support range based offsets in dma-direct
Subsystems (24)
Hi all,
this series adds range-based offsets to the dma-direct implementation. The guts of the change are a patch from Jim with some modifications from me, but to do it nicely we need to ARM patches to prepare for it as well.
Diffstat:
arch/arm/common/dmabounce.c | 2 arch/arm/include/asm/dma-direct.h | 69 +++++++++++++++++ arch/arm/include/asm/dma-mapping.h | 70 ------------------ arch/arm/mach-keystone/keystone.c | 21 +++-- arch/sh/drivers/pci/pcie-sh7786.c | 9 +- arch/x86/pci/sta2x11-fixup.c | 6 + drivers/acpi/arm64/iort.c | 6 + drivers/base/core.c | 2 drivers/gpu/drm/sun4i/sun4i_backend.c | 8 +- drivers/iommu/io-pgtable-arm.c | 2 drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c | 9 ++ drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c | 11 ++ drivers/of/address.c | 73 ++++++++---------- drivers/of/device.c | 44 +++++++---- drivers/of/of_private.h | 11 +- drivers/of/unittest.c | 34 ++++++-- drivers/remoteproc/remoteproc_core.c | 4 - drivers/staging/media/sunxi/cedrus/cedrus_hw.c | 10 ++ drivers/usb/core/message.c | 5 - drivers/usb/core/usb.c | 3 include/linux/device.h | 4 - include/linux/dma-direct.h | 52 +++++++++++-- include/linux/dma-mapping.h | 19 ++++ kernel/dma/coherent.c | 7 - kernel/dma/direct.c | 81 ++++++++++++++++++++- 25 files changed, 373 insertions(+), 189 deletions(-)
PATCH 1/3 ARM/dma-mapping: move various helpers from dma-mapping.h to dma-direct.h
Move the helpers to translate to and from direct mapping DMA addresses to dma-direct.h. This not only is the most logical place, but the new placement also avoids dependency loops with pending commits.
--- arch/arm/common/dmabounce.c | 2 +- arch/arm/include/asm/dma-direct.h | 70 ++++++++++++++++++++++++++++++ arch/arm/include/asm/dma-mapping.h | 70 ------------------------------ 3 files changed, 71 insertions(+), 71 deletions(-)
arch/arm/common/dmabounce.c +1 -1
@@ -24,7 +24,7 @@#include <linux/slab.h> #include <linux/page-flags.h> #include <linux/device.h> -#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> #include <linux/dmapool.h> #include <linux/list.h> #include <linux/scatterlist.h>
arch/arm/include/asm/dma-direct.h +70 -0
76 lines not on this page · raw message
arch/arm/include/asm/dma-mapping.h +0 -70
82 lines not on this page · raw message
Signature · 1 line
2.28.0
Re: [PATCH 1/3] ARM/dma-mapping: move various helpers from dma-mapping.h to dma-direct.h
Quoted 3 lines · Christoph Hellwig
Move the helpers to translate to and from direct mapping DMA addresses to
dma-direct.h. This not only is the most logical place, but the new placement also avoids dependency loops with pending commits.
For the straightforward move as it should be,
However I do wonder how much of this could be cleaned up further...
Quoted 25 lines
Signed-off-by Christoph Hellwig <[email protected]>--- arch/arm/common/dmabounce.c | 2 +- arch/arm/include/asm/dma-direct.h | 70 ++++++++++++++++++++++++++++++ arch/arm/include/asm/dma-mapping.h | 70 ------------------------------ 3 files changed, 71 insertions(+), 71 deletions(-)arch/arm/common/dmabounce.c +1
-1@@ -24,7 +24,7 @@#include <linux/slab.h> #include <linux/page-flags.h> #include <linux/device.h> -#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> #include <linux/dmapool.h> #include <linux/list.h> #include <linux/scatterlist.h> arch/arm/include/asm/dma-direct.h +0
-0
arch/arm/include/asm/dma-direct.h line 9
Quoted 8 lines
arch/arm/include/asm/dma-direct.h +5
-0@@ -2,6 +2,76 @@#ifndef ASM_ARM_DMA_DIRECT_H #define ASM_ARM_DMA_DIRECT_H 1 +#include <asm/memory.h> + +#ifdef __arch_page_to_dma +#error Please update to __arch_pfn_to_dma +#endif
This must be long, long dead by now.
