Skip to contentSkip to conversation

usb: core: setup dma_pfn_offset for USB devices

PATCH Superseded Roger Quadros 3 participants 10 messages Aug 17, 2016 – Sep 13, 2016 USB

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 usb: core: setup dma_pfn_offset for USB devices

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---
 drivers/usb/core/message.c | 1 +
 drivers/usb/core/usb.c     | 1 +
 2 files changed, 2 insertions(+)
drivers/usb/core/message.c +1 -0
index 0406a59..6856b7a 100644
@@ -1863,6 +1863,7 @@ free_interfaces:
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +1 -0
index 5e80697..d81791a 100644
@@ -441,6 +441,7 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;
Signature · 1 line
2.7.4
#1 lore raw

PATCH v2 1/1 usb: core: setup dma_pfn_offset for USB devices and, interfaces

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---
 drivers/usb/core/message.c | 6 ++++++
 drivers/usb/core/usb.c     | 7 +++++++
 2 files changed, 13 insertions(+)
drivers/usb/core/message.c +6 -0
index 0406a59..66364ea 100644
@@ -1863,6 +1863,12 @@ free_interfaces:
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		/* Propagate dma_pfn_offset to USB interface.
+		 * This is especially required by mass storage interface
+		 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+		 * to set the bounce buffer limit based on dma_pfn_offset.
+		 */
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +7 -0
index 5e80697..7dde93d 100644
@@ -441,6 +441,13 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	/* Propagate bus controller's dma_pfn_offset to USB device,
+	 * as it would be needed to be propagated to USB interfaces.
+	 * This is especially required by mass storage interface
+	 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+	 * to set the bounce buffer limit based on dma_pfn_offset.
+	 */
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;
Signature · 1 line
2.7.4
#2 lore raw

Re: [PATCH v2] usb: core: setup dma_pfn_offset for USB devices and, interfaces

Quoted 23 lines · Roger Quadros

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Quoted 4 lines

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---

How does v2 of this patch differ from v1?

Quoted 9 lines
 drivers/usb/core/message.c | 6 ++++++
 drivers/usb/core/usb.c     | 7 +++++++
 2 files changed, 13 insertions(+)
drivers/usb/core/message.c +0 -0
index 0406a59..66364ea 100644

drivers/usb/core/message.c free_interfaces: · line 1871

Quoted 9 lines
drivers/usb/core/message.c +6 -0
@@ -1863,6 +1863,12 @@ free_interfaces:
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		/* Propagate dma_pfn_offset to USB interface.
+		 * This is especially required by mass storage interface
+		 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+		 * to set the bounce buffer limit based on dma_pfn_offset.
+		 */
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;

I really think this comment isn't necessary. It seems pretty obvious that if you're copying DMA-related fields from one device to another then you'll want to copy the dma_pfn_offset along with everything else. No?

The explanation in the Changelog is enough, IMO.

Quoted 8 lines
drivers/usb/core/message.c +0 -0
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +0 -0
index 5e80697..7dde93d 100644

drivers/usb/core/usb.c struct usb_device *usb_alloc_dev(struct usb_device *parent, · line 450

Quoted 10 lines
drivers/usb/core/usb.c +7 -0
@@ -441,6 +441,13 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	/* Propagate bus controller's dma_pfn_offset to USB device,
+	 * as it would be needed to be propagated to USB interfaces.
+	 * This is especially required by mass storage interface
+	 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+	 * to set the bounce buffer limit based on dma_pfn_offset.
+	 */
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;

Same here.

drivers/usb/core/usb.c struct usb_device *usb_alloc_dev(struct usb_device *parent, · line 453

Quoted 3 lines
drivers/usb/core/usb.c +0 -0
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;

Alan Stern

#3 lore raw

Re: [PATCH v2] usb: core: setup dma_pfn_offset for USB devices and, interfaces

Hi Alan,

Quoted 25 lines · Alan Stern
Quoted 23 lines · Roger Quadros

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Quoted 6 lines
Quoted 4 lines

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---

How does v2 of this patch differ from v1?

No functional difference.
Just added more explanation in the commit log.

Quoted 25 lines
Quoted 10 lines
 drivers/usb/core/message.c | 6 ++++++
 drivers/usb/core/usb.c     | 7 +++++++
 2 files changed, 13 insertions(+)
drivers/usb/core/message.c +0 -0
index 0406a59..66364ea 100644

drivers/usb/core/message.c free_interfaces: · line 1871

Quoted 9 lines
drivers/usb/core/message.c +6 -0
@@ -1863,6 +1863,12 @@ free_interfaces:
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		/* Propagate dma_pfn_offset to USB interface.
+		 * This is especially required by mass storage interface
+		 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+		 * to set the bounce buffer limit based on dma_pfn_offset.
+		 */
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;

I really think this comment isn't necessary. It seems pretty obvious that if you're copying DMA-related fields from one device to another then you'll want to copy the dma_pfn_offset along with everything else. No?

