Skip to contentSkip to conversation

usb dma_mask fixups

PATCH In review Christoph Hellwig 4 participants 12 messages Sep 3, 2019 – Sep 4, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

Hi all,

the first patch fixes the ohci-sm501 regression that Guenther reported due to the platform device dma_mask changes. The second one ports that fix to another driver that works the same way. The rest cleans up various loose ends left over from the dma related usb changes in the last two merge windows.

#1 lore raw

PATCH 1/6 usb/ohci-sm501: remove the HCD_DMA flag

This driver doesn't support normal DMA, only direct access to its local memory. Remove the HCD_DMA flag to properly express that fact.

Fixes 1e4946c4412e ("usb: add a HCD_DMA flag instead of guestimating DMA capabilities")
---
 drivers/usb/host/ohci-sm501.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/usb/host/ohci-sm501.c +1 -1
index 0b2aea6e28d4..c158cda9e4b9 100644
@@ -49,7 +49,7 @@ static const struct hc_driver ohci_sm501_hc_driver = {
 	 * generic hardware linkage
 	 */
 	.irq =			ohci_irq,
-	.flags =		HCD_USB11 | HCD_DMA | HCD_MEMORY,
+	.flags =		HCD_USB11 | HCD_MEMORY,
 
 	/*
 	 * basic lifecycle operations
Signature · 1 line
2.20.1
Tested-by on v1 1/6
#2 lore raw

Re: [PATCH 1/6] usb/ohci-sm501: remove the HCD_DMA flag

Quoted 6 lines · Christoph Hellwig

This driver doesn't support normal DMA, only direct access to its local memory. Remove the HCD_DMA flag to properly express that fact.

Fixes 1e4946c4412e ("usb: add a HCD_DMA flag instead of guestimating DMA capabilities")

on top of next-20190903.

Thanks!

Guenter

Quoted 20 lines
---
 drivers/usb/host/ohci-sm501.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/usb/host/ohci-sm501.c +1 -1
index 0b2aea6e28d4..c158cda9e4b9 100644
@@ -49,7 +49,7 @@ static const struct hc_driver ohci_sm501_hc_driver = {
 	 * generic hardware linkage
 	 */
 	.irq =			ohci_irq,
-	.flags =		HCD_USB11 | HCD_DMA | HCD_MEMORY,
+	.flags =		HCD_USB11 | HCD_MEMORY,
 
 	/*
 	 * basic lifecycle operations
Signature · 1 line
2.20.1
#3 lore raw

Re: [PATCH 1/6] usb/ohci-sm501: remove the HCD_DMA flag

In reply to Christoph Hellwig
Quoted 4 lines · Christoph Hellwig

This driver doesn't support normal DMA, only direct access to its local memory. Remove the HCD_DMA flag to properly express that fact.

Fixes 1e4946c4412e ("usb: add a HCD_DMA flag instead of guestimating DMA capabilities")

Wrong sha1. I fixed it up in my tree.

#4 lore raw

PATCH 2/6 usb/ohci-tmio: remove the HCD_DMA flag

In reply to Christoph Hellwig

This driver doesn't support normal DMA, only direct access to its local memory. Remove the HCD_DMA flag to properly express that fact.

Fixes 1e4946c4412e ("usb: add a HCD_DMA flag instead of guestimating DMA capabilities")
---
 drivers/usb/host/ohci-tmio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/usb/host/ohci-tmio.c +1 -1
index 8edbacd3eb17..d5a293a707b6 100644
@@ -153,7 +153,7 @@ static const struct hc_driver ohci_tmio_hc_driver = {
 
 	/* generic hardware linkage */
 	.irq =			ohci_irq,
-	.flags =		HCD_USB11 | HCD_DMA | HCD_MEMORY,
+	.flags =		HCD_USB11 | HCD_MEMORY,
 
 	/* basic lifecycle operations */
 	.start =		ohci_tmio_start,
Signature · 1 line
2.20.1
#5 lore raw

PATCH 3/6 usb-storage: use hcd_uses_dma to check for DMA capabilities

In reply to Christoph Hellwig

The dma_mask on its own doesn't mean much. Instead check for the actual flag.

