devm_reset_control_get_exclusive(dev, "usb_crst"); if (IS_ERR(crst)) { ret = PTR_ERR(crst);@@ -188,6 +190,7 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) goto err; } +skip_usb3_phy: /* * This routes the USB DMA traffic to go through FPD path instead* of reaching DDR directly. This traffic routing is needed to
---Thanks,
Manish
Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
Quoted 113 lines · Manish Narani
Hi Robert,
Quoted 105 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Tuesday, January 25, 2022 12:31 AM
To: Manish Narani <[email protected]>; [email protected]
Cc: Michal Simek <[email protected]>; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 modeQuoted 11 lines · Manish Narani
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 7 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert HancockQuoted 14 lines
Quoted 14 lines
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0
modeIt appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)Quoted 6 lines
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct
Quoted 6 lines
drivers/usb/dwc3/dwc3- renamed from drivers/usb/dwc3/dwc3-xilinx.c +0
-0
xilinx.c
index9cc3ad701a29..dd6218d05159100644drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct
dwc3_xlnxQuoted 28 lines
Quoted 12 lines
drivers/usb/dwc3/dwc3-xilinx.c +0
-0*priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/
writel(FPD_POWER_PRSNT_OPTION,priv_data->regs+XLNX_USB_FPD_POWER_PRSNT);drivers/usb/dwc3/dwc3-xilinx.c +0
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ - writel(PIPE_CLK_SELECT, priv_data->regs +
XLNX_USB_FPD_PIPE_CLK);drivers/usb/dwc3/dwc3-xilinx.c +2
-0+/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) · line 175
Quoted 4 lines
drivers/usb/dwc3/dwc3-xilinx.c +4
-0+* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );When USB3.0 is enabled in the design, FSBL will set this bit to
PIPE_CLK_SELECT
And it's state will be persistent till Linux stage. When this driver finds
the usb3-phy property
In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design has
USB3.0 enabled, then this will clear this bit and ultimately it will fail.Quoted 6 lines
It will be better to skip touching that bit in case the device tree does not
have the usb3-phy property.
This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.So the original patch was tested against hardware that only had USB 2.0 support
and seemed to work fine. However, we've since found an issue with some other
hardware supporting USB 3.0 where either it doesn't detect devices at all, or
they get detected but then seem to drop off the bus very quickly, and we get
this repeatedly:[ 99.858607] usb usb2-port1: Cannot enable. Maybe the USB cable is bad?The same problem is reproducible on the Xilinx ZCU102 board with the same kernel build, where the USB works fine with the Xilinx kernel and a Petalinux
2021.2 pre-built ZCU102 image, so it's no hardware issue.I've been trying to isolate any relevant differences between the Xilinx kernel
and mainline in this respect but haven't had much success. One difference in
this particular dwc3-xilinx code is that the Xilinx kernel has code to reset
the ULPI PHY which is not in the mainline version yet. However, adding that in
doesn't seem to fix the problem.Quoted 5 lines
Have you (or anyone else on the CC list) done any testing of USB 3.0 devices
and USB 3.0 capable hardware on mainline with ZynqMP to know if this is a more
general issue?
Quoted 7 lines
Yes, The USB fixes are work in progress and will be sent to mainline in near future.
This fix is one of them. Your patch is solving the problem in case of USB 2.0
But not the USB 3.0 entirely. There is this corner case which I mentioned, breaks
USB 3.0 functionality with your patch.
You mentioned the case where the design is using USB 3.0 but the usb3-phy is missing from the device tree - with this patch it would switch the clock source into the USB 2 configuration and might break things. I agree it's probably better to just skip the whole reset process and keep the FSBL clock selection if it's not needed in the USB 2 case.
We still have another case where USB 3.0 appears to be broken on a setup where the usb3-phy is specified, and where the Xilinx/Petalinux images work on the same hardware. I haven't yet isolated whether it is related to the kernel or might be caused by the different PL logic or PS configuration settings somehow.
