[PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[FUJITA Tomonori]

This is an updated version of the patchset to add per-device
dma_mapping_ops support for CONFIG_X86_64 like POWER architecture
does:

http://lkml.org/lkml/2008/5/13/36

This is against 2.6.26-rc2-mm1 (the changes since the v1 are pretty
trivial like dropping the change for v850 arch).

This enables us to cleanly fix the Calgary IOMMU issue that some
devices are not behind the IOMMU [1].

I think that per-device dma_mapping_ops support would be also helpful
for KVM people to support PCI passthrough but Andi thinks that this
makes it difficult to support the PCI passthrough (see the above
thread). So I CC'ed this to KVM camp. Comments are appreciated.

A pointer to dma_mapping_ops to struct dev_archdata is added. If the
pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If
it's NULL, the system-wide dma_ops pointer is used as before.

If it's useful for KVM people, I plan to implement a mechanism to
register a hook called when a new pci (or dma capable) device is
created (it works with hot plugging). It enables IOMMUs to set up an
appropriate dma_mapping_ops per device.

The major obstacle is that dma_mapping_error doesn't take a pointer to
the device unlike other DMA operations. So x86 can't have
dma_mapping_ops per device. Note all the POWER IOMMUs use the same
dma_mapping_error function so this is not a problem for POWER but x86
IOMMUs use different dma_mapping_error functions.

The first patch adds the device argument to dma_mapping_error. The
patch is trivial but large since it touches lots of drivers and
dma-mapping.h in all the architecture.

[1] http://lkml.org/lkml/2008/5/8/423

[PATCH v2 -mm 1/2] add the device argument to dma_mapping_error

[FUJITA Tomonori]

dma_mapping_error doesn't take a pointer to the device unlike other
DMA operations. So we can't have dma_mapping_ops per device.

Note that POWER already has dma_mapping_ops per device but all the
POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use
different dma_mapping_error functions. So dma_mapping_error needs the
device argument.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
 Documentation/DMA-API.txt                      |    4 +-
 arch/arm/common/dmabounce.c                    |    2 +-
 arch/ia64/hp/common/hwsw_iommu.c               |    5 ++-
 arch/ia64/hp/common/sba_iommu.c                |    2 +-
 arch/ia64/sn/pci/pci_dma.c                     |    2 +-
 arch/mips/mm/dma-default.c                     |    2 +-
 arch/powerpc/platforms/cell/celleb_scc_pciex.c |    2 +-
 arch/powerpc/platforms/cell/spider-pci.c       |    2 +-
 arch/powerpc/platforms/iseries/mf.c            |    2 +-
 arch/x86/kernel/pci-gart_64.c                  |    1 -
 arch/x86/kernel/pci-nommu.c                    |    2 +-
 drivers/firewire/fw-iso.c                      |    2 +-
 drivers/firewire/fw-ohci.c                     |    2 +-
 drivers/firewire/fw-sbp2.c                     |    8 +++---
 drivers/infiniband/hw/ipath/ipath_sdma.c       |    2 +-
 drivers/infiniband/hw/ipath/ipath_user_sdma.c  |    6 ++--
 drivers/infiniband/hw/mthca/mthca_eq.c         |    2 +-
 drivers/media/dvb/pluto2/pluto2.c              |    2 +-
 drivers/net/arm/ep93xx_eth.c                   |    4 +-
 drivers/net/b44.c                              |   22 +++++++++-------
 drivers/net/bnx2x.c                            |    2 +-
 drivers/net/e100.c                             |    2 +-
 drivers/net/e1000e/ethtool.c                   |    4 +-
 drivers/net/e1000e/netdev.c                    |   11 ++++---
 drivers/net/ibmveth.c                          |   32 ++++++++++++-----------
 drivers/net/iseries_veth.c                     |    4 +-
 drivers/net/mlx4/eq.c                          |    2 +-
 drivers/net/pasemi_mac.c                       |    6 ++--
 drivers/net/qla3xxx.c                          |   12 ++++----
 drivers/net/sfc/rx.c                           |    4 +-
 drivers/net/sfc/tx.c                           |    7 +++--
 drivers/net/spider_net.c                       |    4 +-
 drivers/net/tc35815.c                          |    4 +-
 drivers/net/wireless/ath5k/base.c              |    4 +-
 drivers/net/wireless/b43/dma.c                 |    2 +-
 drivers/net/wireless/b43legacy/dma.c           |    2 +-
 drivers/scsi/ibmvscsi/ibmvscsi.c               |    4 +-
 drivers/scsi/ibmvscsi/ibmvstgt.c               |    2 +-
 drivers/scsi/ibmvscsi/rpa_vscsi.c              |    2 +-
 drivers/spi/atmel_spi.c                        |    4 +-
 drivers/spi/au1550_spi.c                       |    6 ++--
 drivers/spi/omap2_mcspi.c                      |    4 +-
 drivers/spi/pxa2xx_spi.c                       |    4 +-
 drivers/spi/spi_imx.c                          |    6 ++--
 include/asm-alpha/dma-mapping.h                |    6 ++--
 include/asm-alpha/pci.h                        |    2 +-
 include/asm-arm/dma-mapping.h                  |    2 +-
 include/asm-avr32/dma-mapping.h                |    2 +-
 include/asm-cris/dma-mapping.h                 |    2 +-
 include/asm-frv/dma-mapping.h                  |    2 +-
 include/asm-generic/dma-mapping-broken.h       |    2 +-
 include/asm-generic/dma-mapping.h              |    4 +-
 include/asm-generic/pci-dma-compat.h           |    4 +-
 include/asm-ia64/machvec.h                     |    2 +-
 include/asm-m68k/dma-mapping.h                 |    2 +-
 include/asm-mips/dma-mapping.h                 |    2 +-
 include/asm-mn10300/dma-mapping.h              |    2 +-
 include/asm-parisc/dma-mapping.h               |    2 +-
 include/asm-powerpc/dma-mapping.h              |    2 +-
 include/asm-sh/dma-mapping.h                   |    2 +-
 include/asm-sparc/pci.h                        |    3 +-
 include/asm-sparc64/dma-mapping.h              |    2 +-
 include/asm-sparc64/pci.h                      |    5 ++-
 include/asm-x86/dma-mapping.h                  |    7 +++--
 include/asm-x86/swiotlb.h                      |    2 +-
 include/asm-xtensa/dma-mapping.h               |    2 +-
 include/linux/i2o.h                            |    2 +-
 include/rdma/ib_verbs.h                        |    2 +-
 lib/swiotlb.c                                  |    4 +-
 69 files changed, 144 insertions(+), 135 deletions(-)

diff --git a/Documentation/DMA-API.txt b/Documentation/DMA-API.txt
index 80d1504..d8b63d1 100644
--- a/Documentation/DMA-API.txt
+++ b/Documentation/DMA-API.txt
@@ -298,10 +298,10 @@ recommended that you never use these unless you really know what the
 cache width is.
 
 int
-dma_mapping_error(dma_addr_t dma_addr)
+dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 
 int
-pci_dma_mapping_error(dma_addr_t dma_addr)
+pci_dma_mapping_error(struct pci_dev *hwdev, dma_addr_t dma_addr)
 
 In some circumstances dma_map_single and dma_map_page will fail to create
 a mapping. A driver can check for these errors by testing the returned
diff --git a/arch/arm/common/dmabounce.c b/arch/arm/common/dmabounce.c
index 59800a5..f10d348 100644
--- a/arch/arm/common/dmabounce.c
+++ b/arch/arm/common/dmabounce.c
@@ -280,7 +280,7 @@ unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size,
 	/*
 	 * Trying to unmap an invalid mapping
 	 */
-	if (dma_mapping_error(dma_addr)) {
+	if (dma_mapping_error(dev, dma_addr)) {
 		dev_err(dev, "Trying to unmap invalid mapping\n");
 		return;
 	}
diff --git a/arch/ia64/hp/common/hwsw_iommu.c b/arch/ia64/hp/common/hwsw_iommu.c
index 1c44ec2..88b6e6f 100644
--- a/arch/ia64/hp/common/hwsw_iommu.c
+++ b/arch/ia64/hp/common/hwsw_iommu.c
@@ -186,9 +186,10 @@ hwsw_dma_supported (struct device *dev, u64 mask)
 }
 
 int
-hwsw_dma_mapping_error (dma_addr_t dma_addr)
+hwsw_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
-	return hwiommu_dma_mapping_error (dma_addr) || swiotlb_dma_mapping_error(dma_addr);
+	return hwiommu_dma_mapping_error(dev, dma_addr) ||
+		swiotlb_dma_mapping_error(dev, dma_addr);
 }
 
 EXPORT_SYMBOL(hwsw_dma_mapping_error);
diff --git a/arch/ia64/hp/common/sba_iommu.c b/arch/ia64/hp/common/sba_iommu.c
index 34421ae..4956be4 100644
--- a/arch/ia64/hp/common/sba_iommu.c
+++ b/arch/ia64/hp/common/sba_iommu.c
@@ -2147,7 +2147,7 @@ sba_dma_supported (struct device *dev, u64 mask)
 }
 
 int
-sba_dma_mapping_error (dma_addr_t dma_addr)
+sba_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/arch/ia64/sn/pci/pci_dma.c b/arch/ia64/sn/pci/pci_dma.c
index 52175af..53ebb64 100644
--- a/arch/ia64/sn/pci/pci_dma.c
+++ b/arch/ia64/sn/pci/pci_dma.c
@@ -350,7 +350,7 @@ void sn_dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 }
 EXPORT_SYMBOL(sn_dma_sync_sg_for_device);
 
-int sn_dma_mapping_error(dma_addr_t dma_addr)
+int sn_dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
index ae39dd8..891312f 100644
--- a/arch/mips/mm/dma-default.c
+++ b/arch/mips/mm/dma-default.c
@@ -348,7 +348,7 @@ void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nele
 
 EXPORT_SYMBOL(dma_sync_sg_for_device);
 
-int dma_mapping_error(dma_addr_t dma_addr)
+int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/arch/powerpc/platforms/cell/celleb_scc_pciex.c b/arch/powerpc/platforms/cell/celleb_scc_pciex.c
index 31da84c..d3f1102 100644
--- a/arch/powerpc/platforms/cell/celleb_scc_pciex.c
+++ b/arch/powerpc/platforms/cell/celleb_scc_pciex.c
@@ -280,7 +280,7 @@ static int __init scc_pciex_iowa_init(struct iowa_bus *bus, void *data)
 
 	dummy_page_da = dma_map_single(bus->phb->parent, dummy_page_va,
 				       PAGE_SIZE, DMA_FROM_DEVICE);
-	if (dma_mapping_error(dummy_page_da)) {
+	if (dma_mapping_error(bus->phb->parent, dummy_page_da)) {
 		pr_err("PCIEX:Map dummy page failed.\n");
 		kfree(dummy_page_va);
 		return -1;
diff --git a/arch/powerpc/platforms/cell/spider-pci.c b/arch/powerpc/platforms/cell/spider-pci.c
index 418b605..5122ec1 100644
--- a/arch/powerpc/platforms/cell/spider-pci.c
+++ b/arch/powerpc/platforms/cell/spider-pci.c
@@ -111,7 +111,7 @@ static int __init spiderpci_pci_setup_chip(struct pci_controller *phb,
 
 	dummy_page_da = dma_map_single(phb->parent, dummy_page_va,
 				       PAGE_SIZE, DMA_FROM_DEVICE);
-	if (dma_mapping_error(dummy_page_da)) {
+	if (dma_mapping_error(phb->parent, dummy_page_da)) {
 		pr_err("SPIDER-IOWA:Map dummy page filed.\n");
 		kfree(dummy_page_va);
 		return -1;
diff --git a/arch/powerpc/platforms/iseries/mf.c b/arch/powerpc/platforms/iseries/mf.c
index 1dc7295..731d7b1 100644
--- a/arch/powerpc/platforms/iseries/mf.c
+++ b/arch/powerpc/platforms/iseries/mf.c
@@ -871,7 +871,7 @@ static int proc_mf_dump_cmdline(char *page, char **start, off_t off,
 		count = 256 - off;
 
 	dma_addr = iseries_hv_map(page, off + count, DMA_FROM_DEVICE);
-	if (dma_mapping_error(dma_addr))
+	if (dma_mapping_error(NULL, dma_addr))
 		return -ENOMEM;
 	memset(page, 0, off + count);
 	memset(&vsp_cmd, 0, sizeof(vsp_cmd));
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 6371898..020cadd 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -620,7 +620,6 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
 extern int agp_amd64_init(void);
 
 static const struct dma_mapping_ops gart_dma_ops = {
-	.mapping_error			= NULL,
 	.map_single			= gart_map_single,
 	.map_simple			= gart_map_simple,
 	.unmap_single			= gart_unmap_single,
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index aec43d5..05d70b8 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -73,7 +73,7 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg,
 }
 
 /* Make sure we keep the same behaviour */
-static int nommu_mapping_error(dma_addr_t dma_addr)
+static int nommu_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
 {
 #ifdef CONFIG_X86_32
 	return 0;
diff --git a/drivers/firewire/fw-iso.c b/drivers/firewire/fw-iso.c
index bcbe794..e14c03d 100644
--- a/drivers/firewire/fw-iso.c
+++ b/drivers/firewire/fw-iso.c
@@ -50,7 +50,7 @@ fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
 
 		address = dma_map_page(card->device, buffer->pages[i],
 				       0, PAGE_SIZE, direction);
-		if (dma_mapping_error(address)) {
+		if (dma_mapping_error(card->device, address)) {
 			__free_page(buffer->pages[i]);
 			goto out_pages;
 		}
diff --git a/drivers/firewire/fw-ohci.c b/drivers/firewire/fw-ohci.c
index 4f02c55..b63f380 100644
--- a/drivers/firewire/fw-ohci.c
+++ b/drivers/firewire/fw-ohci.c
@@ -953,7 +953,7 @@ at_context_queue_packet(struct context *ctx, struct fw_packet *packet)
 		payload_bus =
 			dma_map_single(ohci->card.device, packet->payload,
 				       packet->payload_length, DMA_TO_DEVICE);
-		if (dma_mapping_error(payload_bus)) {
+		if (dma_mapping_error(ohci->card.device, payload_bus)) {
 			packet->ack = RCODE_SEND_ERROR;
 			return -1;
 		}
diff --git a/drivers/firewire/fw-sbp2.c b/drivers/firewire/fw-sbp2.c
index b2458bb..1abc567 100644
--- a/drivers/firewire/fw-sbp2.c
+++ b/drivers/firewire/fw-sbp2.c
@@ -528,7 +528,7 @@ sbp2_send_management_orb(struct sbp2_logical_unit *lu, int node_id,
 	orb->response_bus =
 		dma_map_single(device->card->device, &orb->response,
 			       sizeof(orb->response), DMA_FROM_DEVICE);
-	if (dma_mapping_error(orb->response_bus))
+	if (dma_mapping_error(device->card->device, orb->response_bus))
 		goto fail_mapping_response;
 
 	orb->request.response.high = 0;
@@ -562,7 +562,7 @@ sbp2_send_management_orb(struct sbp2_logical_unit *lu, int node_id,
 	orb->base.request_bus =
 		dma_map_single(device->card->device, &orb->request,
 			       sizeof(orb->request), DMA_TO_DEVICE);
-	if (dma_mapping_error(orb->base.request_bus))
+	if (dma_mapping_error(device->card->device, orb->base.request_bus))
 		goto fail_mapping_request;
 
 	sbp2_send_orb(&orb->base, lu, node_id, generation,
@@ -1408,7 +1408,7 @@ sbp2_map_scatterlist(struct sbp2_command_orb *orb, struct fw_device *device,
 	orb->page_table_bus =
 		dma_map_single(device->card->device, orb->page_table,
 			       sizeof(orb->page_table), DMA_TO_DEVICE);
-	if (dma_mapping_error(orb->page_table_bus))
+	if (dma_mapping_error(device->card->device, orb->page_table_bus))
 		goto fail_page_table;
 