Quoted 1 line

The explanation in the Changelog is enough, IMO.

Fine with me.

Quoted 13 lines
Quoted 13 lines
drivers/usb/core/message.c +0 -0
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +1 -0
index 5e80697..7dde93d 100644
@@ -441,6 +441,13 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	/* Propagate bus controller's dma_pfn_offset to USB device,
Quoted 8 lines

drivers/usb/core/usb.c struct usb_device *usb_alloc_dev(struct usb_device *parent, · line 450

Quoted 6 lines
drivers/usb/core/usb.c +6 -0
+	 * as it would be needed to be propagated to USB interfaces.
+	 * This is especially required by mass storage interface
+	 * which relies on SCSI layer and scsi_calculate_bounce_limit()
+	 * to set the bounce buffer limit based on dma_pfn_offset.
+	 */
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;

Same here.

OK.

Quoted 7 lines

drivers/usb/core/usb.c struct usb_device *usb_alloc_dev(struct usb_device *parent, · line 453

Quoted 4 lines
drivers/usb/core/usb.c +0 -0
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;

Alan Stern

cheers,
-roger

#4 lore raw

Re: [PATCH v2] usb: core: setup dma_pfn_offset for USB devices and, interfaces

In reply to Alan Stern
Quoted 20 lines · CEST Alan Stern
Quoted 5 lines
drivers/usb/core/message.c +0 -0
index 0406a59..66364ea 100644

drivers/usb/core/message.c free_interfaces: · line 1871

Quoted 9 lines
drivers/usb/core/message.c +6 -0
@@ -1863,6 +1863,12 @@ free_interfaces:
              intf->dev.type = &usb_if_device_type;
              intf->dev.groups = usb_interface_groups;
              intf->dev.dma_mask = dev->dev.dma_mask;
+             /* Propagate dma_pfn_offset to USB interface.
+              * This is especially required by mass storage interface
+              * which relies on SCSI layer and scsi_calculate_bounce_limit()
+              * to set the bounce buffer limit based on dma_pfn_offset.
+              */
+             intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;

I really think this comment isn't necessary. It seems pretty obvious that if you're copying DMA-related fields from one device to another then you'll want to copy the dma_pfn_offset along with everything else. No?

Quoted 1 line

The explanation in the Changelog is enough, IMO.

I asked for a code comment here as what we are doing here is a bit fishy, but the comment (evidently) doesn't quite capture it. I would add (above the dma_mask assignment) something like

/*
* Fake a dma_mask/offset for the USB device:
* We cannot really use the dma-mapping API (dma_alloc_* and
* dma_map_*) for USB devices but instead need to use
* usb_alloc_coherent and pass data in 'urb's, but some subsystems
* manually look into the mask/offset pair to determine whether
* they need bounce buffers.
* Note: calling dma_set_mask() on a USB device would set the
* mask for the entire HCD, so don't do that.
*/

Arnd

#5 lore raw

Re: [PATCH v2] usb: core: setup dma_pfn_offset for USB devices and, interfaces

Quoted 27 lines · Arnd Bergmann
Quoted 20 lines · CEST Alan Stern
Quoted 5 lines
drivers/usb/core/message.c +0 -0
index 0406a59..66364ea 100644

drivers/usb/core/message.c free_interfaces: · line 1871

Quoted 9 lines
drivers/usb/core/message.c +6 -0
@@ -1863,6 +1863,12 @@ free_interfaces:
              intf->dev.type = &usb_if_device_type;
              intf->dev.groups = usb_interface_groups;
              intf->dev.dma_mask = dev->dev.dma_mask;
+             /* Propagate dma_pfn_offset to USB interface.
+              * This is especially required by mass storage interface
+              * which relies on SCSI layer and scsi_calculate_bounce_limit()
+              * to set the bounce buffer limit based on dma_pfn_offset.
+              */
+             intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;

I really think this comment isn't necessary. It seems pretty obvious that if you're copying DMA-related fields from one device to another then you'll want to copy the dma_pfn_offset along with everything else. No?

Quoted 1 line

The explanation in the Changelog is enough, IMO.

I asked for a code comment here as what we are doing here is a bit fishy, but the comment (evidently) doesn't quite capture it. I would add (above the dma_mask assignment) something like

Quoted 10 lines

/*
* Fake a dma_mask/offset for the USB device:
* We cannot really use the dma-mapping API (dma_alloc_* and
* dma_map_*) for USB devices but instead need to use
* usb_alloc_coherent and pass data in 'urb's, but some subsystems
* manually look into the mask/offset pair to determine whether
* they need bounce buffers.
* Note: calling dma_set_mask() on a USB device would set the
* mask for the entire HCD, so don't do that.
*/

I'm okay with this. But at least put this explanation only in usb.c, and have the comment in message.c refer back to it.