Quoted 39 lines
Thanks,
ManishQuoted 35 lines
Quoted 35 lines
---
int ret; u32 reg;(quoted diff) +8
-5- usb3_phy = devm_phy_get(dev, "usb3-phy"); - if (PTR_ERR(usb3_phy) == -EPROBE_DEFER) { - ret = -EPROBE_DEFER; + usb3_phy = devm_phy_optional_get(dev, "usb3-phy"); + if (IS_ERR(usb3_phy)) { + ret = PTR_ERR(usb3_phy); + dev_err_probe(dev, ret, "failed to get USB3 PHY\n"); goto err; - } else if (IS_ERR(usb3_phy)) { - usb3_phy = NULL; } + if (!usb3_phy) + goto skip_usb3_phy; + crst = devm_reset_control_get_exclusive(dev, "usb_crst"); if (IS_ERR(crst)) { ret = PTR_ERR(crst);@@ -188,6 +190,7 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) goto err; } +skip_usb3_phy: /* * This routes the USB DMA traffic to go through FPD path instead* of reaching DDR directly. This traffic routing is needed to
---Thanks,
Manish
Signature · 3 lines
Robert Hancock Senior Hardware Designer, Calian Advanced Technologies www.calian.com
Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
Quoted 122 lines · Robert Hancock
Quoted 113 lines · Manish Narani
Hi Robert,
Quoted 105 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Tuesday, January 25, 2022 12:31 AM
To: Manish Narani <[email protected]>; [email protected]
Cc: Michal Simek <[email protected]>; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 modeQuoted 11 lines · Manish Narani
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 7 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert HancockQuoted 14 lines
Quoted 14 lines
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0
modeIt appears that the PIPE clock should not be selected when only USB 2.0 is being used in the design and no USB 3.0 reference clock is used. Fix to set the correct value depending on whether a USB3 PHY is present.
Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)Quoted 6 lines
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct
Quoted 6 lines
drivers/usb/dwc3/dwc3- renamed from drivers/usb/dwc3/dwc3-xilinx.c +0
-0
xilinx.c
index9cc3ad701a29..dd6218d05159100644drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct
dwc3_xlnxQuoted 28 lines
Quoted 12 lines
drivers/usb/dwc3/dwc3-xilinx.c +0
-0*priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/
writel(FPD_POWER_PRSNT_OPTION,priv_data->regs+XLNX_USB_FPD_POWER_PRSNT);drivers/usb/dwc3/dwc3-xilinx.c +0
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ - writel(PIPE_CLK_SELECT, priv_data->regs +
XLNX_USB_FPD_PIPE_CLK);drivers/usb/dwc3/dwc3-xilinx.c +2
-0+/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) · line 175
Quoted 4 lines
drivers/usb/dwc3/dwc3-xilinx.c +4
-0+* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );When USB3.0 is enabled in the design, FSBL will set this bit to
PIPE_CLK_SELECT
And it's state will be persistent till Linux stage. When this driver finds
the usb3-phy property
In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design has
USB3.0 enabled, then this will clear this bit and ultimately it will fail.Quoted 6 lines
It will be better to skip touching that bit in case the device tree does not
have the usb3-phy property.
This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.So the original patch was tested against hardware that only had USB 2.0 support
and seemed to work fine. However, we've since found an issue with some other
hardware supporting USB 3.0 where either it doesn't detect devices at all, or
they get detected but then seem to drop off the bus very quickly, and we get
this repeatedly:[ 99.858607] usb usb2-port1: Cannot enable. Maybe the USB cable is bad?The same problem is reproducible on the Xilinx ZCU102 board with the same kernel build, where the USB works fine with the Xilinx kernel and a Petalinux
2021.2 pre-built ZCU102 image, so it's no hardware issue.I've been trying to isolate any relevant differences between the Xilinx kernel
and mainline in this respect but haven't had much success. One difference in
this particular dwc3-xilinx code is that the Xilinx kernel has code to reset
the ULPI PHY which is not in the mainline version yet. However, adding that in
doesn't seem to fix the problem.Quoted 5 lines
Have you (or anyone else on the CC list) done any testing of USB 3.0 devices
and USB 3.0 capable hardware on mainline with ZynqMP to know if this is a more
general issue?Quoted 7 lines
Yes, The USB fixes are work in progress and will be sent to mainline in near future.
This fix is one of them. Your patch is solving the problem in case of USB 2.0
But not the USB 3.0 entirely. There is this corner case which I mentioned, breaks
USB 3.0 functionality with your patch.
Quoted 2 lines
You mentioned the case where the design is using USB 3.0 but the usb3-phy is missing from the device tree
Do we even need to support that?
Either phy (with an appropriate phy-names) or usb3-phy should be present for USB3. Mainline has had GTR support since 5.8(?) or so.
--Sean
Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
In reply to Robert HancockQuoted 138 lines · Robert Hancock
Quoted 121 lines · Manish Narani
Hi Robert,
Quoted 113 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Tuesday, January 25, 2022 12:31 AM
To: Manish Narani <[email protected]>; [email protected]
Cc: Michal Simek <[email protected]>; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for
USB2.0 modeQuoted 11 lines · Manish Narani
Hi Robert,
Thanks for the patch! Please see my comments below inline!