 	/*
@@ -1493,7 +1493,7 @@ static int sbp2_scsi_queuecommand(struct scsi_cmnd *cmd, scsi_done_fn_t done)
 	orb->base.request_bus =
 		dma_map_single(device->card->device, &orb->request,
 			       sizeof(orb->request), DMA_TO_DEVICE);
-	if (dma_mapping_error(orb->base.request_bus))
+	if (dma_mapping_error(device->card->device, orb->base.request_bus))
 		goto out;
 
 	sbp2_send_orb(&orb->base, lu, lu->tgt->node_id, lu->generation,
diff --git a/drivers/infiniband/hw/ipath/ipath_sdma.c b/drivers/infiniband/hw/ipath/ipath_sdma.c
index 3697449..1ab6258 100644
--- a/drivers/infiniband/hw/ipath/ipath_sdma.c
+++ b/drivers/infiniband/hw/ipath/ipath_sdma.c
@@ -702,7 +702,7 @@ retry:
 
 	addr = dma_map_single(&dd->pcidev->dev, tx->txreq.map_addr,
 			      tx->map_len, DMA_TO_DEVICE);
-	if (dma_mapping_error(addr)) {
+	if (dma_mapping_error(&dd->pcidev->dev, addr)) {
 		ret = -EIO;
 		goto unlock;
 	}
diff --git a/drivers/infiniband/hw/ipath/ipath_user_sdma.c b/drivers/infiniband/hw/ipath/ipath_user_sdma.c
index 86e0169..82d9a0b 100644
--- a/drivers/infiniband/hw/ipath/ipath_user_sdma.c
+++ b/drivers/infiniband/hw/ipath/ipath_user_sdma.c
@@ -206,7 +206,7 @@ static int ipath_user_sdma_coalesce(const struct ipath_devdata *dd,
 
 	dma_addr = dma_map_page(&dd->pcidev->dev, page, 0, len,
 				DMA_TO_DEVICE);
-	if (dma_mapping_error(dma_addr)) {
+	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 		ret = -ENOMEM;
 		goto free_unmap;
 	}
@@ -301,7 +301,7 @@ static int ipath_user_sdma_pin_pages(const struct ipath_devdata *dd,
 				     pages[j], 0, flen, DMA_TO_DEVICE);
 		unsigned long fofs = addr & ~PAGE_MASK;
 
-		if (dma_mapping_error(dma_addr)) {
+		if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 			ret = -ENOMEM;
 			goto done;
 		}
@@ -508,7 +508,7 @@ static int ipath_user_sdma_queue_pkts(const struct ipath_devdata *dd,
 		if (page) {
 			dma_addr = dma_map_page(&dd->pcidev->dev,
 						page, 0, len, DMA_TO_DEVICE);
-			if (dma_mapping_error(dma_addr)) {
+			if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 				ret = -ENOMEM;
 				goto free_pbc;
 			}
diff --git a/drivers/infiniband/hw/mthca/mthca_eq.c b/drivers/infiniband/hw/mthca/mthca_eq.c
index 8bde7f9..b757c53 100644
--- a/drivers/infiniband/hw/mthca/mthca_eq.c
+++ b/drivers/infiniband/hw/mthca/mthca_eq.c
@@ -782,7 +782,7 @@ int mthca_map_eq_icm(struct mthca_dev *dev, u64 icm_virt)
 		return -ENOMEM;
 	dev->eq_table.icm_dma  = pci_map_page(dev->pdev, dev->eq_table.icm_page, 0,
 					      PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	if (pci_dma_mapping_error(dev->eq_table.icm_dma)) {
+	if (pci_dma_mapping_error(dev->pdev, dev->eq_table.icm_dma)) {
 		__free_page(dev->eq_table.icm_page);
 		return -ENOMEM;
 	}
diff --git a/drivers/media/dvb/pluto2/pluto2.c b/drivers/media/dvb/pluto2/pluto2.c
index 960ed57..36df2b0 100644
--- a/drivers/media/dvb/pluto2/pluto2.c
+++ b/drivers/media/dvb/pluto2/pluto2.c
@@ -242,7 +242,7 @@ static int __devinit pluto_dma_map(struct pluto *pluto)
 	pluto->dma_addr = pci_map_single(pluto->pdev, pluto->dma_buf,
 			TS_DMA_BYTES, PCI_DMA_FROMDEVICE);
 
-	return pci_dma_mapping_error(pluto->dma_addr);
+	return pci_dma_mapping_error(pluto->pdev, pluto->dma_addr);
 }
 
 static void pluto_dma_unmap(struct pluto *pluto)
diff --git a/drivers/net/arm/ep93xx_eth.c b/drivers/net/arm/ep93xx_eth.c
index ecd8fc6..09a7c7b 100644
--- a/drivers/net/arm/ep93xx_eth.c
+++ b/drivers/net/arm/ep93xx_eth.c
@@ -482,7 +482,7 @@ static int ep93xx_alloc_buffers(struct ep93xx_priv *ep)
 			goto err;
 
 		d = dma_map_single(NULL, page, PAGE_SIZE, DMA_FROM_DEVICE);
-		if (dma_mapping_error(d)) {
+		if (dma_mapping_error(NULL, d)) {
 			free_page((unsigned long)page);
 			goto err;
 		}
@@ -505,7 +505,7 @@ static int ep93xx_alloc_buffers(struct ep93xx_priv *ep)
 			goto err;
 
 		d = dma_map_single(NULL, page, PAGE_SIZE, DMA_TO_DEVICE);
-		if (dma_mapping_error(d)) {
+		if (dma_mapping_error(NULL, d)) {
 			free_page((unsigned long)page);
 			goto err;
 		}
diff --git a/drivers/net/b44.c b/drivers/net/b44.c
index 59dce6a..2b51034 100644
--- a/drivers/net/b44.c
+++ b/drivers/net/b44.c
@@ -659,10 +659,10 @@ static int b44_alloc_rx_skb(struct b44 *bp, int src_idx, u32 dest_idx_unmasked)
 
 	/* Hardware bug work-around, the chip is unable to do PCI DMA
 	   to/from anything above 1GB :-( */
-	if (dma_mapping_error(mapping) ||
+	if (dma_mapping_error(bp->sdev->dma_dev, mapping) ||
 		mapping + RX_PKT_BUF_SZ > DMA_30BIT_MASK) {
 		/* Sigh... */
-		if (!dma_mapping_error(mapping))
+		if (!dma_mapping_error(bp->sdev->dma_dev, mapping))
 			dma_unmap_single(bp->sdev->dma_dev, mapping,
 					RX_PKT_BUF_SZ, DMA_FROM_DEVICE);
 		dev_kfree_skb_any(skb);
@@ -672,9 +672,9 @@ static int b44_alloc_rx_skb(struct b44 *bp, int src_idx, u32 dest_idx_unmasked)
 		mapping = dma_map_single(bp->sdev->dma_dev, skb->data,
 					 RX_PKT_BUF_SZ,
 					 DMA_FROM_DEVICE);
-		if (dma_mapping_error(mapping) ||
+		if (dma_mapping_error(bp->sdev->dma_dev, mapping) ||
 			mapping + RX_PKT_BUF_SZ > DMA_30BIT_MASK) {
-			if (!dma_mapping_error(mapping))
+			if (!dma_mapping_error(bp->sdev->dma_dev, mapping))
 				dma_unmap_single(bp->sdev->dma_dev, mapping, RX_PKT_BUF_SZ,DMA_FROM_DEVICE);
 			dev_kfree_skb_any(skb);
 			return -ENOMEM;
@@ -967,11 +967,12 @@ static int b44_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 
 	mapping = dma_map_single(bp->sdev->dma_dev, skb->data, len, DMA_TO_DEVICE);
-	if (dma_mapping_error(mapping) || mapping + len > DMA_30BIT_MASK) {
+	if (dma_mapping_error(bp->sdev->dma_dev, mapping) ||
+	    mapping + len > DMA_30BIT_MASK) {
 		struct sk_buff *bounce_skb;
 
 		/* Chip can't handle DMA to/from >1GB, use bounce buffer */
-		if (!dma_mapping_error(mapping))
+		if (!dma_mapping_error(bp->sdev->dma_dev, mapping))
 			dma_unmap_single(bp->sdev->dma_dev, mapping, len,
 					DMA_TO_DEVICE);
 
@@ -981,8 +982,9 @@ static int b44_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 		mapping = dma_map_single(bp->sdev->dma_dev, bounce_skb->data,
 					 len, DMA_TO_DEVICE);
-		if (dma_mapping_error(mapping) || mapping + len > DMA_30BIT_MASK) {
-			if (!dma_mapping_error(mapping))
+		if (dma_mapping_error(bp->sdev->dma_dev, mapping) ||
+		    mapping + len > DMA_30BIT_MASK) {
+			if (!dma_mapping_error(bp->sdev->dma_dev, mapping))
 				dma_unmap_single(bp->sdev->dma_dev, mapping,
 					 len, DMA_TO_DEVICE);
 			dev_kfree_skb_any(bounce_skb);
@@ -1203,7 +1205,7 @@ static int b44_alloc_consistent(struct b44 *bp, gfp_t gfp)
 			                    DMA_TABLE_BYTES,
 			                    DMA_BIDIRECTIONAL);
 
-		if (dma_mapping_error(rx_ring_dma) ||
+		if (dma_mapping_error(bp->sdev->dma_dev, rx_ring_dma) ||
 			rx_ring_dma + size > DMA_30BIT_MASK) {
 			kfree(rx_ring);
 			goto out_err;
@@ -1230,7 +1232,7 @@ static int b44_alloc_consistent(struct b44 *bp, gfp_t gfp)
 			                    DMA_TABLE_BYTES,
 			                    DMA_TO_DEVICE);
 
-		if (dma_mapping_error(tx_ring_dma) ||
+		if (dma_mapping_error(bp->sdev->dma_dev, tx_ring_dma) ||
 			tx_ring_dma + size > DMA_30BIT_MASK) {
 			kfree(tx_ring);
 			goto out_err;
diff --git a/drivers/net/bnx2x.c b/drivers/net/bnx2x.c
index 7bdb5af..f92eb9d 100644
--- a/drivers/net/bnx2x.c
+++ b/drivers/net/bnx2x.c
@@ -833,7 +833,7 @@ static inline int bnx2x_alloc_rx_skb(struct bnx2x *bp,
 
 	mapping = pci_map_single(bp->pdev, skb->data, bp->rx_buf_use_size,
 				 PCI_DMA_FROMDEVICE);
-	if (unlikely(dma_mapping_error(mapping))) {
+	if (unlikely(dma_mapping_error(bp->pdev, mapping))) {
 
 		dev_kfree_skb(skb);
 		return -ENOMEM;
diff --git a/drivers/net/e100.c b/drivers/net/e100.c
index f3cba5e..7f38cb3 100644
--- a/drivers/net/e100.c
+++ b/drivers/net/e100.c
@@ -1790,7 +1790,7 @@ static int e100_rx_alloc_skb(struct nic *nic, struct rx *rx)
 	rx->dma_addr = pci_map_single(nic->pdev, rx->skb->data,
 		RFD_BUF_LEN, PCI_DMA_BIDIRECTIONAL);
 
-	if (pci_dma_mapping_error(rx->dma_addr)) {
+	if (pci_dma_mapping_error(nic->pdev, rx->dma_addr)) {
 		dev_kfree_skb_any(rx->skb);
 		rx->skb = NULL;
 		rx->dma_addr = 0;
diff --git a/drivers/net/e1000e/ethtool.c b/drivers/net/e1000e/ethtool.c
index a14561f..9350564 100644
--- a/drivers/net/e1000e/ethtool.c
+++ b/drivers/net/e1000e/ethtool.c
@@ -1090,7 +1090,7 @@ static int e1000_setup_desc_rings(struct e1000_adapter *adapter)
 		tx_ring->buffer_info[i].dma =
 			pci_map_single(pdev, skb->data, skb->len,
 				       PCI_DMA_TODEVICE);
-		if (pci_dma_mapping_error(tx_ring->buffer_info[i].dma)) {
+		if (pci_dma_mapping_error(pdev, tx_ring->buffer_info[i].dma)) {
 			ret_val = 4;
 			goto err_nomem;
 		}
@@ -1153,7 +1153,7 @@ static int e1000_setup_desc_rings(struct e1000_adapter *adapter)
 		rx_ring->buffer_info[i].dma =
 			pci_map_single(pdev, skb->data, 2048,
 				       PCI_DMA_FROMDEVICE);
-		if (pci_dma_mapping_error(rx_ring->buffer_info[i].dma)) {
+		if (pci_dma_mapping_error(pdev, rx_ring->buffer_info[i].dma)) {
 			ret_val = 8;
 			goto err_nomem;
 		}
diff --git a/drivers/net/e1000e/netdev.c b/drivers/net/e1000e/netdev.c
index 8cbb40f..41cf3fb 100644
--- a/drivers/net/e1000e/netdev.c
+++ b/drivers/net/e1000e/netdev.c
@@ -196,7 +196,7 @@ map_skb:
 		buffer_info->dma = pci_map_single(pdev, skb->data,
 						  adapter->rx_buffer_len,
 						  PCI_DMA_FROMDEVICE);
-		if (pci_dma_mapping_error(buffer_info->dma)) {
+		if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 			dev_err(&pdev->dev, "RX DMA map failed\n");
 			adapter->rx_dma_failed++;
 			break;
@@ -266,7 +266,7 @@ static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 						   ps_page->page,
 						   0, PAGE_SIZE,
 						   PCI_DMA_FROMDEVICE);
-				if (pci_dma_mapping_error(ps_page->dma)) {
+				if (pci_dma_mapping_error(pdev, ps_page->dma)) {
 					dev_err(&adapter->pdev->dev,
 					  "RX DMA page map failed\n");
 					adapter->rx_dma_failed++;
@@ -301,7 +301,7 @@ static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 		buffer_info->dma = pci_map_single(pdev, skb->data,
 						  adapter->rx_ps_bsize0,
 						  PCI_DMA_FROMDEVICE);
-		if (pci_dma_mapping_error(buffer_info->dma)) {
+		if (pci_dma_mapping_error(pdev, buffer_info->dma)) {
 			dev_err(&pdev->dev, "RX DMA map failed\n");
 			adapter->rx_dma_failed++;
 			/* cleanup skb */
@@ -3342,7 +3342,7 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 				skb->data + offset,
 				size,
 				PCI_DMA_TODEVICE);
-		if (pci_dma_mapping_error(buffer_info->dma)) {
+		if (pci_dma_mapping_error(adapter->pdev, buffer_info->dma)) {
 			dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
 			adapter->tx_dma_failed++;
 			return -1;
@@ -3380,7 +3380,8 @@ static int e1000_tx_map(struct e1000_adapter *adapter,
 					offset,
 					size,
 					PCI_DMA_TODEVICE);
-			if (pci_dma_mapping_error(buffer_info->dma)) {
+			if (pci_dma_mapping_error(adapter->pdev,
+						  buffer_info->dma)) {
 				dev_err(&adapter->pdev->dev,
 					"TX DMA page map failed\n");
 				adapter->tx_dma_failed++;
diff --git a/drivers/net/ibmveth.c b/drivers/net/ibmveth.c
index 0052780..e43d325 100644
--- a/drivers/net/ibmveth.c
+++ b/drivers/net/ibmveth.c
@@ -433,11 +433,11 @@ static void ibmveth_rxq_harvest_buffer(struct ibmveth_adapter *adapter)
 static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 {
 	int i;
+	struct device *dev = &adapter->vdev->dev;
 