(Also, 'urb' doesn't need to be in quotes and should be capitalized:
URBs.)

Alan Stern

#6 lore raw

PATCH v3 1/1 usb: core: setup dma_pfn_offset for USB devices and, interfaces

In reply to Roger Quadros

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---
Changelog:

v3:
- removed comments from code as commit log is sufficient. v2:
- added more information in commit log and code.

 drivers/usb/core/message.c | 1 +
 drivers/usb/core/usb.c     | 1 +
 2 files changed, 2 insertions(+)
drivers/usb/core/message.c +1 -0
index 0406a59..6856b7a 100644
@@ -1863,6 +1863,7 @@ free_interfaces:
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +1 -0
index 5e80697..d81791a 100644
@@ -441,6 +441,7 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;
Signature · 1 line
2.7.4
#7 lore raw

PATCH v4 1/1 usb: core: setup dma_pfn_offset for USB devices and, interfaces

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

---
Changelog:

v4:
- added comment in the code as to why we need to set dma_mask and dma_pfn_offset for usb devices.
v3:
- removed comments from code as commit log is sufficient. v2:
- added more information in commit log and code.

 drivers/usb/core/message.c |  5 +++++
 drivers/usb/core/usb.c     | 11 +++++++++++
 2 files changed, 16 insertions(+)
drivers/usb/core/message.c +5 -0
index 0406a59..bb617df 100644
@@ -1862,7 +1862,12 @@ free_interfaces:
 		intf->dev.bus = &usb_bus_type;
 		intf->dev.type = &usb_if_device_type;
 		intf->dev.groups = usb_interface_groups;
+		/*
+		 * Please refer to usb_alloc_dev() to see why we set
+		 * dma_mask and dma_pfn_offset.
+		 */
 		intf->dev.dma_mask = dev->dev.dma_mask;
+		intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset;
 		INIT_WORK(&intf->reset_ws, __usb_queue_reset_device);
 		intf->minor = -1;
 		device_initialize(&intf->dev);
drivers/usb/core/usb.c +11 -0
index 5e80697..5921514 100644
@@ -440,7 +440,18 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,
 	dev->dev.bus = &usb_bus_type;
 	dev->dev.type = &usb_device_type;
 	dev->dev.groups = usb_device_groups;
+	/*
+	 * Fake a dma_mask/offset for the USB device:
+	 * We cannot really use the dma-mapping API (dma_alloc_* and
+	 * dma_map_*) for USB devices but instead need to use
+	 * usb_alloc_coherent and pass data in 'urb's, but some subsystems
+	 * manually look into the mask/offset pair to determine whether
+	 * they need bounce buffers.
+	 * Note: calling dma_set_mask() on a USB device would set the
+	 * mask for the entire HCD, so don't do that.
+	 */
 	dev->dev.dma_mask = bus->controller->dma_mask;
+	dev->dev.dma_pfn_offset = bus->controller->dma_pfn_offset;
 	set_dev_node(&dev->dev, dev_to_node(bus->controller));
 	dev->state = USB_STATE_ATTACHED;
 	dev->lpm_disable_count = 1;
Signature · 1 line
2.7.4
Acked-by on v4 1/1
#8 lore raw

Re: [PATCH v4] usb: core: setup dma_pfn_offset for USB devices and, interfaces

Quoted 23 lines · CEST Roger Quadros

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Quoted 3 lines

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset

Acked-by
#9 lore raw

Re: [PATCH v4] usb: core: setup dma_pfn_offset for USB devices and, interfaces

Quoted 24 lines · Arnd Bergmann
Quoted 23 lines · CEST Roger Quadros

If dma_pfn_offset is not inherited correctly from the host controller, it might result in sub-optimal configuration as bounce buffer limit might be set to less than optimal level.

Consider the mass storage device case.
USB storage driver creates a scsi host for the mass storage interface in drivers/usb/storage/usb.c
The scsi host parent device is nothing but the the USB interface device. Now, __scsi_init_queue() calls scsi_calculate_bounce_limit() to find out and set the block layer bounce limit. scsi_calculate_bounce_limit() uses dma_max_pfn(host_dev) to get the bounce_limit. host_dev is nothing but the device representing the mass storage interface.
If that device doesn't have the right dma_pfn_offset, then dma_max_pfn() is messed up and the bounce buffer limit is wrong.

e.g. On Keystone 2 systems, dma_max_pfn() is 0x87FFFF and dma_mask_pfn is 0xFFFFF. Consider a mass storage use case: Without this patch, usb scsi host device (usb-storage) will get a dma_pfn_offset of 0 resulting in a dma_max_pfn() of 0xFFFFF within the scsi layer (scsi_calculate_bounce_limit()).
This will result in bounce buffers being unnecessarily used.

Quoted 5 lines
Quoted 3 lines

Hint: On 32-bit ARM platforms dma_max_pfn() = dma_mask_pfn + dma_pfn_offset