Quoted 7 lines
-----Original Message-----
From: Robert Hancock <[email protected]>
Sent: Friday, January 21, 2022 11:49 PM
To: [email protected]
Cc: [email protected]; [email protected]; Michal Simek <[email protected]>; Manish Narani <[email protected]>; [email protected]; Robert HancockQuoted 17 lines
Quoted 17 lines
Subject: [PATCH v6 1/2] usb: dwc3: xilinx: Fix PIPE clock selection for
USB2.0
modeIt appears that the PIPE clock should not be selected when only USB 2.0
is being used in the design and no USB 3.0 reference clock is used. Fix
to set the correct value depending on whether a USB3 PHY is present.Fixes 84770f028fab ("usb: dwc3: Add driver for Xilinx platforms")Signed-off-by Robert Hancock <[email protected]>--- drivers/usb/dwc3/dwc3-xilinx.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)Quoted 6 lines
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct
Quoted 6 lines
drivers/usb/dwc3/dwc3- renamed from drivers/usb/dwc3/dwc3-xilinx.c +0
-0
xilinx.c
index9cc3ad701a29..dd6218d05159100644drivers/usb/dwc3/dwc3-xilinx.c +0
-0@@ -167,8 +167,12 @@ static int dwc3_xlnx_init_zynqmp(struct
dwc3_xlnxQuoted 29 lines
Quoted 12 lines
drivers/usb/dwc3/dwc3-xilinx.c +0
-0*priv_data)/* Set PIPE Power Present signal in FPD Power Present Register*/
writel(FPD_POWER_PRSNT_OPTION,priv_data->regs+XLNX_USB_FPD_POWER_PRSNT);drivers/usb/dwc3/dwc3-xilinx.c +0
-2-/* Set the PIPE Clock Select bit in FPD PIPE Clock register */ - writel(PIPE_CLK_SELECT, priv_data->regs +
XLNX_USB_FPD_PIPE_CLK);drivers/usb/dwc3/dwc3-xilinx.c +2
-0+/* +* Set the PIPE Clock Select bit in FPD PIPE Clock register if a USB3
drivers/usb/dwc3/dwc3-xilinx.c static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) · line 175
Quoted 4 lines
drivers/usb/dwc3/dwc3-xilinx.c +4
-0+* PHY is in use, deselect otherwise +*/ +writel (usb3_phy ?PIPE_CLK_SELECT :PIPE_CLK_DESELECT , + priv_data->regs +XLNX_USB_FPD_PIPE_CLK );When USB3.0 is enabled in the design, FSBL will set this bit to
PIPE_CLK_SELECT
And it's state will be persistent till Linux stage. When this driver finds
the usb3-phy property
In the device tree, it will again set this bit.
But in case if the usb3-phy is not present in the device tree and design
has
USB3.0 enabled, then this will clear this bit and ultimately it will fail.Quoted 7 lines
It will be better to skip touching that bit in case the device tree does
not
have the usb3-phy property.
This will skip the whole sequence of PHY initialization (reset assert/deassert are done in order to help initialize PHY). Something like below should work.So the original patch was tested against hardware that only had USB 2.0 support
and seemed to work fine. However, we've since found an issue with some other
hardware supporting USB 3.0 where either it doesn't detect devices at all,
or
they get detected but then seem to drop off the bus very quickly, and we get
this repeatedly:[ 99.858607] usb usb2-port1: Cannot enable. Maybe the USB cable is bad?The same problem is reproducible on the Xilinx ZCU102 board with the same kernel build, where the USB works fine with the Xilinx kernel and a Petalinux
2021.2 pre-built ZCU102 image, so it's no hardware issue.I've been trying to isolate any relevant differences between the Xilinx kernel
and mainline in this respect but haven't had much success. One difference in
this particular dwc3-xilinx code is that the Xilinx kernel has code to reset
the ULPI PHY which is not in the mainline version yet. However, adding that
in
doesn't seem to fix the problem.Quoted 5 lines
Have you (or anyone else on the CC list) done any testing of USB 3.0 devices
and USB 3.0 capable hardware on mainline with ZynqMP to know if this is a more
general issue?Quoted 8 lines
Yes, The USB fixes are work in progress and will be sent to mainline in near
future.
This fix is one of them. Your patch is solving the problem in case of USB 2.0
But not the USB 3.0 entirely. There is this corner case which I mentioned, breaks
USB 3.0 functionality with your patch.You mentioned the case where the design is using USB 3.0 but the usb3-phy is missing from the device tree - with this patch it would switch the clock source
into the USB 2 configuration and might break things. I agree it's probably better to just skip the whole reset process and keep the FSBL clock selection if it's not needed in the USB 2 case.