 	if(adapter->buffer_list_addr != NULL) {
-		if(!dma_mapping_error(adapter->buffer_list_dma)) {
-			dma_unmap_single(&adapter->vdev->dev,
-					adapter->buffer_list_dma, 4096,
+		if (!dma_mapping_error(dev, adapter->buffer_list_dma)) {
+			dma_unmap_single(dev, adapter->buffer_list_dma, 4096,
 					DMA_BIDIRECTIONAL);
 			adapter->buffer_list_dma = DMA_ERROR_CODE;
 		}
@@ -446,9 +446,8 @@ static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 	}
 
 	if(adapter->filter_list_addr != NULL) {
-		if(!dma_mapping_error(adapter->filter_list_dma)) {
-			dma_unmap_single(&adapter->vdev->dev,
-					adapter->filter_list_dma, 4096,
+		if (!dma_mapping_error(dev, adapter->filter_list_dma)) {
+			dma_unmap_single(dev, adapter->filter_list_dma, 4096,
 					DMA_BIDIRECTIONAL);
 			adapter->filter_list_dma = DMA_ERROR_CODE;
 		}
@@ -457,8 +456,8 @@ static void ibmveth_cleanup(struct ibmveth_adapter *adapter)
 	}
 
 	if(adapter->rx_queue.queue_addr != NULL) {
-		if(!dma_mapping_error(adapter->rx_queue.queue_dma)) {
-			dma_unmap_single(&adapter->vdev->dev,
+		if (!dma_mapping_error(dev, adapter->rx_queue.queue_dma)) {
+			dma_unmap_single(dev,
 					adapter->rx_queue.queue_dma,
 					adapter->rx_queue.queue_len,
 					DMA_BIDIRECTIONAL);
@@ -508,6 +507,7 @@ static int ibmveth_open(struct net_device *netdev)
 	int rc;
 	union ibmveth_buf_desc rxq_desc;
 	int i;
+	struct device *dev;
 
 	ibmveth_debug_printk("open starting\n");
 
@@ -536,17 +536,19 @@ static int ibmveth_open(struct net_device *netdev)
 		return -ENOMEM;
 	}
 
-	adapter->buffer_list_dma = dma_map_single(&adapter->vdev->dev,
+	dev = &adapter->vdev->dev;
+
+	adapter->buffer_list_dma = dma_map_single(dev,
 			adapter->buffer_list_addr, 4096, DMA_BIDIRECTIONAL);
-	adapter->filter_list_dma = dma_map_single(&adapter->vdev->dev,
+	adapter->filter_list_dma = dma_map_single(dev,
 			adapter->filter_list_addr, 4096, DMA_BIDIRECTIONAL);
-	adapter->rx_queue.queue_dma = dma_map_single(&adapter->vdev->dev,
+	adapter->rx_queue.queue_dma = dma_map_single(dev,
 			adapter->rx_queue.queue_addr,
 			adapter->rx_queue.queue_len, DMA_BIDIRECTIONAL);
 
-	if((dma_mapping_error(adapter->buffer_list_dma) ) ||
-	   (dma_mapping_error(adapter->filter_list_dma)) ||
-	   (dma_mapping_error(adapter->rx_queue.queue_dma))) {
+	if ((dma_mapping_error(dev, adapter->buffer_list_dma)) ||
+	    (dma_mapping_error(dev, adapter->filter_list_dma)) ||
+	    (dma_mapping_error(dev, adapter->rx_queue.queue_dma))) {
 		ibmveth_error_printk("unable to map filter or buffer list pages\n");
 		ibmveth_cleanup(adapter);
 		napi_disable(&adapter->napi);
@@ -875,7 +877,7 @@ static int ibmveth_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 		buf[1] = 0;
 	}
 
-	if (dma_mapping_error(desc.fields.address)) {
+	if (dma_mapping_error(&adapter->vdev->dev, desc.fields.address)) {
 		ibmveth_error_printk("tx: unable to map xmit buffer\n");
 		tx_map_failed++;
 		tx_dropped++;
diff --git a/drivers/net/iseries_veth.c b/drivers/net/iseries_veth.c
index b8d0639..c46864d 100644
--- a/drivers/net/iseries_veth.c
+++ b/drivers/net/iseries_veth.c
@@ -1128,7 +1128,7 @@ static int veth_transmit_to_one(struct sk_buff *skb, HvLpIndex rlp,
 	msg->data.addr[0] = dma_map_single(port->dev, skb->data,
 				skb->len, DMA_TO_DEVICE);
 
-	if (dma_mapping_error(msg->data.addr[0]))
+	if (dma_mapping_error(port->dev, msg->data.addr[0]))
 		goto recycle_and_drop;
 
 	msg->dev = port->dev;
@@ -1226,7 +1226,7 @@ static void veth_recycle_msg(struct veth_lpar_connection *cnx,
 		dma_address = msg->data.addr[0];
 		dma_length = msg->data.len[0];
 
-		if (!dma_mapping_error(dma_address))
+		if (!dma_mapping_error(msg->dev, dma_address))
 			dma_unmap_single(msg->dev, dma_address, dma_length,
 					DMA_TO_DEVICE);
 
diff --git a/drivers/net/mlx4/eq.c b/drivers/net/mlx4/eq.c
index e141a15..4ca78aa 100644
--- a/drivers/net/mlx4/eq.c
+++ b/drivers/net/mlx4/eq.c
@@ -525,7 +525,7 @@ int mlx4_map_eq_icm(struct mlx4_dev *dev, u64 icm_virt)
 		return -ENOMEM;
 	priv->eq_table.icm_dma  = pci_map_page(dev->pdev, priv->eq_table.icm_page, 0,
 					       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	if (pci_dma_mapping_error(priv->eq_table.icm_dma)) {
+	if (pci_dma_mapping_error(dev->pdev, priv->eq_table.icm_dma)) {
 		__free_page(priv->eq_table.icm_page);
 		return -ENOMEM;
 	}
diff --git a/drivers/net/pasemi_mac.c b/drivers/net/pasemi_mac.c
index 3b2a6c5..9e40f23 100644
--- a/drivers/net/pasemi_mac.c
+++ b/drivers/net/pasemi_mac.c
@@ -650,7 +650,7 @@ static void pasemi_mac_replenish_rx_ring(const struct net_device *dev,
 				     mac->bufsz - LOCAL_SKB_ALIGN,
 				     PCI_DMA_FROMDEVICE);
 
-		if (unlikely(dma_mapping_error(dma))) {
+		if (unlikely(mac->dma_pdev, dma_mapping_error(dma))) {
 			dev_kfree_skb_irq(info->skb);
 			break;
 		}
@@ -1519,7 +1519,7 @@ static int pasemi_mac_start_tx(struct sk_buff *skb, struct net_device *dev)
 	map[0] = pci_map_single(mac->dma_pdev, skb->data, skb_headlen(skb),
 				PCI_DMA_TODEVICE);
 	map_size[0] = skb_headlen(skb);
-	if (dma_mapping_error(map[0]))
+	if (dma_mapping_error(mac->dma_pdev, map[0]))
 		goto out_err_nolock;
 
 	for (i = 0; i < nfrags; i++) {
@@ -1529,7 +1529,7 @@ static int pasemi_mac_start_tx(struct sk_buff *skb, struct net_device *dev)
 					frag->page_offset, frag->size,
 					PCI_DMA_TODEVICE);
 		map_size[i+1] = frag->size;
-		if (dma_mapping_error(map[i+1])) {
+		if (dma_mapping_error(mac->dma_pdev, map[i+1])) {
 			nfrags = i;
 			goto out_err_nolock;
 		}
diff --git a/drivers/net/qla3xxx.c b/drivers/net/qla3xxx.c
index b7f7b22..d585753 100644
--- a/drivers/net/qla3xxx.c
+++ b/drivers/net/qla3xxx.c
@@ -328,7 +328,7 @@ static void ql_release_to_lrg_buf_free_list(struct ql3_adapter *qdev,
 					     qdev->lrg_buffer_len -
 					     QL_HEADER_SPACE,
 					     PCI_DMA_FROMDEVICE);
-			err = pci_dma_mapping_error(map);
+			err = pci_dma_mapping_error(qdev->pdev, map);
 			if(err) {
 				printk(KERN_ERR "%s: PCI mapping failed with error: %d\n",
 				       qdev->ndev->name, err);
@@ -1919,7 +1919,7 @@ static int ql_populate_free_queue(struct ql3_adapter *qdev)
 						     QL_HEADER_SPACE,
 						     PCI_DMA_FROMDEVICE);
 
-				err = pci_dma_mapping_error(map);
+				err = pci_dma_mapping_error(qdev->pdev, map);
 				if(err) {
 					printk(KERN_ERR "%s: PCI mapping failed with error: %d\n",
 					       qdev->ndev->name, err);
@@ -2454,7 +2454,7 @@ static int ql_send_map(struct ql3_adapter *qdev,
 	 */
 	map = pci_map_single(qdev->pdev, skb->data, len, PCI_DMA_TODEVICE);
 
-	err = pci_dma_mapping_error(map);
+	err = pci_dma_mapping_error(qdev->pdev, map);
 	if(err) {
 		printk(KERN_ERR "%s: PCI mapping failed with error: %d\n",
 		       qdev->ndev->name, err);
@@ -2487,7 +2487,7 @@ static int ql_send_map(struct ql3_adapter *qdev,
 						     sizeof(struct oal),
 						     PCI_DMA_TODEVICE);
 
-				err = pci_dma_mapping_error(map);
+				err = pci_dma_mapping_error(qdev->pdev, map);
 				if(err) {
 
 					printk(KERN_ERR "%s: PCI mapping outbound address list with error: %d\n",
@@ -2514,7 +2514,7 @@ static int ql_send_map(struct ql3_adapter *qdev,
 					 frag->page_offset, frag->size,
 					 PCI_DMA_TODEVICE);
 
-			err = pci_dma_mapping_error(map);
+			err = pci_dma_mapping_error(qdev->pdev, map);
 			if(err) {
 				printk(KERN_ERR "%s: PCI mapping frags failed with error: %d\n",
 				       qdev->ndev->name, err);
@@ -2916,7 +2916,7 @@ static int ql_alloc_large_buffers(struct ql3_adapter *qdev)
 					     QL_HEADER_SPACE,
 					     PCI_DMA_FROMDEVICE);
 
-			err = pci_dma_mapping_error(map);
+			err = pci_dma_mapping_error(qdev->pdev, map);
 			if(err) {
 				printk(KERN_ERR "%s: PCI mapping failed with error: %d\n",
 				       qdev->ndev->name, err);
diff --git a/drivers/net/sfc/rx.c b/drivers/net/sfc/rx.c
index 6706223..a1bc380 100644
--- a/drivers/net/sfc/rx.c
+++ b/drivers/net/sfc/rx.c
@@ -230,7 +230,7 @@ static inline int efx_init_rx_buffer_skb(struct efx_rx_queue *rx_queue,
 					  rx_buf->data, rx_buf->len,
 					  PCI_DMA_FROMDEVICE);
 
-	if (unlikely(pci_dma_mapping_error(rx_buf->dma_addr))) {
+	if (unlikely(pci_dma_mapping_error(efx->pci_dev, rx_buf->dma_addr))) {
 		dev_kfree_skb_any(rx_buf->skb);
 		rx_buf->skb = NULL;
 		return -EIO;
@@ -272,7 +272,7 @@ static inline int efx_init_rx_buffer_page(struct efx_rx_queue *rx_queue,
 					0, RX_PAGE_SIZE(efx),
 					PCI_DMA_FROMDEVICE);
 
-		if (unlikely(pci_dma_mapping_error(dma_addr))) {
+		if (unlikely(pci_dma_mapping_error(efx->pci_dev, dma_addr))) {
 			__free_pages(rx_buf->page, efx->rx_buffer_order);
 			rx_buf->page = NULL;
 			return -EIO;
diff --git a/drivers/net/sfc/tx.c b/drivers/net/sfc/tx.c
index 9b436f5..16c2d6d 100644
--- a/drivers/net/sfc/tx.c
+++ b/drivers/net/sfc/tx.c
@@ -172,7 +172,7 @@ static inline int efx_enqueue_skb(struct efx_tx_queue *tx_queue,
 
 	/* Process all fragments */
 	while (1) {
-		if (unlikely(pci_dma_mapping_error(dma_addr)))
+		if (unlikely(pci_dma_mapping_error(pci_dev, dma_addr)))
 			goto pci_err;
 