---
 drivers/usb/storage/scsiglue.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/usb/storage/scsiglue.c +3 -3
index 05b80211290d..7e0bc640c237 100644
@@ -40,6 +40,7 @@
 #include <scsi/scsi_eh.h>
 
 #include "usb.h"
+#include <linux/usb/hcd.h>
 #include "scsiglue.h"
 #include "debug.h"
 #include "transport.h"
@@ -141,11 +142,10 @@ static int slave_configure(struct scsi_device *sdev)
 
 	/*
 	 * Some USB host controllers can't do DMA; they have to use PIO.
-	 * They indicate this by setting their dma_mask to NULL.  For
-	 * such controllers we need to make sure the block layer sets
+	 * For such controllers we need to make sure the block layer sets
 	 * up bounce buffers in addressable memory.
 	 */
-	if (!us->pusb_dev->bus->controller->dma_mask)
+	if (!hcd_uses_dma(bus_to_hcd(us->pusb_dev->bus)))
 		blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_HIGH);
 
 	/*
Signature · 1 line
2.20.1
#6 lore raw

PATCH 4/6 usb: remove a stale comment in hcd_alloc_coherent

In reply to Christoph Hellwig

Now that we have the local memory pool implemented there is no need to use dma_declare_coherent_memory.

---
 drivers/usb/core/hcd.c | 3 ---
 1 file changed, 3 deletions(-)
drivers/usb/core/hcd.c +0 -3
index add2af4af766..4ccfc8e59604 100644
@@ -1249,9 +1249,6 @@ EXPORT_SYMBOL_GPL(usb_hcd_unlink_urb_from_ep);
  * To support host controllers with limited dma capabilities we provide dma
  * bounce buffers. This feature can be enabled by initializing
  * hcd->localmem_pool using usb_hcd_setup_local_mem().
- * For this to work properly the host controller code must first use the
- * function dma_declare_coherent_memory() to point out which memory area
- * that should be used for dma allocations.
  *
  * The initialized hcd->localmem_pool then tells the usb code to allocate all
  * data for dma using the genalloc API.
Signature · 1 line
2.20.1
#7 lore raw

PATCH 5/6 usb: remove commented out dma wrappers

In reply to Christoph Hellwig

These wrappers have never seen use and have been commented out for a long time. Remove them for good.

---
 drivers/usb/core/usb.c | 222 -----------------------------------------
 1 file changed, 222 deletions(-)
drivers/usb/core/usb.c +0 -222
index 0ab8738047da..f16c26dc079d 100644
@@ -933,228 +933,6 @@ void usb_free_coherent(struct usb_device *dev, size_t size, void *addr,
 }
 EXPORT_SYMBOL_GPL(usb_free_coherent);
 