Quoted 6 lines
We still have another case where USB 3.0 appears to be broken on a setup where
the usb3-phy is specified, and where the Xilinx/Petalinux images work on the same hardware. I haven't yet isolated whether it is related to the kernel or might be caused by the different PL logic or PS configuration settings somehow.
Just to follow up, it appears this problem was due to a bug in the phy-zynqmp driver corrupting the PS-GTR settings for other lanes when SGMII mode was enabled on any lane. I'll be submitting a separate patch for this shortly. I'll also update this patch set to skip the resets and USB3-only register settings when no USB3 PHY is specified.
Quoted 40 lines
Quoted 40 lines
Thanks,
ManishQuoted 36 lines
Quoted 36 lines
---
int ret; u32 reg;(quoted diff) +8
-5- usb3_phy = devm_phy_get(dev, "usb3-phy"); - if (PTR_ERR(usb3_phy) == -EPROBE_DEFER) { - ret = -EPROBE_DEFER; + usb3_phy = devm_phy_optional_get(dev, "usb3-phy"); + if (IS_ERR(usb3_phy)) { + ret = PTR_ERR(usb3_phy); + dev_err_probe(dev, ret, "failed to get USB3 PHY\n"); goto err; - } else if (IS_ERR(usb3_phy)) { - usb3_phy = NULL; } + if (!usb3_phy) + goto skip_usb3_phy; + crst = devm_reset_control_get_exclusive(dev, "usb_crst"); if (IS_ERR(crst)) { ret = PTR_ERR(crst);@@ -188,6 +190,7 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data) goto err; } +skip_usb3_phy: /* * This routes the USB DMA traffic to go through FPD pathinstead
* of reaching DDR directly. This traffic routing is needed to
---Thanks,
Manish
Signature · 3 lines
Robert Hancock Senior Hardware Designer, Calian Advanced Technologies www.calian.com
PATCH v6 2/2 usb: dwc3: xilinx: Fix error handling when getting USB3 PHY
In reply to Robert HancockThe code that looked up the USB3 PHY was ignoring all errors other than EPROBE_DEFER in an attempt to handle the PHY not being present. Fix and simplify the code by using devm_phy_optional_get and dev_err_probe so that a missing PHY is not treated as an error and unexpected errors are handled properly.
--- drivers/usb/dwc3/dwc3-xilinx.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
drivers/usb/dwc3/dwc3-xilinx.c +4 -5
@@ -102,12 +102,11 @@ static int dwc3_xlnx_init_zynqmp(struct dwc3_xlnx *priv_data)int ret;u32 reg; - usb3_phy =devm_phy_get (dev,"usb3-phy" ); -if (PTR_ERR (usb3_phy) == -EPROBE_DEFER ) { - ret = -EPROBE_DEFER ; + usb3_phy =devm_phy_optional_get (dev,"usb3-phy" ); +if (IS_ERR (usb3_phy)) { + ret =PTR_ERR (usb3_phy); +dev_err_probe (dev, ret,"failed to get USB3 PHY\n" ); goto err; - }else if (IS_ERR (usb3_phy)) { - usb3_phy =NULL ; } crst =devm_reset_control_get_exclusive (dev,"usb_crst" );
Signature · 1 line
2.31.1
Hi Robert,
Quoted 23 lines · Robert Hancock
Some fixes related to the Xilinx ZynqMP DWC3 wrapper driver to allow ZynqMP USB to work properly when the hardware is configured in USB 2.0-only mode.
Changes since v5:
-code formatting fixes, no functional changeChanges since v4:
-dropped DWC3 core patches as they are superseded by Sean Anderson's patchset "usb: dwc3: Calculate REFCLKPER et. al. from reference clock", ZynqMP-specific patches unchangedChanges since v3:
-fixed DT schema dt-doc-validate errorChanges since v2:
-additional kerneldoc fixesChanges since v1:
(quoted diff) +0
-2-added DT binding documentation for new attribute -kerneldoc formatting and reworded comments
Quoted 7 lines
Robert Hancock (2):
usb: dwc3: xilinx: Fix PIPE clock selection for USB2.0 mode
usb: dwc3: xilinx: Fix error handling when getting USB3 PHYdrivers/usb/dwc3/dwc3-xilinx.c | 17 ++++++++++------- 1 file changed, 10 insertions(+), 7 deletions(-)
Looks like this got my usb working, thanks!
For ZynqMP,