 		/* Store fields for marking in the per-fragment final
@@ -660,7 +660,8 @@ efx_tsoh_heap_alloc(struct efx_tx_queue *tx_queue, size_t header_len)
 	tsoh->dma_addr = pci_map_single(tx_queue->efx->pci_dev,
 					TSOH_BUFFER(tsoh), header_len,
 					PCI_DMA_TODEVICE);
-	if (unlikely(pci_dma_mapping_error(tsoh->dma_addr))) {
+	if (unlikely(pci_dma_mapping_error(tx_queue->efx->pci_dev,
+					   tsoh->dma_addr))) {
 		kfree(tsoh);
 		return NULL;
 	}
@@ -862,7 +863,7 @@ static inline int tso_get_fragment(struct tso_state *st, struct efx_nic *efx,
 
 	st->ifc.unmap_addr = pci_map_page(efx->pci_dev, page, page_off,
 					  len, PCI_DMA_TODEVICE);
-	if (likely(!pci_dma_mapping_error(st->ifc.unmap_addr))) {
+	if (likely(!pci_dma_mapping_error(efx->pci_dev, st->ifc.unmap_addr))) {
 		st->ifc.unmap_len = len;
 		st->ifc.len = len;
 		st->ifc.dma_addr = st->ifc.unmap_addr;
diff --git a/drivers/net/spider_net.c b/drivers/net/spider_net.c
index 4776716..fdcc4ab 100644
--- a/drivers/net/spider_net.c
+++ b/drivers/net/spider_net.c
@@ -452,7 +452,7 @@ spider_net_prepare_rx_descr(struct spider_net_card *card,
 	/* iommu-map the skb */
 	buf = pci_map_single(card->pdev, descr->skb->data,
 			SPIDER_NET_MAX_FRAME, PCI_DMA_FROMDEVICE);
-	if (pci_dma_mapping_error(buf)) {
+	if (pci_dma_mapping_error(card->pdev, buf)) {
 		dev_kfree_skb_any(descr->skb);
 		descr->skb = NULL;
 		if (netif_msg_rx_err(card) && net_ratelimit())
@@ -691,7 +691,7 @@ spider_net_prepare_tx_descr(struct spider_net_card *card,
 	unsigned long flags;
 
 	buf = pci_map_single(card->pdev, skb->data, skb->len, PCI_DMA_TODEVICE);
-	if (pci_dma_mapping_error(buf)) {
+	if (pci_dma_mapping_error(card->pdev, buf)) {
 		if (netif_msg_tx_err(card) && net_ratelimit())
 			dev_err(&card->netdev->dev, "could not iommu-map packet (%p, %i). "
 				  "Dropping packet\n", skb->data, skb->len);
diff --git a/drivers/net/tc35815.c b/drivers/net/tc35815.c
index 10e4e85..1fc9c78 100644
--- a/drivers/net/tc35815.c
+++ b/drivers/net/tc35815.c
@@ -506,7 +506,7 @@ static void *alloc_rxbuf_page(struct pci_dev *hwdev, dma_addr_t *dma_handle)
 		return NULL;
 	*dma_handle = pci_map_single(hwdev, buf, PAGE_SIZE,
 				     PCI_DMA_FROMDEVICE);
-	if (pci_dma_mapping_error(*dma_handle)) {
+	if (pci_dma_mapping_error(hwdev, *dma_handle)) {
 		free_page((unsigned long)buf);
 		return NULL;
 	}
@@ -536,7 +536,7 @@ static struct sk_buff *alloc_rxbuf_skb(struct net_device *dev,
 		return NULL;
 	*dma_handle = pci_map_single(hwdev, skb->data, RX_BUF_SIZE,
 				     PCI_DMA_FROMDEVICE);
-	if (pci_dma_mapping_error(*dma_handle)) {
+	if (pci_dma_mapping_error(hwdev, *dma_handle)) {
 		dev_kfree_skb_any(skb);
 		return NULL;
 	}
diff --git a/drivers/net/wireless/ath5k/base.c b/drivers/net/wireless/ath5k/base.c
index 3201c16..daa2e57 100644
--- a/drivers/net/wireless/ath5k/base.c
+++ b/drivers/net/wireless/ath5k/base.c
@@ -1260,7 +1260,7 @@ ath5k_rxbuf_setup(struct ath5k_softc *sc, struct ath5k_buf *bf)
 		bf->skb = skb;
 		bf->skbaddr = pci_map_single(sc->pdev,
 			skb->data, sc->rxbufsize, PCI_DMA_FROMDEVICE);
-		if (unlikely(pci_dma_mapping_error(bf->skbaddr))) {
+		if (unlikely(pci_dma_mapping_error(sc->pdev, bf->skbaddr))) {
 			ATH5K_ERR(sc, "%s: DMA mapping failed\n", __func__);
 			dev_kfree_skb(skb);
 			bf->skb = NULL;
@@ -2029,7 +2029,7 @@ ath5k_beacon_setup(struct ath5k_softc *sc, struct ath5k_buf *bf,
 	ATH5K_DBG(sc, ATH5K_DEBUG_BEACON, "skb %p [data %p len %u] "
 			"skbaddr %llx\n", skb, skb->data, skb->len,
 			(unsigned long long)bf->skbaddr);
-	if (pci_dma_mapping_error(bf->skbaddr)) {
+	if (pci_dma_mapping_error(sc->pdev, bf->skbaddr)) {
 		ATH5K_ERR(sc, "beacon DMA mapping failed\n");
 		return -EIO;
 	}
diff --git a/drivers/net/wireless/b43/dma.c b/drivers/net/wireless/b43/dma.c
index f50e201..364840b 100644
--- a/drivers/net/wireless/b43/dma.c
+++ b/drivers/net/wireless/b43/dma.c
@@ -518,7 +518,7 @@ static bool b43_dma_mapping_error(struct b43_dmaring *ring,
 				  dma_addr_t addr,
 				  size_t buffersize, bool dma_to_device)
 {
-	if (unlikely(dma_mapping_error(addr)))
+	if (unlikely(dma_mapping_error(ring->dev->dev->dma_dev, addr)))
 		return 1;
 
 	switch (ring->type) {
diff --git a/drivers/net/wireless/b43legacy/dma.c b/drivers/net/wireless/b43legacy/dma.c
index d6686f7..7ccc149 100644
--- a/drivers/net/wireless/b43legacy/dma.c
+++ b/drivers/net/wireless/b43legacy/dma.c
@@ -589,7 +589,7 @@ static bool b43legacy_dma_mapping_error(struct b43legacy_dmaring *ring,
 					 size_t buffersize,
 					 bool dma_to_device)
 {
-	if (unlikely(dma_mapping_error(addr)))
+	if (unlikely(dma_mapping_error(ring->dev->dev->dma_dev, addr)))
 		return 1;
 
 	switch (ring->type) {
diff --git a/drivers/scsi/ibmvscsi/ibmvscsi.c b/drivers/scsi/ibmvscsi/ibmvscsi.c
index ccfd8ac..b68423e 100644
--- a/drivers/scsi/ibmvscsi/ibmvscsi.c
+++ b/drivers/scsi/ibmvscsi/ibmvscsi.c
@@ -854,7 +854,7 @@ static void send_mad_adapter_info(struct ibmvscsi_host_data *hostdata)
 					    sizeof(hostdata->madapter_info),
 					    DMA_BIDIRECTIONAL);
 
-	if (dma_mapping_error(req->buffer)) {
+	if (dma_mapping_error(hostdata->dev, req->buffer)) {
 		dev_err(hostdata->dev, "Unable to map request_buffer for adapter_info!\n");
 		free_event_struct(&hostdata->pool, evt_struct);
 		return;
@@ -1399,7 +1399,7 @@ static int ibmvscsi_do_host_config(struct ibmvscsi_host_data *hostdata,
 						    length,
 						    DMA_BIDIRECTIONAL);
 
-	if (dma_mapping_error(host_config->buffer)) {
+	if (dma_mapping_error(hostdata->dev, host_config->buffer)) {
 		dev_err(hostdata->dev, "dma_mapping error getting host config\n");
 		free_event_struct(&hostdata->pool, evt_struct);
 		return -1;
diff --git a/drivers/scsi/ibmvscsi/ibmvstgt.c b/drivers/scsi/ibmvscsi/ibmvstgt.c
index d6f68f2..2a5b29d 100644
--- a/drivers/scsi/ibmvscsi/ibmvstgt.c
+++ b/drivers/scsi/ibmvscsi/ibmvstgt.c
@@ -564,7 +564,7 @@ static int crq_queue_create(struct crq_queue *queue, struct srp_target *target)
 					  queue->size * sizeof(*queue->msgs),
 					  DMA_BIDIRECTIONAL);
 
-	if (dma_mapping_error(queue->msg_token))
+	if (dma_mapping_error(target->dev, queue->msg_token))
 		goto map_failed;
 
 	err = h_reg_crq(vport->dma_dev->unit_address, queue->msg_token,
diff --git a/drivers/scsi/ibmvscsi/rpa_vscsi.c b/drivers/scsi/ibmvscsi/rpa_vscsi.c
index 1821461..462a857 100644
--- a/drivers/scsi/ibmvscsi/rpa_vscsi.c
+++ b/drivers/scsi/ibmvscsi/rpa_vscsi.c
@@ -253,7 +253,7 @@ static int rpavscsi_init_crq_queue(struct crq_queue *queue,
 					  queue->size * sizeof(*queue->msgs),
 					  DMA_BIDIRECTIONAL);
 
-	if (dma_mapping_error(queue->msg_token))
+	if (dma_mapping_error(hostdata->dev, queue->msg_token))
 		goto map_failed;
 
 	gather_partition_info();
diff --git a/drivers/spi/atmel_spi.c b/drivers/spi/atmel_spi.c
index e81d59d..0c71656 100644
--- a/drivers/spi/atmel_spi.c
+++ b/drivers/spi/atmel_spi.c
@@ -313,14 +313,14 @@ atmel_spi_dma_map_xfer(struct atmel_spi *as, struct spi_transfer *xfer)
 		xfer->tx_dma = dma_map_single(dev,
 				(void *) xfer->tx_buf, xfer->len,
 				DMA_TO_DEVICE);
-		if (dma_mapping_error(xfer->tx_dma))
+		if (dma_mapping_error(dev, xfer->tx_dma))
 			return -ENOMEM;
 	}
 	if (xfer->rx_buf) {
 		xfer->rx_dma = dma_map_single(dev,
 				xfer->rx_buf, xfer->len,
 				DMA_FROM_DEVICE);
-		if (dma_mapping_error(xfer->rx_dma)) {
+		if (dma_mapping_error(dev, xfer->rx_dma)) {
 			if (xfer->tx_buf)
 				dma_unmap_single(dev,
 						xfer->tx_dma, xfer->len,
diff --git a/drivers/spi/au1550_spi.c b/drivers/spi/au1550_spi.c
index 072c4a5..183e0f3 100644
--- a/drivers/spi/au1550_spi.c
+++ b/drivers/spi/au1550_spi.c
@@ -330,7 +330,7 @@ static int au1550_spi_dma_rxtmp_alloc(struct au1550_spi *hw, unsigned size)
 	hw->dma_rx_tmpbuf_size = size;
 	hw->dma_rx_tmpbuf_addr = dma_map_single(hw->dev, hw->dma_rx_tmpbuf,
 			size, DMA_FROM_DEVICE);
-	if (dma_mapping_error(hw->dma_rx_tmpbuf_addr)) {
+	if (dma_mapping_error(hw->dev, hw->dma_rx_tmpbuf_addr)) {
 		kfree(hw->dma_rx_tmpbuf);
 		hw->dma_rx_tmpbuf = 0;
 		hw->dma_rx_tmpbuf_size = 0;
@@ -374,7 +374,7 @@ static int au1550_spi_dma_txrxb(struct spi_device *spi, struct spi_transfer *t)
 			dma_rx_addr = dma_map_single(hw->dev,
 					(void *)t->rx_buf,
 					t->len, DMA_FROM_DEVICE);
-			if (dma_mapping_error(dma_rx_addr))
+			if (dma_mapping_error(hw->dev, dma_rx_addr))
 				dev_err(hw->dev, "rx dma map error\n");
 		}
 	} else {
@@ -397,7 +397,7 @@ static int au1550_spi_dma_txrxb(struct spi_device *spi, struct spi_transfer *t)
 			dma_tx_addr = dma_map_single(hw->dev,
 					(void *)t->tx_buf,
 					t->len, DMA_TO_DEVICE);
-			if (dma_mapping_error(dma_tx_addr))
+			if (dma_mapping_error(hw->dev, dma_tx_addr))
 				dev_err(hw->dev, "tx dma map error\n");
 		}
 	} else {
diff --git a/drivers/spi/omap2_mcspi.c b/drivers/spi/omap2_mcspi.c
index b1cc148..f6f987b 100644
--- a/drivers/spi/omap2_mcspi.c
+++ b/drivers/spi/omap2_mcspi.c
@@ -836,7 +836,7 @@ static int omap2_mcspi_transfer(struct spi_device *spi, struct spi_message *m)
 		if (tx_buf != NULL) {
 			t->tx_dma = dma_map_single(&spi->dev, (void *) tx_buf,
 					len, DMA_TO_DEVICE);
-			if (dma_mapping_error(t->tx_dma)) {
+			if (dma_mapping_error(&spi->dev, t->tx_dma)) {
 				dev_dbg(&spi->dev, "dma %cX %d bytes error\n",
 						'T', len);
 				return -EINVAL;
@@ -845,7 +845,7 @@ static int omap2_mcspi_transfer(struct spi_device *spi, struct spi_message *m)
 		if (rx_buf != NULL) {
 			t->rx_dma = dma_map_single(&spi->dev, rx_buf, t->len,
 					DMA_FROM_DEVICE);
-			if (dma_mapping_error(t->rx_dma)) {
+			if (dma_mapping_error(&spi->dev, t->rx_dma)) {
 				dev_dbg(&spi->dev, "dma %cX %d bytes error\n",
 						'R', len);
 				if (tx_buf != NULL)
diff --git a/drivers/spi/pxa2xx_spi.c b/drivers/spi/pxa2xx_spi.c
index 0c452c4..067299d 100644
--- a/drivers/spi/pxa2xx_spi.c
+++ b/drivers/spi/pxa2xx_spi.c
@@ -353,7 +353,7 @@ static int map_dma_buffers(struct driver_data *drv_data)
 	drv_data->rx_dma = dma_map_single(dev, drv_data->rx,
 						drv_data->rx_map_len,
 						DMA_FROM_DEVICE);
-	if (dma_mapping_error(drv_data->rx_dma))
+	if (dma_mapping_error(dev, drv_data->rx_dma))
 		return 0;
 
 	/* Stream map the tx buffer */
@@ -361,7 +361,7 @@ static int map_dma_buffers(struct driver_data *drv_data)
 						drv_data->tx_map_len,
 						DMA_TO_DEVICE);
 
