dma-mapping: treat dev->bus_dma_mask as a DMA limit

Using a mask to represent bus DMA constraints has a set of limitations.
The biggest one being it can only hold a power of two (minus one). The
DMA mapping code is already aware of this and treats dev->bus_dma_mask
as a limit. This quirk is already used by some architectures although
still rare.

With the introduction of the Raspberry Pi 4 we've found a new contender
for the use of bus DMA limits, as its PCIe bus can only address the
lower 3GB of memory (of a total of 4GB). This is impossible to represent
with a mask. To make things worse the device-tree code rounds non power
of two bus DMA limits to the next power of two, which is unacceptable in
this case.

In the light of this, rename dev->bus_dma_mask to dev->bus_dma_limit all
over the tree and treat it as such. Note that dev->bus_dma_limit should
contain the higher accessible DMA address.

Signed-off-by: Nicolas Saenz Julienne <[e-mail]>
Reviewed-by: Robin Murphy <[e-mail]>
Signed-off-by: Christoph Hellwig <[e-mail]>

arch/mips/pci/fixup-sb1250.c

···
 
 /*
  * The BCM1250, etc. PCI host bridge does not support DAC on its 32-bit
- * bus, so we set the bus's DMA mask accordingly.  However the HT link
+ * bus, so we set the bus's DMA limit accordingly.  However the HT link
  * down the artificial PCI-HT bridge supports 40-bit addressing and the
  * SP1011 HT-PCI bridge downstream supports both DAC and a 64-bit bus
  * width, so we record the PCI-HT bridge's secondary and subordinate bus
- * numbers and do not set the mask for devices present in the inclusive
+ * numbers and do not set the limit for devices present in the inclusive
  * range of those.
  */
-struct sb1250_bus_dma_mask_exclude {
+struct sb1250_bus_dma_limit_exclude {
 	bool set;
 	unsigned char start;
 	unsigned char end;
 };
 
-static int sb1250_bus_dma_mask(struct pci_dev *dev, void *data)
+static int sb1250_bus_dma_limit(struct pci_dev *dev, void *data)
 {
-	struct sb1250_bus_dma_mask_exclude *exclude = data;
+	struct sb1250_bus_dma_limit_exclude *exclude = data;
 	bool exclude_this;
 	bool ht_bridge;
 
···
 			exclude->start, exclude->end);
 	} else {
 		dev_dbg(&dev->dev, "disabling DAC for device");
-		dev->dev.bus_dma_mask = DMA_BIT_MASK(32);
+		dev->dev.bus_dma_limit = DMA_BIT_MASK(32);
 	}
 
 	return 0;
···
 
 static void quirk_sb1250_pci_dac(struct pci_dev *dev)
 {
-	struct sb1250_bus_dma_mask_exclude exclude = { .set = false };
+	struct sb1250_bus_dma_limit_exclude exclude = { .set = false };
 
-	pci_walk_bus(dev->bus, sb1250_bus_dma_mask, &exclude);
+	pci_walk_bus(dev->bus, sb1250_bus_dma_limit, &exclude);
 }
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_SIBYTE, PCI_DEVICE_ID_BCM1250_PCI,
 			quirk_sb1250_pci_dac);

arch/powerpc/sysdev/fsl_pci.c

···
 {
 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
 
-	pdev->dev.bus_dma_mask =
-		hose->dma_window_base_cur + hose->dma_window_size;
+	pdev->dev.bus_dma_limit =
+		hose->dma_window_base_cur + hose->dma_window_size - 1;
 }
 
 static void setup_swiotlb_ops(struct pci_controller *hose)
···
 	 * mapping that allows addressing any RAM address from across PCI.
 	 */
 	if (dev_is_pci(dev) && dma_mask >= pci64_dma_offset * 2 - 1) {
-		dev->bus_dma_mask = 0;
+		dev->bus_dma_limit = 0;
 		dev->archdata.dma_offset = pci64_dma_offset;
 	}
 }

arch/x86/kernel/pci-dma.c

···
 
 static int via_no_dac_cb(struct pci_dev *pdev, void *data)
 {
-	pdev->dev.bus_dma_mask = DMA_BIT_MASK(32);
+	pdev->dev.bus_dma_limit = DMA_BIT_MASK(32);
 	return 0;
 }
 

arch/x86/mm/mem_encrypt.c

···
 	if (sme_active()) {
 		u64 dma_enc_mask = DMA_BIT_MASK(__ffs64(sme_me_mask));
 		u64 dma_dev_mask = min_not_zero(dev->coherent_dma_mask,
-						dev->bus_dma_mask);
+						dev->bus_dma_limit);
 
 		if (dma_dev_mask <= dma_enc_mask)
 			return true;

arch/x86/pci/sta2x11-fixup.c

···
 
 	dev->dma_pfn_offset = PFN_DOWN(-amba_base);
 