-/**
- * usb_buffer_map - create DMA mapping(s) for an urb
- * @urb: urb whose transfer_buffer/setup_packet will be mapped
- *
- * URB_NO_TRANSFER_DMA_MAP is added to urb->transfer_flags if the operation
- * succeeds. If the device is connected to this system through a non-DMA
- * controller, this operation always succeeds.
- *
- * This call would normally be used for an urb which is reused, perhaps
- * as the target of a large periodic transfer, with usb_buffer_dmasync()
- * calls to synchronize memory and dma state.
- *
- * Reverse the effect of this call with usb_buffer_unmap().
- *
- * Return: Either %NULL (indicating no buffer could be mapped), or @urb.
- *
- */
-#if 0
-struct urb *usb_buffer_map(struct urb *urb)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!urb
-			|| !urb->dev
-			|| !(bus = urb->dev->bus)
-			|| !(controller = bus->sysdev))
-		return NULL;
-
-	if (controller->dma_mask) {
-		urb->transfer_dma = dma_map_single(controller,
-			urb->transfer_buffer, urb->transfer_buffer_length,
-			usb_pipein(urb->pipe)
-				? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-	/* FIXME generic api broken like pci, can't report errors */
-	/* if (urb->transfer_dma == DMA_ADDR_INVALID) return 0; */
-	} else
-		urb->transfer_dma = ~0;
-	urb->transfer_flags |= URB_NO_TRANSFER_DMA_MAP;
-	return urb;
-}
-EXPORT_SYMBOL_GPL(usb_buffer_map);
-#endif  /*  0  */
-
-/* XXX DISABLED, no users currently.  If you wish to re-enable this
- * XXX please determine whether the sync is to transfer ownership of
- * XXX the buffer from device to cpu or vice verse, and thusly use the
- * XXX appropriate _for_{cpu,device}() method.  -DaveM
- */
-#if 0
-
-/**
- * usb_buffer_dmasync - synchronize DMA and CPU view of buffer(s)
- * @urb: urb whose transfer_buffer/setup_packet will be synchronized
- */
-void usb_buffer_dmasync(struct urb *urb)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!urb
-			|| !(urb->transfer_flags & URB_NO_TRANSFER_DMA_MAP)
-			|| !urb->dev
-			|| !(bus = urb->dev->bus)
-			|| !(controller = bus->sysdev))
-		return;
-
-	if (controller->dma_mask) {
-		dma_sync_single_for_cpu(controller,
-			urb->transfer_dma, urb->transfer_buffer_length,
-			usb_pipein(urb->pipe)
-				? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-		if (usb_pipecontrol(urb->pipe))
-			dma_sync_single_for_cpu(controller,
-					urb->setup_dma,
-					sizeof(struct usb_ctrlrequest),
-					DMA_TO_DEVICE);
-	}
-}
-EXPORT_SYMBOL_GPL(usb_buffer_dmasync);
-#endif
-
-/**
- * usb_buffer_unmap - free DMA mapping(s) for an urb
- * @urb: urb whose transfer_buffer will be unmapped
- *
- * Reverses the effect of usb_buffer_map().
- */
-#if 0
-void usb_buffer_unmap(struct urb *urb)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!urb
-			|| !(urb->transfer_flags & URB_NO_TRANSFER_DMA_MAP)
-			|| !urb->dev
-			|| !(bus = urb->dev->bus)
-			|| !(controller = bus->sysdev))
-		return;
-
-	if (controller->dma_mask) {
-		dma_unmap_single(controller,
-			urb->transfer_dma, urb->transfer_buffer_length,
-			usb_pipein(urb->pipe)
-				? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-	}
-	urb->transfer_flags &= ~URB_NO_TRANSFER_DMA_MAP;
-}
-EXPORT_SYMBOL_GPL(usb_buffer_unmap);
-#endif  /*  0  */
-
-#if 0
-/**
- * usb_buffer_map_sg - create scatterlist DMA mapping(s) for an endpoint
- * @dev: device to which the scatterlist will be mapped
- * @is_in: mapping transfer direction
- * @sg: the scatterlist to map
- * @nents: the number of entries in the scatterlist
- *
- * Return: Either < 0 (indicating no buffers could be mapped), or the
- * number of DMA mapping array entries in the scatterlist.
- *
- * Note:
- * The caller is responsible for placing the resulting DMA addresses from
- * the scatterlist into URB transfer buffer pointers, and for setting the
- * URB_NO_TRANSFER_DMA_MAP transfer flag in each of those URBs.
- *
- * Top I/O rates come from queuing URBs, instead of waiting for each one
- * to complete before starting the next I/O.   This is particularly easy
- * to do with scatterlists.  Just allocate and submit one URB for each DMA
- * mapping entry returned, stopping on the first error or when all succeed.
- * Better yet, use the usb_sg_*() calls, which do that (and more) for you.
- *
- * This call would normally be used when translating scatterlist requests,
- * rather than usb_buffer_map(), since on some hardware (with IOMMUs) it
- * may be able to coalesce mappings for improved I/O efficiency.
- *
- * Reverse the effect of this call with usb_buffer_unmap_sg().
- */
-int usb_buffer_map_sg(const struct usb_device *dev, int is_in,
-		      struct scatterlist *sg, int nents)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!dev
-			|| !(bus = dev->bus)
-			|| !(controller = bus->sysdev)
-			|| !controller->dma_mask)
-		return -EINVAL;
-
-	/* FIXME generic api broken like pci, can't report errors */
-	return dma_map_sg(controller, sg, nents,
-			is_in ? DMA_FROM_DEVICE : DMA_TO_DEVICE) ? : -ENOMEM;
-}
-EXPORT_SYMBOL_GPL(usb_buffer_map_sg);
-#endif
-
-/* XXX DISABLED, no users currently.  If you wish to re-enable this
- * XXX please determine whether the sync is to transfer ownership of
- * XXX the buffer from device to cpu or vice verse, and thusly use the
- * XXX appropriate _for_{cpu,device}() method.  -DaveM
- */
-#if 0
-
-/**
- * usb_buffer_dmasync_sg - synchronize DMA and CPU view of scatterlist buffer(s)
- * @dev: device to which the scatterlist will be mapped
- * @is_in: mapping transfer direction
- * @sg: the scatterlist to synchronize
- * @n_hw_ents: the positive return value from usb_buffer_map_sg
- *
- * Use this when you are re-using a scatterlist's data buffers for
- * another USB request.
- */
-void usb_buffer_dmasync_sg(const struct usb_device *dev, int is_in,
-			   struct scatterlist *sg, int n_hw_ents)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!dev
-			|| !(bus = dev->bus)
-			|| !(controller = bus->sysdev)
-			|| !controller->dma_mask)
-		return;
-
-	dma_sync_sg_for_cpu(controller, sg, n_hw_ents,
-			    is_in ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-}
-EXPORT_SYMBOL_GPL(usb_buffer_dmasync_sg);
-#endif
-
-#if 0
-/**
- * usb_buffer_unmap_sg - free DMA mapping(s) for a scatterlist
- * @dev: device to which the scatterlist will be mapped
- * @is_in: mapping transfer direction
- * @sg: the scatterlist to unmap
- * @n_hw_ents: the positive return value from usb_buffer_map_sg
- *
- * Reverses the effect of usb_buffer_map_sg().
- */
-void usb_buffer_unmap_sg(const struct usb_device *dev, int is_in,
-			 struct scatterlist *sg, int n_hw_ents)
-{
-	struct usb_bus		*bus;
-	struct device		*controller;
-
-	if (!dev
-			|| !(bus = dev->bus)
-			|| !(controller = bus->sysdev)
-			|| !controller->dma_mask)
-		return;
-
-	dma_unmap_sg(controller, sg, n_hw_ents,
-			is_in ? DMA_FROM_DEVICE : DMA_TO_DEVICE);
-}
-EXPORT_SYMBOL_GPL(usb_buffer_unmap_sg);
-#endif
-
 /*
  * Notifications of device and interface registration
  */
Signature · 1 line
2.20.1
#8 lore raw