-	if (dma_mapping_error(drv_data->tx_dma)) {
+	if (dma_mapping_error(dev, drv_data->tx_dma)) {
 		dma_unmap_single(dev, drv_data->rx_dma,
 					drv_data->rx_map_len, DMA_FROM_DEVICE);
 		return 0;
diff --git a/drivers/spi/spi_imx.c b/drivers/spi/spi_imx.c
index c730d05..0712e37 100644
--- a/drivers/spi/spi_imx.c
+++ b/drivers/spi/spi_imx.c
@@ -488,7 +488,7 @@ static int map_dma_buffers(struct driver_data *drv_data)
 							buf,
 							drv_data->tx_map_len,
 							DMA_TO_DEVICE);
-			if (dma_mapping_error(drv_data->tx_dma))
+			if (dma_mapping_error(dev, drv_data->tx_dma))
 				return -1;
 
 			drv_data->tx_dma_needs_unmap = 1;
@@ -513,7 +513,7 @@ static int map_dma_buffers(struct driver_data *drv_data)
 					buf,
 					drv_data->len,
 					DMA_FROM_DEVICE);
-		if (dma_mapping_error(drv_data->rx_dma))
+		if (dma_mapping_error(dev, drv_data->rx_dma))
 			return -1;
 		drv_data->rx_dma_needs_unmap = 1;
 	}
@@ -531,7 +531,7 @@ static int map_dma_buffers(struct driver_data *drv_data)
 					buf,
 					drv_data->tx_map_len,
 					DMA_TO_DEVICE);
-	if (dma_mapping_error(drv_data->tx_dma)) {
+	if (dma_mapping_error(dev, drv_data->tx_dma)) {
 		if (drv_data->rx_dma) {
 			dma_unmap_single(dev,
 					drv_data->rx_dma,
diff --git a/include/asm-alpha/dma-mapping.h b/include/asm-alpha/dma-mapping.h
index db351d1..a5801ae 100644
--- a/include/asm-alpha/dma-mapping.h
+++ b/include/asm-alpha/dma-mapping.h
@@ -24,8 +24,8 @@
 		pci_unmap_sg(alpha_gendev_to_pci(dev), sg, nents, dir)
 #define dma_supported(dev, mask)			\
 		pci_dma_supported(alpha_gendev_to_pci(dev), mask)
-#define dma_mapping_error(addr)				\
-		pci_dma_mapping_error(addr)
+#define dma_mapping_error(dev, addr)				\
+		pci_dma_mapping_error(alpha_gendev_to_pci(dev), addr)
 
 #else	/* no PCI - no IOMMU. */
 
@@ -45,7 +45,7 @@ int dma_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 #define dma_unmap_page(dev, addr, size, dir)	((void)0)
 #define dma_unmap_sg(dev, sg, nents, dir)	((void)0)
 
-#define dma_mapping_error(addr)  (0)
+#define dma_mapping_error(dev, addr)  (0)
 
 #endif	/* !CONFIG_PCI */
 
diff --git a/include/asm-alpha/pci.h b/include/asm-alpha/pci.h
index d31fd49..2a14302 100644
--- a/include/asm-alpha/pci.h
+++ b/include/asm-alpha/pci.h
@@ -106,7 +106,7 @@ extern dma_addr_t pci_map_page(struct pci_dev *, struct page *,
 /* Test for pci_map_single or pci_map_page having generated an error.  */
 
 static inline int
-pci_dma_mapping_error(dma_addr_t dma_addr)
+pci_dma_mapping_error(struct pci_dev *pdev, dma_addr_t dma_addr)
 {
 	return dma_addr == 0;
 }
diff --git a/include/asm-arm/dma-mapping.h b/include/asm-arm/dma-mapping.h
index e99406a..f41335b 100644
--- a/include/asm-arm/dma-mapping.h
+++ b/include/asm-arm/dma-mapping.h
@@ -56,7 +56,7 @@ static inline int dma_is_consistent(struct device *dev, dma_addr_t handle)
 /*
  * DMA errors are defined by all-bits-set in the DMA address.
  */
-static inline int dma_mapping_error(dma_addr_t dma_addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return dma_addr == ~0;
 }
diff --git a/include/asm-avr32/dma-mapping.h b/include/asm-avr32/dma-mapping.h
index 57dc672..0399359 100644
--- a/include/asm-avr32/dma-mapping.h
+++ b/include/asm-avr32/dma-mapping.h
@@ -35,7 +35,7 @@ static inline int dma_set_mask(struct device *dev, u64 dma_mask)
 /*
  * dma_map_single can't fail as it is implemented now.
  */
-static inline int dma_mapping_error(dma_addr_t addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t addr)
 {
 	return 0;
 }
diff --git a/include/asm-cris/dma-mapping.h b/include/asm-cris/dma-mapping.h
index edc8d1b..cb2fb25 100644
--- a/include/asm-cris/dma-mapping.h
+++ b/include/asm-cris/dma-mapping.h
@@ -120,7 +120,7 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nelems,
 }
 
 static inline int
-dma_mapping_error(dma_addr_t dma_addr)
+dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/include/asm-frv/dma-mapping.h b/include/asm-frv/dma-mapping.h
index 2e8966c..b289887 100644
--- a/include/asm-frv/dma-mapping.h
+++ b/include/asm-frv/dma-mapping.h
@@ -126,7 +126,7 @@ void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nele
 }
 
 static inline
-int dma_mapping_error(dma_addr_t dma_addr)
+int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/include/asm-generic/dma-mapping-broken.h b/include/asm-generic/dma-mapping-broken.h
index e2468f8..82cd0cb 100644
--- a/include/asm-generic/dma-mapping-broken.h
+++ b/include/asm-generic/dma-mapping-broken.h
@@ -61,7 +61,7 @@ dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, int nelems,
 #define dma_sync_sg_for_device dma_sync_sg_for_cpu
 
 extern int
-dma_mapping_error(dma_addr_t dma_addr);
+dma_mapping_error(struct device *dev, dma_addr_t dma_addr);
 
 extern int
 dma_supported(struct device *dev, u64 mask);
diff --git a/include/asm-generic/dma-mapping.h b/include/asm-generic/dma-mapping.h
index 783ab99..189486c 100644
--- a/include/asm-generic/dma-mapping.h
+++ b/include/asm-generic/dma-mapping.h
@@ -144,9 +144,9 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nelems,
 }
 
 static inline int
-dma_mapping_error(dma_addr_t dma_addr)
+dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
-	return pci_dma_mapping_error(dma_addr);
+	return pci_dma_mapping_error(to_pci_dev(dev), dma_addr);
 }
 
 
diff --git a/include/asm-generic/pci-dma-compat.h b/include/asm-generic/pci-dma-compat.h
index 25c10e9..37b3706 100644
--- a/include/asm-generic/pci-dma-compat.h
+++ b/include/asm-generic/pci-dma-compat.h
@@ -99,9 +99,9 @@ pci_dma_sync_sg_for_device(struct pci_dev *hwdev, struct scatterlist *sg,
 }
 
 static inline int
-pci_dma_mapping_error(dma_addr_t dma_addr)
+pci_dma_mapping_error(struct pci_dev *pdev, dma_addr_t dma_addr)
 {
-	return dma_mapping_error(dma_addr);
+	return dma_mapping_error(&pdev->dev, dma_addr);
 }
 
 #endif
diff --git a/include/asm-ia64/machvec.h b/include/asm-ia64/machvec.h
index 9f020eb..50055a3 100644
--- a/include/asm-ia64/machvec.h
+++ b/include/asm-ia64/machvec.h
@@ -54,7 +54,7 @@ typedef void ia64_mv_dma_sync_single_for_cpu (struct device *, dma_addr_t, size_
 typedef void ia64_mv_dma_sync_sg_for_cpu (struct device *, struct scatterlist *, int, int);
 typedef void ia64_mv_dma_sync_single_for_device (struct device *, dma_addr_t, size_t, int);
 typedef void ia64_mv_dma_sync_sg_for_device (struct device *, struct scatterlist *, int, int);
-typedef int ia64_mv_dma_mapping_error (dma_addr_t dma_addr);
+typedef int ia64_mv_dma_mapping_error(struct device *, dma_addr_t dma_addr);
 typedef int ia64_mv_dma_supported (struct device *, u64);
 
 typedef dma_addr_t ia64_mv_dma_map_single_attrs (struct device *, void *, size_t, int, struct dma_attrs *);
diff --git a/include/asm-m68k/dma-mapping.h b/include/asm-m68k/dma-mapping.h
index a26cdeb..91f7944 100644
--- a/include/asm-m68k/dma-mapping.h
+++ b/include/asm-m68k/dma-mapping.h
@@ -84,7 +84,7 @@ static inline void dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *s
 {
 }
 
-static inline int dma_mapping_error(dma_addr_t handle)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t handle)
 {
 	return 0;
 }
diff --git a/include/asm-mips/dma-mapping.h b/include/asm-mips/dma-mapping.h
index 230b3f1..c64afb4 100644
--- a/include/asm-mips/dma-mapping.h
+++ b/include/asm-mips/dma-mapping.h
@@ -42,7 +42,7 @@ extern void dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 	int nelems, enum dma_data_direction direction);
 extern void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 	int nelems, enum dma_data_direction direction);
-extern int dma_mapping_error(dma_addr_t dma_addr);
+extern int dma_mapping_error(struct device *dev, dma_addr_t dma_addr);
 extern int dma_supported(struct device *dev, u64 mask);
 
 static inline int
diff --git a/include/asm-mn10300/dma-mapping.h b/include/asm-mn10300/dma-mapping.h
index 7c882fc..ccae8f6 100644
--- a/include/asm-mn10300/dma-mapping.h
+++ b/include/asm-mn10300/dma-mapping.h
@@ -182,7 +182,7 @@ void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 }
 
 static inline
-int dma_mapping_error(dma_addr_t dma_addr)
+int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/include/asm-parisc/dma-mapping.h b/include/asm-parisc/dma-mapping.h
index c6c0e9f..53af696 100644
--- a/include/asm-parisc/dma-mapping.h
+++ b/include/asm-parisc/dma-mapping.h
@@ -248,6 +248,6 @@ void * sba_get_iommu(struct parisc_device *dev);
 #endif
 
 /* At the moment, we panic on error for IOMMU resource exaustion */
-#define dma_mapping_error(x)	0
+#define dma_mapping_error(dev, x)	0
 
 #endif
diff --git a/include/asm-powerpc/dma-mapping.h b/include/asm-powerpc/dma-mapping.h
index bbefb69..14a93a1 100644
--- a/include/asm-powerpc/dma-mapping.h
+++ b/include/asm-powerpc/dma-mapping.h
@@ -350,7 +350,7 @@ static inline void dma_sync_sg_for_device(struct device *dev,
 		__dma_sync_page(sg_page(sg), sg->offset, sg->length, direction);
 }
 
-static inline int dma_mapping_error(dma_addr_t dma_addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 #ifdef CONFIG_PPC64
 	return (dma_addr == DMA_ERROR_CODE);
diff --git a/include/asm-sh/dma-mapping.h b/include/asm-sh/dma-mapping.h
index 22cc419..6c0b8a2 100644
--- a/include/asm-sh/dma-mapping.h
+++ b/include/asm-sh/dma-mapping.h
@@ -171,7 +171,7 @@ static inline int dma_get_cache_alignment(void)
 	return L1_CACHE_BYTES;
 }
 
-static inline int dma_mapping_error(dma_addr_t dma_addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return dma_addr == 0;
 }
diff --git a/include/asm-sparc/pci.h b/include/asm-sparc/pci.h
index b93b6c7..03e9381 100644
--- a/include/asm-sparc/pci.h
+++ b/include/asm-sparc/pci.h
@@ -154,7 +154,8 @@ static inline void pci_dma_burst_advice(struct pci_dev *pdev,
 
 #define PCI_DMA_ERROR_CODE      (~(dma_addr_t)0x0)
 
-static inline int pci_dma_mapping_error(dma_addr_t dma_addr)
+static inline int pci_dma_mapping_error(struct pci_device *pdev,
+					dma_addr_t dma_addr)
 {
         return (dma_addr == PCI_DMA_ERROR_CODE);
 }
diff --git a/include/asm-sparc64/dma-mapping.h b/include/asm-sparc64/dma-mapping.h
index 38cbec7..bfa64f9 100644
--- a/include/asm-sparc64/dma-mapping.h
+++ b/include/asm-sparc64/dma-mapping.h
@@ -135,7 +135,7 @@ static inline void dma_sync_sg_for_device(struct device *dev,
 	/* No flushing needed to sync cpu writes to the device.  */
 }
 
-static inline int dma_mapping_error(dma_addr_t dma_addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return (dma_addr == DMA_ERROR_CODE);
 }
diff --git a/include/asm-sparc64/pci.h b/include/asm-sparc64/pci.h
index f59f257..4f79a54 100644
--- a/include/asm-sparc64/pci.h
+++ b/include/asm-sparc64/pci.h
@@ -140,9 +140,10 @@ extern int pci_dma_supported(struct pci_dev *hwdev, u64 mask);
 #define PCI64_REQUIRED_MASK	(~(dma64_addr_t)0)
 #define PCI64_ADDR_BASE		0xfffc000000000000UL
 
-static inline int pci_dma_mapping_error(dma_addr_t dma_addr)
+static inline int pci_dma_mapping_error(struct pci_dev *pdev,
+					dma_addr_t dma_addr)
 {
-	return dma_mapping_error(dma_addr);
+	return dma_mapping_error(&pdev->dev, dma_addr);
 }
 
 #ifdef CONFIG_PCI
diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h
index a1a4dc7..69626b6 100644
--- a/include/asm-x86/dma-mapping.h
+++ b/include/asm-x86/dma-mapping.h
@@ -18,7 +18,8 @@ extern int forbid_dac;
 extern int force_iommu;
 
 struct dma_mapping_ops {
-	int             (*mapping_error)(dma_addr_t dma_addr);
+	int             (*mapping_error)(struct device *dev,
+					 dma_addr_t dma_addr);
 	void*           (*alloc_coherent)(struct device *dev, size_t size,
 				dma_addr_t *dma_handle, gfp_t gfp);
 	void            (*free_coherent)(struct device *dev, size_t size,
@@ -59,10 +60,10 @@ struct dma_mapping_ops {
 
 extern const struct dma_mapping_ops *dma_ops;
 