-	dev->bus_dma_mask = max_amba_addr;
+	dev->bus_dma_limit = max_amba_addr;
 	pci_set_consistent_dma_mask(pdev, max_amba_addr);
 	pci_set_dma_mask(pdev, max_amba_addr);
 

drivers/acpi/arm64/iort.c

···
  */
 void iort_dma_setup(struct device *dev, u64 *dma_addr, u64 *dma_size)
 {
-	u64 mask, dmaaddr = 0, size = 0, offset = 0;
-	int ret, msb;
+	u64 end, mask, dmaaddr = 0, size = 0, offset = 0;
+	int ret;
 
 	/*
 	 * If @dev is expected to be DMA-capable then the bus code that created
···
 	}
 
 	if (!ret) {
-		msb = fls64(dmaaddr + size - 1);
 		/*
-		 * Round-up to the power-of-two mask or set
-		 * the mask to the whole 64-bit address space
-		 * in case the DMA region covers the full
-		 * memory window.
+		 * Limit coherent and dma mask based on size retrieved from
+		 * firmware.
 		 */
-		mask = msb == 64 ? U64_MAX : (1ULL << msb) - 1;
-		/*
-		 * Limit coherent and dma mask based on size
-		 * retrieved from firmware.
-		 */
-		dev->bus_dma_mask = mask;
+		end = dmaaddr + size - 1;
+		mask = DMA_BIT_MASK(ilog2(end) + 1);
+		dev->bus_dma_limit = end;
 		dev->coherent_dma_mask = mask;
 		*dev->dma_mask = mask;
 	}

drivers/ata/ahci.c

···
 	 * value, don't extend it here. This happens on STA2X11, for example.
 	 *
 	 * XXX: manipulating the DMA mask from platform code is completely
-	 * bogus, platform code should use dev->bus_dma_mask instead..
+	 * bogus, platform code should use dev->bus_dma_limit instead..
 	 */
 	if (pdev->dma_mask && pdev->dma_mask < DMA_BIT_MASK(32))
 		return 0;

drivers/iommu/dma-iommu.c

···
 	if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1)))
 		iova_len = roundup_pow_of_two(iova_len);
 
-	if (dev->bus_dma_mask)
-		dma_limit &= dev->bus_dma_mask;
+	dma_limit = min_not_zero(dma_limit, dev->bus_dma_limit);
 
 	if (domain->geometry.force_aperture)
 		dma_limit = min(dma_limit, domain->geometry.aperture_end);

drivers/of/device.c

···
 	bool coherent;
 	unsigned long offset;
 	const struct iommu_ops *iommu;
-	u64 mask;
+	u64 mask, end;
 
 	ret = of_dma_get_range(np, &dma_addr, &paddr, &size);
 	if (ret < 0) {
···
 	 * Limit coherent and dma mask based on size and default mask
 	 * set by the driver.
 	 */
-	mask = DMA_BIT_MASK(ilog2(dma_addr + size - 1) + 1);
+	end = dma_addr + size - 1;
+	mask = DMA_BIT_MASK(ilog2(end) + 1);
 	dev->coherent_dma_mask &= mask;
 	*dev->dma_mask &= mask;
-	/* ...but only set bus mask if we found valid dma-ranges earlier */
+	/* ...but only set bus limit if we found valid dma-ranges earlier */
 	if (!ret)
-		dev->bus_dma_mask = mask;
+		dev->bus_dma_limit = end;
 
 	coherent = of_dma_is_coherent(np);
 	dev_dbg(dev, "device is%sdma coherent\n",

include/linux/device.h

···
  * @coherent_dma_mask: Like dma_mask, but for alloc_coherent mapping as not all
  * 		hardware supports 64-bit addresses for consistent allocations
  * 		such descriptors.
- * @bus_dma_mask: Mask of an upstream bridge or bus which imposes a smaller DMA
- *		limit than the device itself supports.
+ * @bus_dma_limit: Limit of an upstream bridge or bus which imposes a smaller
+ *		DMA limit than the device itself supports.
  * @dma_pfn_offset: offset of DMA memory range relatively of RAM
  * @dma_parms:	A low level driver may set these to teach IOMMU code about
  * 		segment limitations.
···
 					     not all hardware supports
 					     64 bit addresses for consistent
 					     allocations such descriptors. */
-	u64		bus_dma_mask;	/* upstream dma_mask constraint */
+	u64		bus_dma_limit;	/* upstream dma constraint */
 	unsigned long	dma_pfn_offset;
 
 	struct device_dma_parameters *dma_parms;

include/linux/dma-direct.h

···
 	    min(addr, end) < phys_to_dma(dev, PFN_PHYS(min_low_pfn)))
 		return false;
 