PATCH 6/6 mfd: don't select DMA_DECLARE_COHERENT for the sm501 and tc6393xb drivers

In reply to Christoph Hellwig

Now that these drivers use the usb localmem pool there is no need to select DMA_DECLARE_COHERENT.

---
 drivers/mfd/Kconfig | 2 --
 1 file changed, 2 deletions(-)
drivers/mfd/Kconfig +0 -2
index f129f9678940..c8cbde59bbf6 100644
@@ -1105,7 +1105,6 @@ config MFD_SI476X_CORE
 config MFD_SM501
 	tristate "Silicon Motion SM501"
 	depends on HAS_DMA
-	select DMA_DECLARE_COHERENT
 	 ---help---
 	  This is the core driver for the Silicon Motion SM501 multimedia
 	  companion chip. This device is a multifunction device which may
@@ -1714,7 +1713,6 @@ config MFD_TC6393XB
 	select GPIOLIB
 	select MFD_CORE
 	select MFD_TMIO
-	select DMA_DECLARE_COHERENT
 	help
 	  Support for Toshiba Mobile IO Controller TC6393XB
 
Signature · 1 line
2.20.1
#9 lore raw
In reply to Christoph Hellwig
Quoted 7 lines · Christoph Hellwig

Hi all,

the first patch fixes the ohci-sm501 regression that Guenther reported due to the platform device dma_mask changes. The second one ports that fix to another driver that works the same way. The rest cleans up various loose ends left over from the dma related usb changes in the last two merge windows.

Thanks for these, all now queued up.

greg k-h

Acked-by
#10 lore raw
Quoted 10 lines · Greg Kroah-Hartman
Quoted 7 lines · Christoph Hellwig

Hi all,

the first patch fixes the ohci-sm501 regression that Guenther reported due to the platform device dma_mask changes. The second one ports that fix to another driver that works the same way. The rest cleans up various loose ends left over from the dma related usb changes in the last two merge windows.

Thanks for these, all now queued up.

Did you queue the MFD patch too?

If so (and you can rebase ;) ), please feel free to add my:

If not, no sweat.

Signature · 4 lines
Lee Jones [李琼斯]
Linaro Services Technical Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog
#11 lore raw
Quoted 13 lines · Lee Jones
Quoted 10 lines · Greg Kroah-Hartman
Quoted 7 lines · Christoph Hellwig

Hi all,

the first patch fixes the ohci-sm501 regression that Guenther reported due to the platform device dma_mask changes. The second one ports that fix to another driver that works the same way. The rest cleans up various loose ends left over from the dma related usb changes in the last two merge windows.

Thanks for these, all now queued up.

Quoted 7 lines

Did you queue the MFD patch too?

If so (and you can rebase ;) ), please feel free to add my:

If not, no sweat.

I already applied it, and no, sorry, I can't rebase. But thanks for the review!

greg k-h