-static inline int dma_mapping_error(dma_addr_t dma_addr)
+static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	if (dma_ops->mapping_error)
-		return dma_ops->mapping_error(dma_addr);
+		return dma_ops->mapping_error(dev, dma_addr);
 
 	return (dma_addr == bad_dma_address);
 }
diff --git a/include/asm-x86/swiotlb.h b/include/asm-x86/swiotlb.h
index f5d9e74..746eeab 100644
--- a/include/asm-x86/swiotlb.h
+++ b/include/asm-x86/swiotlb.h
@@ -35,7 +35,7 @@ extern int swiotlb_map_sg(struct device *hwdev, struct scatterlist *sg,
 			  int nents, int direction);
 extern void swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sg,
 			     int nents, int direction);
-extern int swiotlb_dma_mapping_error(dma_addr_t dma_addr);
+extern int swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr);
 extern void swiotlb_free_coherent(struct device *hwdev, size_t size,
 				  void *vaddr, dma_addr_t dma_handle);
 extern int swiotlb_dma_supported(struct device *hwdev, u64 mask);
diff --git a/include/asm-xtensa/dma-mapping.h b/include/asm-xtensa/dma-mapping.h
index 3c7d537..51882ae 100644
--- a/include/asm-xtensa/dma-mapping.h
+++ b/include/asm-xtensa/dma-mapping.h
@@ -139,7 +139,7 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, int nelems,
 		consistent_sync(sg_virt(sg), sg->length, dir);
 }
 static inline int
-dma_mapping_error(dma_addr_t dma_addr)
+dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
 	return 0;
 }
diff --git a/include/linux/i2o.h b/include/linux/i2o.h
index 7d51cbc..75ae6d8 100644
--- a/include/linux/i2o.h
+++ b/include/linux/i2o.h
@@ -758,7 +758,7 @@ static inline dma_addr_t i2o_dma_map_single(struct i2o_controller *c, void *ptr,
 	}
 
 	dma_addr = dma_map_single(&c->pdev->dev, ptr, size, direction);
-	if (!dma_mapping_error(dma_addr)) {
+	if (!dma_mapping_error(&c->pdev->dev, dma_addr)) {
 #ifdef CONFIG_I2O_EXT_ADAPTEC_DMA64
 		if ((sizeof(dma_addr_t) > 4) && c->pae_support) {
 			*mptr++ = cpu_to_le32(0x7C020002);
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 911a661..972657e 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -1506,7 +1506,7 @@ static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
 {
 	if (dev->dma_ops)
 		return dev->dma_ops->mapping_error(dev, dma_addr);
-	return dma_mapping_error(dma_addr);
+	return dma_mapping_error(dev->dma_device, dma_addr);
 }
 
 /**
diff --git a/lib/swiotlb.c b/lib/swiotlb.c
index d568894..977edbd 100644
--- a/lib/swiotlb.c
+++ b/lib/swiotlb.c
@@ -492,7 +492,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
 		 */
 		dma_addr_t handle;
 		handle = swiotlb_map_single(NULL, NULL, size, DMA_FROM_DEVICE);
-		if (swiotlb_dma_mapping_error(handle))
+		if (swiotlb_dma_mapping_error(hwdev, handle))
 			return NULL;
 
 		ret = bus_to_virt(handle);
@@ -824,7 +824,7 @@ swiotlb_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg,
 }
 
 int
-swiotlb_dma_mapping_error(dma_addr_t dma_addr)
+swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
 {
 	return (dma_addr == virt_to_bus(io_tlb_overflow_buffer));
 }
-- 
1.5.4.2

[PATCH v2 -mm 2/2] x86: per-device dma_mapping_ops support

[FUJITA Tomonori]

This adds per-device dma_mapping_ops support for CONFIG_X86_64.

A pointer to dma_mapping_ops to struct dev_archdata is added. If the
pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If
it's NULL, the system-wide dma_ops pointer is used as before.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
---
 arch/x86/kernel/pci-calgary_64.c |    2 +-
 arch/x86/kernel/pci-dma.c        |    2 +-
 arch/x86/kernel/pci-gart_64.c    |    2 +-
 arch/x86/kernel/pci-nommu.c      |   14 +-----
 arch/x86/kernel/pci-swiotlb_64.c |    2 +-
 include/asm-x86/device.h         |    3 +
 include/asm-x86/dma-mapping.h    |   95 ++++++++++++++++++++++++++------------
 7 files changed, 74 insertions(+), 46 deletions(-)

diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c
index e28ec49..dca9a82 100644
--- a/arch/x86/kernel/pci-calgary_64.c
+++ b/arch/x86/kernel/pci-calgary_64.c
@@ -541,7 +541,7 @@ error:
 	return ret;
 }
 
-static const struct dma_mapping_ops calgary_dma_ops = {
+static struct dma_mapping_ops calgary_dma_ops = {
 	.alloc_coherent = calgary_alloc_coherent,
 	.map_single = calgary_map_single,
 	.unmap_single = calgary_unmap_single,
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index 9c4614a..d2f2304 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -11,7 +11,7 @@
 int forbid_dac __read_mostly;
 EXPORT_SYMBOL(forbid_dac);
 
-const struct dma_mapping_ops *dma_ops;
+struct dma_mapping_ops *dma_ops;
 EXPORT_SYMBOL(dma_ops);
 
 static int iommu_sac_force __read_mostly;
diff --git a/arch/x86/kernel/pci-gart_64.c b/arch/x86/kernel/pci-gart_64.c
index 020cadd..638c214 100644
--- a/arch/x86/kernel/pci-gart_64.c
+++ b/arch/x86/kernel/pci-gart_64.c
@@ -619,7 +619,7 @@ static __init int init_k8_gatt(struct agp_kern_info *info)
 
 extern int agp_amd64_init(void);
 
-static const struct dma_mapping_ops gart_dma_ops = {
+static struct dma_mapping_ops gart_dma_ops = {
 	.map_single			= gart_map_single,
 	.map_simple			= gart_map_simple,
 	.unmap_single			= gart_unmap_single,
diff --git a/arch/x86/kernel/pci-nommu.c b/arch/x86/kernel/pci-nommu.c
index 05d70b8..67cc5ee 100644
--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -72,21 +72,9 @@ static int nommu_map_sg(struct device *hwdev, struct scatterlist *sg,
 	return nents;
 }
 
-/* Make sure we keep the same behaviour */
-static int nommu_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
-{
-#ifdef CONFIG_X86_32
-	return 0;
-#else
-	return (dma_addr == bad_dma_address);
-#endif
-}
-
-
-const struct dma_mapping_ops nommu_dma_ops = {
+struct dma_mapping_ops nommu_dma_ops = {
 	.map_single = nommu_map_single,
 	.map_sg = nommu_map_sg,
-	.mapping_error = nommu_mapping_error,
 	.is_phys = 1,
 };
 
diff --git a/arch/x86/kernel/pci-swiotlb_64.c b/arch/x86/kernel/pci-swiotlb_64.c
index 490da7f..a464016 100644
--- a/arch/x86/kernel/pci-swiotlb_64.c
+++ b/arch/x86/kernel/pci-swiotlb_64.c
@@ -18,7 +18,7 @@ swiotlb_map_single_phys(struct device *hwdev, phys_addr_t paddr, size_t size,
 	return swiotlb_map_single(hwdev, phys_to_virt(paddr), size, direction);
 }
 
-const struct dma_mapping_ops swiotlb_dma_ops = {
+struct dma_mapping_ops swiotlb_dma_ops = {
 	.mapping_error = swiotlb_dma_mapping_error,
 	.alloc_coherent = swiotlb_alloc_coherent,
 	.free_coherent = swiotlb_free_coherent,
diff --git a/include/asm-x86/device.h b/include/asm-x86/device.h
index 87a7153..3c034f4 100644
--- a/include/asm-x86/device.h
+++ b/include/asm-x86/device.h
@@ -5,6 +5,9 @@ struct dev_archdata {
 #ifdef CONFIG_ACPI
 	void	*acpi_handle;
 #endif
+#ifdef CONFIG_X86_64
+struct dma_mapping_ops *dma_ops;
+#endif
 #ifdef CONFIG_DMAR
 	void *iommu; /* hook for IOMMU specific extension */
 #endif
diff --git a/include/asm-x86/dma-mapping.h b/include/asm-x86/dma-mapping.h
index 69626b6..feb9a53 100644
--- a/include/asm-x86/dma-mapping.h
+++ b/include/asm-x86/dma-mapping.h
@@ -58,14 +58,33 @@ struct dma_mapping_ops {
 	int		is_phys;
 };
 
-extern const struct dma_mapping_ops *dma_ops;
+extern struct dma_mapping_ops *dma_ops;
 
+static inline struct dma_mapping_ops *get_dma_ops(struct device *dev)
+{
+#ifdef CONFIG_X86_32
+	return dma_ops;
+#else
+	if (unlikely(!dev) || !dev->archdata.dma_ops)
+		return dma_ops;
+	else
+		return dev->archdata.dma_ops;
+#endif
+}
+
+/* Make sure we keep the same behaviour */
 static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr)
 {
-	if (dma_ops->mapping_error)
-		return dma_ops->mapping_error(dev, dma_addr);
+	struct dma_mapping_ops *ops = get_dma_ops(dev);
+
+#ifdef CONFIG_X86_32
+	return 0;
+#else
+	if (ops->mapping_error)
+		return ops->mapping_error(dev, dma_addr);
 
 	return (dma_addr == bad_dma_address);
+#endif
 }
 
 #define dma_alloc_noncoherent(d, s, h, f) dma_alloc_coherent(d, s, h, f)
@@ -85,44 +104,53 @@ static inline dma_addr_t
 dma_map_single(struct device *hwdev, void *ptr, size_t size,
 	       int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	return dma_ops->map_single(hwdev, virt_to_phys(ptr), size, direction);
+	return ops->map_single(hwdev, virt_to_phys(ptr), size, direction);
 }
 
 static inline void
 dma_unmap_single(struct device *dev, dma_addr_t addr, size_t size,
 		 int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(dev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->unmap_single)
-		dma_ops->unmap_single(dev, addr, size, direction);
+	if (ops->unmap_single)
+		ops->unmap_single(dev, addr, size, direction);
 }
 
 static inline int
 dma_map_sg(struct device *hwdev, struct scatterlist *sg,
 	   int nents, int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	return dma_ops->map_sg(hwdev, sg, nents, direction);
+	return ops->map_sg(hwdev, sg, nents, direction);
 }
 
 static inline void
 dma_unmap_sg(struct device *hwdev, struct scatterlist *sg, int nents,
 	     int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->unmap_sg)
-		dma_ops->unmap_sg(hwdev, sg, nents, direction);
+	if (ops->unmap_sg)
+		ops->unmap_sg(hwdev, sg, nents, direction);
 }
 
 static inline void
 dma_sync_single_for_cpu(struct device *hwdev, dma_addr_t dma_handle,
 			size_t size, int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_single_for_cpu)
-		dma_ops->sync_single_for_cpu(hwdev, dma_handle, size,
-					     direction);
+	if (ops->sync_single_for_cpu)
+		ops->sync_single_for_cpu(hwdev, dma_handle, size, direction);
 	flush_write_buffers();
 }
 
@@ -130,10 +158,11 @@ static inline void
 dma_sync_single_for_device(struct device *hwdev, dma_addr_t dma_handle,
 			   size_t size, int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_single_for_device)
-		dma_ops->sync_single_for_device(hwdev, dma_handle, size,
-						direction);
+	if (ops->sync_single_for_device)
+		ops->sync_single_for_device(hwdev, dma_handle, size, direction);
 	flush_write_buffers();
 }
 
@@ -141,11 +170,12 @@ static inline void
 dma_sync_single_range_for_cpu(struct device *hwdev, dma_addr_t dma_handle,
 			      unsigned long offset, size_t size, int direction)
 {
-	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_single_range_for_cpu)
-		dma_ops->sync_single_range_for_cpu(hwdev, dma_handle, offset,
-						   size, direction);
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
 
+	BUG_ON(!valid_dma_direction(direction));
+	if (ops->sync_single_range_for_cpu)
+		ops->sync_single_range_for_cpu(hwdev, dma_handle, offset,
+					       size, direction);
 	flush_write_buffers();
 }
 
@@ -154,11 +184,12 @@ dma_sync_single_range_for_device(struct device *hwdev, dma_addr_t dma_handle,
 				 unsigned long offset, size_t size,
 				 int direction)
 {
-	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_single_range_for_device)
-		dma_ops->sync_single_range_for_device(hwdev, dma_handle,
-						      offset, size, direction);
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
 
+	BUG_ON(!valid_dma_direction(direction));
+	if (ops->sync_single_range_for_device)
+		ops->sync_single_range_for_device(hwdev, dma_handle,
+						  offset, size, direction);
 	flush_write_buffers();
 }
 
@@ -166,9 +197,11 @@ static inline void
 dma_sync_sg_for_cpu(struct device *hwdev, struct scatterlist *sg,
 		    int nelems, int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_sg_for_cpu)
-		dma_ops->sync_sg_for_cpu(hwdev, sg, nelems, direction);
+	if (ops->sync_sg_for_cpu)
+		ops->sync_sg_for_cpu(hwdev, sg, nelems, direction);
 	flush_write_buffers();
 }
 
@@ -176,9 +209,11 @@ static inline void
 dma_sync_sg_for_device(struct device *hwdev, struct scatterlist *sg,
 		       int nelems, int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	if (dma_ops->sync_sg_for_device)
-		dma_ops->sync_sg_for_device(hwdev, sg, nelems, direction);
+	if (ops->sync_sg_for_device)
+		ops->sync_sg_for_device(hwdev, sg, nelems, direction);
 
 	flush_write_buffers();
 }
@@ -187,9 +222,11 @@ static inline dma_addr_t dma_map_page(struct device *dev, struct page *page,
 				      size_t offset, size_t size,
 				      int direction)
 {
+	struct dma_mapping_ops *ops = get_dma_ops(dev);
+
 	BUG_ON(!valid_dma_direction(direction));
-	return dma_ops->map_single(dev, page_to_phys(page)+offset,
-				   size, direction);
+	return ops->map_single(dev, page_to_phys(page) + offset,
+			       size, direction);
 }
 
 static inline void dma_unmap_page(struct device *dev, dma_addr_t addr,
-- 
1.5.4.2

Re: [PATCH v2 -mm 1/2] add the device argument to dma_mapping_error

[Andrew Morton]

On Mon, 19 May 2008 15:31:28 +0900 FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> wrote:

> dma_mapping_error doesn't take a pointer to the device unlike other
> DMA operations. So we can't have dma_mapping_ops per device.
> 
> Note that POWER already has dma_mapping_ops per device but all the
> POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use
> different dma_mapping_error functions. So dma_mapping_error needs the
> device argument.

This patch continues to turn my hair grey.

I'm currently staring at this, in include/linux/ssb/ssb.h:

static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
{
	switch (dev->bus->bustype) {
	case SSB_BUSTYPE_PCI:
		return pci_dma_mapping_error(dev->dev, addr);
	case SSB_BUSTYPE_SSB:
		return dma_mapping_error(dev->dev, addr);
	default:
		__ssb_dma_not_implemented(dev);
	}
	return -ENOSYS;
}

How do I go from an ssb_device* to a pci_dev*?

Dunno.  I think I'll cheat and do:

static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
{
	switch (dev->bus->bustype) {
	case SSB_BUSTYPE_PCI:
		return dma_mapping_error(dev->dev, addr);
	case SSB_BUSTYPE_SSB:
		return dma_mapping_error(dev->dev, addr);
	default:
		__ssb_dma_not_implemented(dev);
	}
	return -ENOSYS;
}

please take a look, see if we can do better?

