Skip to contentSkip to conversation

usb: chipidea: udc: enforce write to the memory.

PATCH In review Tomer Maimon 3 participants 6 messages Jul 10, 2024 – Jul 26, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 usb: chipidea: udc: enforce write to the memory.

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

---
 drivers/usb/chipidea/udc.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
drivers/usb/chipidea/udc.c +12 -3
index 0b7bd3c643c3..0b14a1d54d59 100644
@@ -190,12 +190,21 @@ static int hw_ep_get_halt(struct ci_hdrc *ci, int num, int dir)
  *
  * This function returns an error code
  */
-static int hw_ep_prime(struct ci_hdrc *ci, int num, int dir, int is_ctrl)
+static int hw_ep_prime(struct ci_hdrc *ci, struct ci_hw_ep *hwep, int num,
+		       int dir, int is_ctrl)
 {
 	int n = hw_ep_bit(num, dir);
 
 	/* Synchronize before ep prime */
 	wmb();
+	
+	/* 
+	 * We add the read from qh.ptr->td.token to make sure the previous
+	 * write to it indeed got into the mamory so when we prime the DMA
+	 * will read the updated data
+	 */ 
+	if (hwep->qh.ptr->td.token & 0x80000000)
+		pr_info("%s(): hwep->qh.ptr->td.token=%08x\n", __func__, hwep->qh.ptr->td.token);
 
 	if (is_ctrl && dir == RX && hw_read(ci, OP_ENDPTSETUPSTAT, BIT(num)))
 		return -EAGAIN;
@@ -632,7 +641,7 @@ static int _hardware_enqueue(struct ci_hw_ep *hwep, struct ci_hw_req *hwreq)
 		hwep->qh.ptr->cap |= cpu_to_le32(mul << __ffs(QH_MULT));
 	}
 
-	ret = hw_ep_prime(ci, hwep->num, hwep->dir,
+	ret = hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
 			   hwep->type == USB_ENDPOINT_XFER_CONTROL);
 done:
 	return ret;
@@ -658,7 +667,7 @@ static int reprime_dtd(struct ci_hdrc *ci, struct ci_hw_ep *hwep,
 	hwep->qh.ptr->td.token &=
 		cpu_to_le32(~(TD_STATUS_HALTED | TD_STATUS_ACTIVE));
 
-	return hw_ep_prime(ci, hwep->num, hwep->dir,
+	return hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
 				hwep->type == USB_ENDPOINT_XFER_CONTROL);
 }
 
Signature · 1 line
2.34.1
#1 lore raw

Dear Tomer,

Thank you for your patch.

Am 10.07.24 um 14:41 schrieb Tomer Maimon:

Quoted 3 lines

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

Please document the datasheet name, revision and section. If this actually caused user visible problems, please also document the test case.

Quoted 15 lines
---
  drivers/usb/chipidea/udc.c | 15 ++++++++++++---
  1 file changed, 12 insertions(+), 3 deletions(-)
drivers/usb/chipidea/udc.c +1 -1
index 0b7bd3c643c3..0b14a1d54d59 100644
@@ -190,12 +190,21 @@ static int hw_ep_get_halt(struct ci_hdrc *ci, int num, int dir)
   *
   * This function returns an error code
   */
-static int hw_ep_prime(struct ci_hdrc *ci, int num, int dir, int is_ctrl)
+static int hw_ep_prime(struct ci_hdrc *ci, struct ci_hw_ep *hwep, int num,

drivers/usb/chipidea/udc.c static int hw_ep_get_halt(struct ci_hdrc *ci, int num, int dir) · line 203

Quoted 10 lines
drivers/usb/chipidea/udc.c +5 -0
+		       int dir, int is_ctrl)
  {
  	int n = hw_ep_bit(num, dir);
 
  	/* Synchronize before ep prime */
  	wmb();
+	
+	/*
+	 * We add the read from qh.ptr->td.token to make sure the previous
+	 * write to it indeed got into the mamory so when we prime the DMA

m*e*mory

Quoted 17 lines
drivers/usb/chipidea/udc.c +5 -1
+	 * will read the updated data
+	 */
+	if (hwep->qh.ptr->td.token & 0x80000000)
+		pr_info("%s(): hwep->qh.ptr->td.token=%08x\n", __func__, hwep->qh.ptr->td.token);
 
  	if (is_ctrl && dir == RX && hw_read(ci, OP_ENDPTSETUPSTAT, BIT(num)))
  		return -EAGAIN;
@@ -632,7 +641,7 @@ static int _hardware_enqueue(struct ci_hw_ep *hwep, struct ci_hw_req *hwreq)
  		hwep->qh.ptr->cap |= cpu_to_le32(mul << __ffs(QH_MULT));
  	}
 
-	ret = hw_ep_prime(ci, hwep->num, hwep->dir,
+	ret = hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
  			   hwep->type == USB_ENDPOINT_XFER_CONTROL);
  done:
  	return ret;

drivers/usb/chipidea/udc.c static int reprime_dtd(struct ci_hdrc *ci, struct ci_hw_ep *hwep, · line 673

Quoted 8 lines
drivers/usb/chipidea/udc.c +1 -1
@@ -658,7 +667,7 @@ static int reprime_dtd(struct ci_hdrc *ci, struct ci_hw_ep *hwep,
  	hwep->qh.ptr->td.token &=
  		cpu_to_le32(~(TD_STATUS_HALTED | TD_STATUS_ACTIVE));
 
-	return hw_ep_prime(ci, hwep->num, hwep->dir,
+	return hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
  				hwep->type == USB_ENDPOINT_XFER_CONTROL);
  }
 

Kind regards,

Paul

#2 lore raw

Hi Paul,

Thanks a lot for your comments, it will be addressed in the next version.