-	return end <= min_not_zero(*dev->dma_mask, dev->bus_dma_mask);
+	return end <= min_not_zero(*dev->dma_mask, dev->bus_dma_limit);
 }
 
 u64 dma_direct_get_required_mask(struct device *dev);

include/linux/dma-mapping.h

···
  */
 static inline bool dma_addressing_limited(struct device *dev)
 {
-	return min_not_zero(dma_get_mask(dev), dev->bus_dma_mask) <
+	return min_not_zero(dma_get_mask(dev), dev->bus_dma_limit) <
 			    dma_get_required_mask(dev);
 }
 

kernel/dma/direct.c

···
 {
 	if (!dev->dma_mask) {
 		dev_err_once(dev, "DMA map on device without dma_mask\n");
-	} else if (*dev->dma_mask >= DMA_BIT_MASK(32) || dev->bus_dma_mask) {
+	} else if (*dev->dma_mask >= DMA_BIT_MASK(32) || dev->bus_dma_limit) {
 		dev_err_once(dev,
-			"overflow %pad+%zu of DMA mask %llx bus mask %llx\n",
-			&dma_addr, size, *dev->dma_mask, dev->bus_dma_mask);
+			"overflow %pad+%zu of DMA mask %llx bus limit %llx\n",
+			&dma_addr, size, *dev->dma_mask, dev->bus_dma_limit);
 	}
 	WARN_ON_ONCE(1);
 }
···
 }
 
 static gfp_t __dma_direct_optimal_gfp_mask(struct device *dev, u64 dma_mask,
-		u64 *phys_mask)
+		u64 *phys_limit)
 {
-	if (dev->bus_dma_mask && dev->bus_dma_mask < dma_mask)
-		dma_mask = dev->bus_dma_mask;
+	u64 dma_limit = min_not_zero(dma_mask, dev->bus_dma_limit);
 
 	if (force_dma_unencrypted(dev))
-		*phys_mask = __dma_to_phys(dev, dma_mask);
+		*phys_limit = __dma_to_phys(dev, dma_limit);
 	else
-		*phys_mask = dma_to_phys(dev, dma_mask);
+		*phys_limit = dma_to_phys(dev, dma_limit);
 
 	/*
 	 * Optimistically try the zone that the physical address mask falls
···
 	 * Note that GFP_DMA32 and GFP_DMA are no ops without the corresponding
 	 * zones.
 	 */
-	if (*phys_mask <= DMA_BIT_MASK(zone_dma_bits))
+	if (*phys_limit <= DMA_BIT_MASK(zone_dma_bits))
 		return GFP_DMA;
-	if (*phys_mask <= DMA_BIT_MASK(32))
+	if (*phys_limit <= DMA_BIT_MASK(32))
 		return GFP_DMA32;
 	return 0;
 }
···
 static bool dma_coherent_ok(struct device *dev, phys_addr_t phys, size_t size)
 {
 	return phys_to_dma_direct(dev, phys) + size - 1 <=
-			min_not_zero(dev->coherent_dma_mask, dev->bus_dma_mask);
+			min_not_zero(dev->coherent_dma_mask, dev->bus_dma_limit);
 }
 
 struct page *__dma_direct_alloc_pages(struct device *dev, size_t size,
···
 	size_t alloc_size = PAGE_ALIGN(size);
 	int node = dev_to_node(dev);
 	struct page *page = NULL;
-	u64 phys_mask;
+	u64 phys_limit;
 
 	if (attrs & DMA_ATTR_NO_WARN)
 		gfp |= __GFP_NOWARN;
···
 	/* we always manually zero the memory once we are done: */
 	gfp &= ~__GFP_ZERO;
 	gfp |= __dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
-			&phys_mask);
+			&phys_limit);
 	page = dma_alloc_contiguous(dev, alloc_size, gfp);
 	if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
 		dma_free_contiguous(dev, page, alloc_size);
···
 		page = NULL;
 
 		if (IS_ENABLED(CONFIG_ZONE_DMA32) &&
-		    phys_mask < DMA_BIT_MASK(64) &&
+		    phys_limit < DMA_BIT_MASK(64) &&
 		    !(gfp & (GFP_DMA32 | GFP_DMA))) {
 			gfp |= GFP_DMA32;
 			goto again;