Re: [PATCH v2 -mm 1/2] add the device argument to dma_mapping_error

[Michael Buesch]

On Wednesday 02 July 2008 12:07:23 Andrew Morton wrote:
> On Mon, 19 May 2008 15:31:28 +0900 FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> wrote:
> 
> > dma_mapping_error doesn't take a pointer to the device unlike other
> > DMA operations. So we can't have dma_mapping_ops per device.
> > 
> > Note that POWER already has dma_mapping_ops per device but all the
> > POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use
> > different dma_mapping_error functions. So dma_mapping_error needs the
> > device argument.
> 
> This patch continues to turn my hair grey.
> 
> I'm currently staring at this, in include/linux/ssb/ssb.h:
> 
> static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
> {
> 	switch (dev->bus->bustype) {
> 	case SSB_BUSTYPE_PCI:
> 		return pci_dma_mapping_error(dev->dev, addr);
> 	case SSB_BUSTYPE_SSB:
> 		return dma_mapping_error(dev->dev, addr);
> 	default:
> 		__ssb_dma_not_implemented(dev);
> 	}
> 	return -ENOSYS;
> }
> 
> How do I go from an ssb_device* to a pci_dev*?
> 
> Dunno.  I think I'll cheat and do:
> 
> static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
> {
> 	switch (dev->bus->bustype) {
> 	case SSB_BUSTYPE_PCI:
> 		return dma_mapping_error(dev->dev, addr);

The statement above is wrong. The PCI-specific dma-mapping-error function
must be used here. (I hope such a thing exists. Otherwise the API is broken).
So
		return pci_mapping_error(dev->bus->host_pci, addr);

> 	case SSB_BUSTYPE_SSB:
> 		return dma_mapping_error(dev->dev, addr);
> 	default:
> 		__ssb_dma_not_implemented(dev);
> 	}
> 	return -ENOSYS;
> }



-- 
Greetings Michael.

Re: [PATCH v2 -mm 1/2] add the device argument to dma_mapping_error

[Michael Buesch]

On Wednesday 02 July 2008 12:18:40 Michael Buesch wrote:
> On Wednesday 02 July 2008 12:07:23 Andrew Morton wrote:
> > On Mon, 19 May 2008 15:31:28 +0900 FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> wrote:
> > 
> > > dma_mapping_error doesn't take a pointer to the device unlike other
> > > DMA operations. So we can't have dma_mapping_ops per device.
> > > 
> > > Note that POWER already has dma_mapping_ops per device but all the
> > > POWER IOMMUs use the same dma_mapping_error function. x86 IOMMUs use
> > > different dma_mapping_error functions. So dma_mapping_error needs the
> > > device argument.
> > 
> > This patch continues to turn my hair grey.
> > 
> > I'm currently staring at this, in include/linux/ssb/ssb.h:
> > 
> > static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
> > {
> > 	switch (dev->bus->bustype) {
> > 	case SSB_BUSTYPE_PCI:
> > 		return pci_dma_mapping_error(dev->dev, addr);
> > 	case SSB_BUSTYPE_SSB:
> > 		return dma_mapping_error(dev->dev, addr);
> > 	default:
> > 		__ssb_dma_not_implemented(dev);
> > 	}
> > 	return -ENOSYS;
> > }
> > 
> > How do I go from an ssb_device* to a pci_dev*?
> > 
> > Dunno.  I think I'll cheat and do:
> > 
> > static inline int ssb_dma_mapping_error(struct ssb_device *dev, dma_addr_t addr)
> > {
> > 	switch (dev->bus->bustype) {
> > 	case SSB_BUSTYPE_PCI:
> > 		return dma_mapping_error(dev->dev, addr);
> 
> The statement above is wrong. The PCI-specific dma-mapping-error function
> must be used here. (I hope such a thing exists. Otherwise the API is broken).
> So
> 		return pci_mapping_error(dev->bus->host_pci, addr);

Oh wait I see you quoted the function above. So the statement should be:

		return pci_dma_mapping_error(dev->bus->host_pci, addr);


-- 
Greetings Michael.

Re: [PATCH v2 -mm 1/2] add the device argument to dma_mapping_error

[Andrew Morton]

On Wed, 2 Jul 2008 12:20:43 +0200 Michael Buesch <mb@bu3sch.de> wrote:

> 		return pci_dma_mapping_error(dev->bus->host_pci, addr);

that works, thanks.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Amit Shah]

On Monday 19 May 2008 12:01:27 FUJITA Tomonori wrote:
> This is an updated version of the patchset to add per-device
> dma_mapping_ops support for CONFIG_X86_64 like POWER architecture
> does:
>
> http://lkml.org/lkml/2008/5/13/36
>
> This is against 2.6.26-rc2-mm1 (the changes since the v1 are pretty
> trivial like dropping the change for v850 arch).
>
> This enables us to cleanly fix the Calgary IOMMU issue that some
> devices are not behind the IOMMU [1].
>
> I think that per-device dma_mapping_ops support would be also helpful
> for KVM people to support PCI passthrough but Andi thinks that this
> makes it difficult to support the PCI passthrough (see the above
> thread). So I CC'ed this to KVM camp. Comments are appreciated.

I think what's more useful is a chain with a properly defined order or 
hierarchy (based on what Muli suggested last time we discussed this

http://lkml.org/lkml/2007/11/12/44 )

The suggested order was (in calling order):
pvdma->hardare->nommu/swiotlb

The discussion in the thread pointed to above has details as to why.

> A pointer to dma_mapping_ops to struct dev_archdata is added. If the
> pointer is non NULL, DMA operations in asm/dma-mapping.h use it. If
> it's NULL, the system-wide dma_ops pointer is used as before.


> If it's useful for KVM people, I plan to implement a mechanism to
> register a hook called when a new pci (or dma capable) device is

OK; this sounds helpful. the hook can make a hypercall and confirm with the 
host kernel if the device in question is an assigned physical device. If yes, 
we replace the dma_ops. Though, the original intent of having stackable ops 
is that we might want to go through the swiotlb in the guest even for an 
assigned device if the guest dma addresses are not in the addressable range 
of the guest chipset.

> created (it works with hot plugging). It enables IOMMUs to set up an
> appropriate dma_mapping_ops per device.

>From what we've discussed so far, it looks like stackable dma ops will 
definitely be needed. Does this patchset provide something that stacking 
won't?

> [1] http://lkml.org/lkml/2008/5/8/423

Amit.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Muli Ben-Yehuda]

On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:

> OK; this sounds helpful. the hook can make a hypercall and confirm
> with the host kernel if the device in question is an assigned
> physical device. If yes, we replace the dma_ops. Though, the
> original intent of having stackable ops is that we might want to go
> through the swiotlb in the guest even for an assigned device if the
> guest dma addresses are not in the addressable range of the guest
> chipset.
> 
> > created (it works with hot plugging). It enables IOMMUs to set up an
> > appropriate dma_mapping_ops per device.
> 
> From what we've discussed so far, it looks like stackable dma ops will 
> definitely be needed. Does this patchset provide something that stacking 
> won't?

Yes---this patchset let's you have a per-device dma-ops, whereas with
stackable you only get global dma-ops. I think it's clear we need
both, and I think per-device dma-ops are the first thing that's
needed. Stacking can then be introduced on a per-device basis.

Cheers,
Muli

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Amit Shah]

On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > OK; this sounds helpful. the hook can make a hypercall and confirm
> > with the host kernel if the device in question is an assigned
> > physical device. If yes, we replace the dma_ops. Though, the
> > original intent of having stackable ops is that we might want to go
> > through the swiotlb in the guest even for an assigned device if the
> > guest dma addresses are not in the addressable range of the guest
> > chipset.
> >
> > > created (it works with hot plugging). It enables IOMMUs to set up an
> > > appropriate dma_mapping_ops per device.
> >
> > From what we've discussed so far, it looks like stackable dma ops will
> > definitely be needed. Does this patchset provide something that stacking
> > won't?
>
> Yes---this patchset let's you have a per-device dma-ops, whereas with
> stackable you only get global dma-ops. I think it's clear we need
> both, and I think per-device dma-ops are the first thing that's
> needed. Stacking can then be introduced on a per-device basis.

When we would want stacking, we'll want it globally and not per-device, isn't 
it? Or at least for devices on a particular bus.

When an IOMMU driver registers itself, it should tell which devices it's 
interested in (each device behind a bus or by enumerating each device it 
cares for). This should take care of all the scenarios and we won't have the 
need for per-device dma_ops.

For something like pvdma, we can walk through the list of pci devices and make 
a hypercall for each of them to get this information and have the pvdma 
version of dma_ops registered for that device. This sounds like it's 
per-device dma_ops, but it's not -- internally, the dma operations walk 
through each of the IOMMUs registered and call them in sequence.

Does this work?

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[FUJITA Tomonori]

On Mon, 26 May 2008 09:39:20 +0530
Amit Shah <amit.shah@qumranet.com> wrote:

> On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> > On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > > OK; this sounds helpful. the hook can make a hypercall and confirm
> > > with the host kernel if the device in question is an assigned
> > > physical device. If yes, we replace the dma_ops. Though, the
> > > original intent of having stackable ops is that we might want to go
> > > through the swiotlb in the guest even for an assigned device if the
> > > guest dma addresses are not in the addressable range of the guest
> > > chipset.
> > >
> > > > created (it works with hot plugging). It enables IOMMUs to set up an
> > > > appropriate dma_mapping_ops per device.
> > >
> > > From what we've discussed so far, it looks like stackable dma ops will
> > > definitely be needed. Does this patchset provide something that stacking
> > > won't?
> >
> > Yes---this patchset let's you have a per-device dma-ops, whereas with
> > stackable you only get global dma-ops. I think it's clear we need
> > both, and I think per-device dma-ops are the first thing that's
> > needed. Stacking can then be introduced on a per-device basis.
> 
> When we would want stacking, we'll want it globally and not per-device, isn't 
> it? Or at least for devices on a particular bus.
> 
> When an IOMMU driver registers itself, it should tell which devices it's 
> interested in (each device behind a bus or by enumerating each device it 
> cares for). This should take care of all the scenarios and we won't have the 
> need for per-device dma_ops.

Well, without per-device dma_ops, IOMMUs could live. But it's pretty
hacky. Every time a dma operation is called, IOMMUs need to figure out
how a device should be handled.

If IOMMUs can set dma_ops for the device when a new device is created,
IOMMUs don't care anything any more. That's much clean. That's What
the POWER architecture does.


> For something like pvdma, we can walk through the list of pci devices and make 
> a hypercall for each of them to get this information and have the pvdma 
> version of dma_ops registered for that device. This sounds like it's 
> per-device dma_ops, but it's not -- internally, the dma operations walk 
> through each of the IOMMUs registered and call them in sequence.

As Muli poinsted out, For pvdma, you need stacking per-device
dma_ops. With per-device dma_ops, you don't need hack like adding
is_pv_device hook in dma_ops. You can set your dma_ops to only pci
devices that you are interested.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Amit Shah]

On Monday 26 May 2008 11:41:52 FUJITA Tomonori wrote:
> On Mon, 26 May 2008 09:39:20 +0530
>
> Amit Shah <amit.shah@qumranet.com> wrote:
> > On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> > > On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > > > OK; this sounds helpful. the hook can make a hypercall and confirm
> > > > with the host kernel if the device in question is an assigned
> > > > physical device. If yes, we replace the dma_ops. Though, the
> > > > original intent of having stackable ops is that we might want to go
> > > > through the swiotlb in the guest even for an assigned device if the
> > > > guest dma addresses are not in the addressable range of the guest
> > > > chipset.
> > > >
> > > > > created (it works with hot plugging). It enables IOMMUs to set up
> > > > > an appropriate dma_mapping_ops per device.
> > > >
> > > > From what we've discussed so far, it looks like stackable dma ops
> > > > will definitely be needed. Does this patchset provide something that
> > > > stacking won't?
> > >
> > > Yes---this patchset let's you have a per-device dma-ops, whereas with
> > > stackable you only get global dma-ops. I think it's clear we need
> > > both, and I think per-device dma-ops are the first thing that's
> > > needed. Stacking can then be introduced on a per-device basis.
> >
> > When we would want stacking, we'll want it globally and not per-device,
> > isn't it? Or at least for devices on a particular bus.
> >
> > When an IOMMU driver registers itself, it should tell which devices it's
> > interested in (each device behind a bus or by enumerating each device it
> > cares for). This should take care of all the scenarios and we won't have
> > the need for per-device dma_ops.
>
> Well, without per-device dma_ops, IOMMUs could live. But it's pretty
> hacky. Every time a dma operation is called, IOMMUs need to figure out
> how a device should be handled.

What if this information could be hidden behind (a slightly complicated) 
get_dma_ops()? Also, each of the operations in dma_ops will see if there's 
something else down the stack that might be interested in the current device.

My contention is that we are going to need stackable ops, and a full-fledged 
stackable implementation is going to solve this problem as well. However, 
this current implementation of per-device dma_ops looks like a really simple 
and non-intrusive solution to one problem, that of getting rid of some 
overheads in the IOMMU code.

> If IOMMUs can set dma_ops for the device when a new device is created,
> IOMMUs don't care anything any more. That's much clean. That's What
> the POWER architecture does.
>
> > For something like pvdma, we can walk through the list of pci devices and
> > make a hypercall for each of them to get this information and have the
> > pvdma version of dma_ops registered for that device. This sounds like
> > it's per-device dma_ops, but it's not -- internally, the dma operations
> > walk through each of the IOMMUs registered and call them in sequence.
>
> As Muli poinsted out, For pvdma, you need stacking per-device
> dma_ops. With per-device dma_ops, you don't need hack like adding
> is_pv_device hook in dma_ops. You can set your dma_ops to only pci
> devices that you are interested.

The hack was added only because there's no stackable dma api we have now. 
Sure, per-device dma_ops is going to solve this problem and I like it. I'm 
only saying we're also going to need stacking ops and in effect, per-device 
dma_ops would just be replaced by them once we get the complete solution.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[FUJITA Tomonori]

On Mon, 26 May 2008 22:14:34 +0530
Amit Shah <amit.shah@qumranet.com> wrote:

> On Monday 26 May 2008 11:41:52 FUJITA Tomonori wrote:
> > On Mon, 26 May 2008 09:39:20 +0530
> >
> > Amit Shah <amit.shah@qumranet.com> wrote:
> > > On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> > > > On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > > > > OK; this sounds helpful. the hook can make a hypercall and confirm
> > > > > with the host kernel if the device in question is an assigned
> > > > > physical device. If yes, we replace the dma_ops. Though, the
> > > > > original intent of having stackable ops is that we might want to go
> > > > > through the swiotlb in the guest even for an assigned device if the
> > > > > guest dma addresses are not in the addressable range of the guest
> > > > > chipset.
> > > > >
> > > > > > created (it works with hot plugging). It enables IOMMUs to set up
> > > > > > an appropriate dma_mapping_ops per device.
> > > > >
> > > > > From what we've discussed so far, it looks like stackable dma ops
> > > > > will definitely be needed. Does this patchset provide something that
> > > > > stacking won't?
> > > >
> > > > Yes---this patchset let's you have a per-device dma-ops, whereas with
> > > > stackable you only get global dma-ops. I think it's clear we need
> > > > both, and I think per-device dma-ops are the first thing that's
> > > > needed. Stacking can then be introduced on a per-device basis.
> > >
> > > When we would want stacking, we'll want it globally and not per-device,
> > > isn't it? Or at least for devices on a particular bus.
> > >
> > > When an IOMMU driver registers itself, it should tell which devices it's
> > > interested in (each device behind a bus or by enumerating each device it
> > > cares for). This should take care of all the scenarios and we won't have
> > > the need for per-device dma_ops.
> >
> > Well, without per-device dma_ops, IOMMUs could live. But it's pretty
> > hacky. Every time a dma operation is called, IOMMUs need to figure out
> > how a device should be handled.
> 
> What if this information could be hidden behind (a slightly complicated) 
> get_dma_ops()? Also, each of the operations in dma_ops will see if there's 
> something else down the stack that might be interested in the current device.

dma_ops can't do anything since only IOMMUs know what to do against a
device.