Tomer

Quoted 72 lines · Paul Menzel

Dear Tomer,

Thank you for your patch.

Am 10.07.24 um 14:41 schrieb Tomer Maimon:

Quoted 3 lines

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

Please document the datasheet name, revision and section. If this actually caused user visible problems, please also document the test case.

Quoted 15 lines
---
  drivers/usb/chipidea/udc.c | 15 ++++++++++++---
  1 file changed, 12 insertions(+), 3 deletions(-)
drivers/usb/chipidea/udc.c +1 -1
index 0b7bd3c643c3..0b14a1d54d59 100644
@@ -190,12 +190,21 @@ static int hw_ep_get_halt(struct ci_hdrc *ci, int num, int dir)
   *
   * This function returns an error code
   */
-static int hw_ep_prime(struct ci_hdrc *ci, int num, int dir, int is_ctrl)
+static int hw_ep_prime(struct ci_hdrc *ci, struct ci_hw_ep *hwep, int num,

drivers/usb/chipidea/udc.c static int hw_ep_get_halt(struct ci_hdrc *ci, int num, int dir) · line 203

Quoted 10 lines
drivers/usb/chipidea/udc.c +5 -0
+                    int dir, int is_ctrl)
  {
      int n = hw_ep_bit(num, dir);
 
      /* Synchronize before ep prime */
      wmb();
+
+     /*
+      * We add the read from qh.ptr->td.token to make sure the previous
+      * write to it indeed got into the mamory so when we prime the DMA

m*e*mory

Quoted 17 lines
drivers/usb/chipidea/udc.c +5 -1
+      * will read the updated data
+      */
+     if (hwep->qh.ptr->td.token & 0x80000000)
+             pr_info("%s(): hwep->qh.ptr->td.token=%08x\n", __func__, hwep->qh.ptr->td.token);
 
      if (is_ctrl && dir == RX && hw_read(ci, OP_ENDPTSETUPSTAT, BIT(num)))
              return -EAGAIN;
@@ -632,7 +641,7 @@ static int _hardware_enqueue(struct ci_hw_ep *hwep, struct ci_hw_req *hwreq)
              hwep->qh.ptr->cap |= cpu_to_le32(mul << __ffs(QH_MULT));
      }
 
-     ret = hw_ep_prime(ci, hwep->num, hwep->dir,
+     ret = hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
                         hwep->type == USB_ENDPOINT_XFER_CONTROL);
  done:
      return ret;

drivers/usb/chipidea/udc.c static int reprime_dtd(struct ci_hdrc *ci, struct ci_hw_ep *hwep, · line 673

Quoted 8 lines
drivers/usb/chipidea/udc.c +1 -1
@@ -658,7 +667,7 @@ static int reprime_dtd(struct ci_hdrc *ci, struct ci_hw_ep *hwep,
      hwep->qh.ptr->td.token &=
              cpu_to_le32(~(TD_STATUS_HALTED | TD_STATUS_ACTIVE));
 
-     return hw_ep_prime(ci, hwep->num, hwep->dir,
+     return hw_ep_prime(ci, hwep, hwep->num, hwep->dir,
                              hwep->type == USB_ENDPOINT_XFER_CONTROL);
  }
 

Kind regards,

Paul

#3 lore raw
In reply to Tomer Maimon
Quoted 8 lines · Tomer Maimon

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

---
 drivers/usb/chipidea/udc.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

What's the state of this patch with respect to upstream? Is this something specific to the openbmc/linux dev-6.6 tree, or is there an upstream equivalent? If the latter, can you please link to the relevant patch?

Andrew

#4 lore raw

Hi Andrew,

Quoted 11 lines · Andrew Jeffery
Quoted 8 lines · Tomer Maimon

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

---
 drivers/usb/chipidea/udc.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
Quoted 4 lines

What's the state of this patch with respect to upstream? Is this something specific to the openbmc/linux dev-6.6 tree, or is there an upstream equivalent? If the latter, can you please link to the relevant patch?

This patch didn't upstream yet to OpenBMC vanilla.
Its not specific to openbmc/linux dev-6.6 tree, but it is something that related to NPCM UDC module and I am not sure that Chipidea will approve to upstream it to the main chipidea driver
I am planning to start the upstream in the next Linux version 6.11.

Quoted 2 lines

Andrew

Thanks,

Tomer

#5 lore raw
Quoted 15 lines · Tomer Maimon

Hi Andrew,

Quoted 11 lines · Andrew Jeffery
Quoted 8 lines · Tomer Maimon

In the prime endpoint function, we need to read from qh.ptr->td.token to ensure that the previous write to it has indeed been committed to memory.

---
 drivers/usb/chipidea/udc.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
Quoted 8 lines
Quoted 4 lines

What's the state of this patch with respect to upstream? Is this something specific to the openbmc/linux dev-6.6 tree, or is there an upstream equivalent? If the latter, can you please link to the relevant patch?

This patch didn't upstream yet to OpenBMC vanilla.
Its not specific to openbmc/linux dev-6.6 tree, but it is something that related to NPCM UDC module and I am not sure that Chipidea will approve to upstream it to the main chipidea driver

Given it's a driver for shared IP I prefer you get some feedback on the upstream lists before we apply this to the openbmc tree.

Andrew