Whatever you implement in dma_ops, without per-device dma_ops, IOMMUs
need to figure out what to do a device every time a dma operation is
called.


> My contention is that we are going to need stackable ops, and a full-fledged 
> stackable implementation is going to solve this problem as well. However, 
> this current implementation of per-device dma_ops looks like a really simple 
> and non-intrusive solution to one problem, that of getting rid of some 
> overheads in the IOMMU code.

I don't think that stackable ops solve the problem that some IOMMUs have.


> > If IOMMUs can set dma_ops for the device when a new device is created,
> > IOMMUs don't care anything any more. That's much clean. That's What
> > the POWER architecture does.
> >
> > > For something like pvdma, we can walk through the list of pci devices and
> > > make a hypercall for each of them to get this information and have the
> > > pvdma version of dma_ops registered for that device. This sounds like
> > > it's per-device dma_ops, but it's not -- internally, the dma operations
> > > walk through each of the IOMMUs registered and call them in sequence.
> >
> > As Muli poinsted out, For pvdma, you need stacking per-device
> > dma_ops. With per-device dma_ops, you don't need hack like adding
> > is_pv_device hook in dma_ops. You can set your dma_ops to only pci
> > devices that you are interested.
> 
> The hack was added only because there's no stackable dma api we have now. 
> Sure, per-device dma_ops is going to solve this problem and I like it. I'm 
> only saying we're also going to need stacking ops and in effect, per-device 
> dma_ops would just be replaced by them once we get the complete solution.

Again, stackable ops can't cleanly solve the problem that per-device
dma_ops tries to solve. For example, you stack dma_ops like
pvdma->hardare->nommu/swiotlb. How can pvdma_ops know if pvdma_ops
needs to handle a device or not? pvdma_ops needs to skip some devices
and handle some. per-device dma_ops enables us not to stack pvdma_ops
for devices that pvdma_ops are not instrested in. That's much clean.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Amit Shah]

On Tuesday 27 May 2008 05:20:54 FUJITA Tomonori wrote:
> On Mon, 26 May 2008 22:14:34 +0530
>
> Amit Shah <amit.shah@qumranet.com> wrote:
> > On Monday 26 May 2008 11:41:52 FUJITA Tomonori wrote:
> > > On Mon, 26 May 2008 09:39:20 +0530
> > >
> > > Amit Shah <amit.shah@qumranet.com> wrote:
> > > > On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> > > > > On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > > > > > OK; this sounds helpful. the hook can make a hypercall and
> > > > > > confirm with the host kernel if the device in question is an
> > > > > > assigned physical device. If yes, we replace the dma_ops. Though,
> > > > > > the original intent of having stackable ops is that we might want
> > > > > > to go through the swiotlb in the guest even for an assigned
> > > > > > device if the guest dma addresses are not in the addressable
> > > > > > range of the guest chipset.
> > > > > >
> > > > > > > created (it works with hot plugging). It enables IOMMUs to set
> > > > > > > up an appropriate dma_mapping_ops per device.
> > > > > >
> > > > > > From what we've discussed so far, it looks like stackable dma ops
> > > > > > will definitely be needed. Does this patchset provide something
> > > > > > that stacking won't?
> > > > >
> > > > > Yes---this patchset let's you have a per-device dma-ops, whereas
> > > > > with stackable you only get global dma-ops. I think it's clear we
> > > > > need both, and I think per-device dma-ops are the first thing
> > > > > that's needed. Stacking can then be introduced on a per-device
> > > > > basis.
> > > >
> > > > When we would want stacking, we'll want it globally and not
> > > > per-device, isn't it? Or at least for devices on a particular bus.
> > > >
> > > > When an IOMMU driver registers itself, it should tell which devices
> > > > it's interested in (each device behind a bus or by enumerating each
> > > > device it cares for). This should take care of all the scenarios and
> > > > we won't have the need for per-device dma_ops.
> > >
> > > Well, without per-device dma_ops, IOMMUs could live. But it's pretty
> > > hacky. Every time a dma operation is called, IOMMUs need to figure out
> > > how a device should be handled.
> >
> > What if this information could be hidden behind (a slightly complicated)
> > get_dma_ops()? Also, each of the operations in dma_ops will see if
> > there's something else down the stack that might be interested in the
> > current device.
>
> dma_ops can't do anything since only IOMMUs know what to do against a
> device.

Instead of each device calling a function to check which IOMMU is right, I am 
suggesting each IOMMU come in and tell which devices it is interested in.

> Again, stackable ops can't cleanly solve the problem that per-device
> dma_ops tries to solve. For example, you stack dma_ops like
> pvdma->hardare->nommu/swiotlb. How can pvdma_ops know if pvdma_ops
> needs to handle a device or not? pvdma_ops needs to skip some devices
> and handle some. per-device dma_ops enables us not to stack pvdma_ops
> for devices that pvdma_ops are not instrested in. That's much clean.

OK; how about this:

An example with per-device dma_ops and stacking will look like this:

pvdma->hardware->nommu/swiotlb
    ^             ^
     |              |
 e1000     rtl8139

And this scheme is going to suit everyone, agreed?

This is simple and doesn't need too many changes all around.

I was suggesting something more than this that can handle cases like an iommu 
wanting to have each device behind a bus to pass through it (it's still 
possible, but needs a per-device walk). Also, in the scenario depicted above, 
each device will start by pointing to the first iommu in the chain (pvdma in 
this case) and the iommu will then determine if that device needs to be 
passed via its translations.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[FUJITA Tomonori]

On Tue, 27 May 2008 10:23:21 +0530
Amit Shah <amit.shah@qumranet.com> wrote:

> On Tuesday 27 May 2008 05:20:54 FUJITA Tomonori wrote:
> > On Mon, 26 May 2008 22:14:34 +0530
> >
> > Amit Shah <amit.shah@qumranet.com> wrote:
> > > On Monday 26 May 2008 11:41:52 FUJITA Tomonori wrote:
> > > > On Mon, 26 May 2008 09:39:20 +0530
> > > >
> > > > Amit Shah <amit.shah@qumranet.com> wrote:
> > > > > On Sunday 25 May 2008 12:50:11 Muli Ben-Yehuda wrote:
> > > > > > On Thu, May 22, 2008 at 04:13:02PM +0530, Amit Shah wrote:
> > > > > > > OK; this sounds helpful. the hook can make a hypercall and
> > > > > > > confirm with the host kernel if the device in question is an
> > > > > > > assigned physical device. If yes, we replace the dma_ops. Though,
> > > > > > > the original intent of having stackable ops is that we might want
> > > > > > > to go through the swiotlb in the guest even for an assigned
> > > > > > > device if the guest dma addresses are not in the addressable
> > > > > > > range of the guest chipset.
> > > > > > >
> > > > > > > > created (it works with hot plugging). It enables IOMMUs to set
> > > > > > > > up an appropriate dma_mapping_ops per device.
> > > > > > >
> > > > > > > From what we've discussed so far, it looks like stackable dma ops
> > > > > > > will definitely be needed. Does this patchset provide something
> > > > > > > that stacking won't?
> > > > > >
> > > > > > Yes---this patchset let's you have a per-device dma-ops, whereas
> > > > > > with stackable you only get global dma-ops. I think it's clear we
> > > > > > need both, and I think per-device dma-ops are the first thing
> > > > > > that's needed. Stacking can then be introduced on a per-device
> > > > > > basis.
> > > > >
> > > > > When we would want stacking, we'll want it globally and not
> > > > > per-device, isn't it? Or at least for devices on a particular bus.
> > > > >
> > > > > When an IOMMU driver registers itself, it should tell which devices
> > > > > it's interested in (each device behind a bus or by enumerating each
> > > > > device it cares for). This should take care of all the scenarios and
> > > > > we won't have the need for per-device dma_ops.
> > > >
> > > > Well, without per-device dma_ops, IOMMUs could live. But it's pretty
> > > > hacky. Every time a dma operation is called, IOMMUs need to figure out
> > > > how a device should be handled.
> > >
> > > What if this information could be hidden behind (a slightly complicated)
> > > get_dma_ops()? Also, each of the operations in dma_ops will see if
> > > there's something else down the stack that might be interested in the
> > > current device.
> >
> > dma_ops can't do anything since only IOMMUs know what to do against a
> > device.
> 
> Instead of each device calling a function to check which IOMMU is right, I am 
> suggesting each IOMMU come in and tell which devices it is interested in.

It means that you need to register IOMMU information per
device. That's same to per-device dma_ops.

Or It means you need put devices (an IOMMU is interested in) to a
list. Every time dma operation is called, you check the list to see
who is interested in a device. That's not clean (not effective too).

 
> > Again, stackable ops can't cleanly solve the problem that per-device
> > dma_ops tries to solve. For example, you stack dma_ops like
> > pvdma->hardare->nommu/swiotlb. How can pvdma_ops know if pvdma_ops
> > needs to handle a device or not? pvdma_ops needs to skip some devices
> > and handle some. per-device dma_ops enables us not to stack pvdma_ops
> > for devices that pvdma_ops are not instrested in. That's much clean.
> 
> OK; how about this:
> 
> An example with per-device dma_ops and stacking will look like this:
> 
> pvdma->hardware->nommu/swiotlb
>     ^             ^
>      |              |
>  e1000     rtl8139
> 
> And this scheme is going to suit everyone, agreed?
> 
> This is simple and doesn't need too many changes all around.

Sorry, I'm not sure what this picture represents.

BTW, without pvdma, there is no need to hardware->nommu/swiotlb
stacking for IOMMUs like Calgary. Per-device dma_ops wor for them.


> I was suggesting something more than this that can handle cases like an iommu 
> wanting to have each device behind a bus to pass through it (it's still 
> possible, but needs a per-device walk). Also, in the scenario depicted above, 
> each device will start by pointing to the first iommu in the chain (pvdma in 
> this case) and the iommu will then determine if that device needs to be 
> passed via its translations.

No, IOMMUs doesn't need to do that. We need to put a stacking
mechanism in dma-mapping.h. A stacking mechanism should not be visible
to IOMMUs.

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[Amit Shah]

On Tuesday 27 May 2008 10:54:06 FUJITA Tomonori wrote:

> > An example with per-device dma_ops and stacking will look like this:
> >
> > pvdma->hardware->nommu/swiotlb
> >     ^             ^
> >
> >  e1000     rtl8139
> >
> > And this scheme is going to suit everyone, agreed?
> >
> > This is simple and doesn't need too many changes all around.
>
> Sorry, I'm not sure what this picture represents.

It meant to show just e1000 needs to go through the pvdma translations. 
rtl8139 goes via the other iommus. e1000 also goes through the other iommus 
(mainly if it's going to be the swiotlb that a guest might need).

> BTW, without pvdma, there is no need to hardware->nommu/swiotlb
> stacking for IOMMUs like Calgary. Per-device dma_ops wor for them.

Hmm, ok. Then this argument doesn't count.

> > I was suggesting something more than this that can handle cases like an
> > iommu wanting to have each device behind a bus to pass through it (it's
> > still possible, but needs a per-device walk). Also, in the scenario
> > depicted above, each device will start by pointing to the first iommu in
> > the chain (pvdma in this case) and the iommu will then determine if that
> > device needs to be passed via its translations.
>
> No, IOMMUs doesn't need to do that. We need to put a stacking
> mechanism in dma-mapping.h. A stacking mechanism should not be visible
> to IOMMUs.

OK; then just per-device dma_ops will work and for the pvdma case, we'll have 
to have the stacking. Since this is a special case, any kind of generic APIs 
shouldn't be needed as well.

What is the plan with this patch then? When do you plan to ask for mainline 
merging?

Re: [PATCH v2 -mm 0/2] x86: per-device dma_mapping_ops

[FUJITA Tomonori]

On Tue, 27 May 2008 11:24:08 +0530
Amit Shah <amit.shah@qumranet.com> wrote:

> On Tuesday 27 May 2008 10:54:06 FUJITA Tomonori wrote:
> 
> > > An example with per-device dma_ops and stacking will look like this:
> > >
> > > pvdma->hardware->nommu/swiotlb
> > >     ^             ^
> > >
> > >  e1000     rtl8139
> > >
> > > And this scheme is going to suit everyone, agreed?
> > >
> > > This is simple and doesn't need too many changes all around.
> >
> > Sorry, I'm not sure what this picture represents.
> 
> It meant to show just e1000 needs to go through the pvdma translations. 
> rtl8139 goes via the other iommus. e1000 also goes through the other iommus 
> (mainly if it's going to be the swiotlb that a guest might need).
> 
> > BTW, without pvdma, there is no need to hardware->nommu/swiotlb
> > stacking for IOMMUs like Calgary. Per-device dma_ops wor for them.
> 
> Hmm, ok. Then this argument doesn't count.
> 
> > > I was suggesting something more than this that can handle cases like an
> > > iommu wanting to have each device behind a bus to pass through it (it's
> > > still possible, but needs a per-device walk). Also, in the scenario
> > > depicted above, each device will start by pointing to the first iommu in
> > > the chain (pvdma in this case) and the iommu will then determine if that
> > > device needs to be passed via its translations.
> >
> > No, IOMMUs doesn't need to do that. We need to put a stacking
> > mechanism in dma-mapping.h. A stacking mechanism should not be visible
> > to IOMMUs.
> 
> OK; then just per-device dma_ops will work and for the pvdma case, we'll have 
> to have the stacking. Since this is a special case, any kind of generic APIs 
> shouldn't be needed as well.

I'm not sure what you mean exactly, but there might be other people
who want to use the dma_ops stacking though I'm not sure yet how they
use it:

http://lkml.org/lkml/2008/5/15/79


> What is the plan with this patch then? When do you plan to ask for mainline 
> merging?

Andrew already put this patchset in -mm. Unless someone comes with a
new reason against this patchset, it will be merged, I think.

BTW, Andrew, really sorry about several compile bugs due to the first
patch (changing dma_mapping_error) in this patchset. And thanks a lot
for fixing them.