Quoted 13 lines
arch/arm/include/asm/dma-direct.h +13
-0+ +/* +* dma_to_pfn/pfn_to_dma/dma_to_virt/virt_to_dma are architecture private +* functions used internally by the DMA-mapping API to provide DMA +* addresses. They must not be used by drivers. +*/ +#ifndef __arch_pfn_to_dma +static inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) +{ +if (dev) + pfn -= dev->dma_pfn_offset; +return (dma_addr_t )__pfn_to_bus (pfn); +}
arch/arm/include/asm/dma-direct.h line 32
Quoted 10 lines
arch/arm/include/asm/dma-direct.h +10
-0+ +static inline unsigned long dma_to_pfn (struct device *dev,dma_addr_t addr) +{ +unsigned long pfn =__bus_to_pfn (addr); + +if (dev) + pfn += dev->dma_pfn_offset; + +return pfn; +}
These are only overridden for OMAP1510, and it looks like it wouldn't take much for the platform code or ohci-omap driver to set up a generic DMA offset for the relevant device.
arch/arm/include/asm/dma-direct.h line 43
Quoted 11 lines
arch/arm/include/asm/dma-direct.h +11
-0+ +static inline void *dma_to_virt (struct device *dev,dma_addr_t addr) +{ +if (dev) { +unsigned long pfn =dma_to_pfn (dev, addr); + +return phys_to_virt (__pfn_to_phys (pfn)); + } + +return (void *)__bus_to_virt ((unsigned long )addr); +}
This appears entirely unused.
arch/arm/include/asm/dma-direct.h line 51
Quoted 8 lines
arch/arm/include/asm/dma-direct.h +8
-0+ +static inline dma_addr_t virt_to_dma (struct device *dev,void *addr) +{ +if (dev) +return pfn_to_dma (dev,virt_to_pfn (addr)); + +return (dma_addr_t )__virt_to_bus ((unsigned long )(addr)); +}
And this is only used for some debug prints in dmabounce.
Similarly the __bus_to_*()/__*_to_bus() calls themselves only appear significant to mach-footbridge any more, and could probably also be evolved into regular DMA offsets now that all API calls must have a non-NULL device. I think I might come back and take a closer look at all this at some point in future... :)
Robin.
Quoted 115 lines
arch/arm/include/asm/dma-direct.h +23
-0+ +#else +static inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) +{ +return __arch_pfn_to_dma (dev, pfn); +} + +static inline unsigned long dma_to_pfn (struct device *dev,dma_addr_t addr) +{ +return __arch_dma_to_pfn (dev, addr); +} + +static inline void *dma_to_virt (struct device *dev,dma_addr_t addr) +{ +return __arch_dma_to_virt (dev, addr); +} + +static inline dma_addr_t virt_to_dma (struct device *dev,void *addr) +{ +return __arch_virt_to_dma (dev, addr); +} +#endif +static inline dma_addr_t __phys_to_dma (struct device *dev,phys_addr_t paddr) {unsigned int offset = paddr & ~PAGE_MASK ;arch/arm/include/asm/dma-mapping.h +0
-70@@ -8,8 +8,6 @@#include <linux/scatterlist.h> #include <linux/dma-debug.h> -#include <asm/memory.h> -#include <xen/xen.h> #include <asm/xen/hypervisor.h> @@ -23,74 +21,6 @@ static inline const struct dma_map_ops *get_arch_dma_ops(struct bus_type *bus)return NULL ; } -#ifdef __arch_page_to_dma -#error Please update to __arch_pfn_to_dma -#endif - -/* -* dma_to_pfn/pfn_to_dma/dma_to_virt/virt_to_dma are architecture private -* functions used internally by the DMA-mapping API to provide DMA -* addresses. They must not be used by drivers. -*/ -#ifndef __arch_pfn_to_dma -static inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) -{ -if (dev) - pfn -= dev->dma_pfn_offset; -return (dma_addr_t )__pfn_to_bus (pfn); -} - -static inline unsigned long dma_to_pfn (struct device *dev,dma_addr_t addr) -{ -unsigned long pfn =__bus_to_pfn (addr); - -if (dev) - pfn += dev->dma_pfn_offset; - -return pfn; -} - -static inline void *dma_to_virt (struct device *dev,dma_addr_t addr) -{ -if (dev) { -unsigned long pfn =dma_to_pfn (dev, addr); - -return phys_to_virt (__pfn_to_phys (pfn)); - } - -return (void *)__bus_to_virt ((unsigned long )addr); -} - -static inline dma_addr_t virt_to_dma (struct device *dev,void *addr) -{ -if (dev) -return pfn_to_dma (dev,virt_to_pfn (addr)); - -return (dma_addr_t )__virt_to_bus ((unsigned long )(addr)); -} - -#else -static inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) -{ -return __arch_pfn_to_dma (dev, pfn); -} - -static inline unsigned long dma_to_pfn (struct device *dev,dma_addr_t addr) -{ -return __arch_dma_to_pfn (dev, addr); -} - -static inline void *dma_to_virt (struct device *dev,dma_addr_t addr) -{ -return __arch_dma_to_virt (dev, addr); -} - -static inline dma_addr_t virt_to_dma (struct device *dev,void *addr) -{ -return __arch_virt_to_dma (dev, addr); -} -#endif -/** * arm_dma_alloc - allocate consistent memory for DMA * @dev: valid struct device pointer, or NULL for ISA and EISA-like devices
Re: [PATCH 1/3] ARM/dma-mapping: move various helpers from dma-mapping.h to dma-direct.h
Quoted 9 lines · Robin Murphy
Quoted 3 lines · Christoph Hellwig
Move the helpers to translate to and from direct mapping DMA addresses to
dma-direct.h. This not only is the most logical place, but the new placement also avoids dependency loops with pending commits.For the straightforward move as it should be,
Reviewed-by Robin Murphy <[email protected]>
Quoted 7 lines
However I do wonder how much of this could be cleaned up further...
arch/arm/include/asm/dma-direct.h line 9
Quoted 4 lines
arch/arm/include/asm/dma-direct.h +4
-0+ +#ifdef __arch_page_to_dma +#error Please update to __arch_pfn_to_dma +#endif This must be long, long dead by now.
Yeah. I had a patch to remove this which lead me into the rabbit hole your described later. A few patches in I decided to give up and just do the trivial move. But it probably makes sense to pick up at least the two trivial dead code removal patches..
Quoted 5 lines
Quoted 5 lines
arch/arm/include/asm/dma-direct.h +5
-0+static inline unsigned long dma_to_pfn (struct device *dev,dma_addr_t addr) +{ +unsigned long pfn =__bus_to_pfn (addr); + +if (dev)
Quoted 8 lines
arch/arm/include/asm/dma-direct.h line 32
Quoted 4 lines
arch/arm/include/asm/dma-direct.h +4
-0+ pfn += dev->dma_pfn_offset; + +return pfn; +}These are only overridden for OMAP1510, and it looks like it wouldn't take much for the platform code or ohci-omap driver to set up a generic DMA offset for the relevant device.
I sent a ping to the omap maintainers earlier this week to ask for that :)
Quoted 10 lines
arch/arm/include/asm/dma-direct.h line 51
Quoted 7 lines
arch/arm/include/asm/dma-direct.h +7
-0+static inline dma_addr_t virt_to_dma (struct device *dev,void *addr) +{ +if (dev) +return pfn_to_dma (dev,virt_to_pfn (addr)); + +return (dma_addr_t )__virt_to_bus ((unsigned long )(addr)); +}And this is only used for some debug prints in dmabounce.
Quoted 5 lines
Similarly the __bus_to_*()/__*_to_bus() calls themselves only appear significant to mach-footbridge any more, and could probably also be evolved into regular DMA offsets now that all API calls must have a non-NULL device. I think I might come back and take a closer look at all this at some point in future... :)
Yes, pretty much all of this should eventually go away. I just don't want to bock the ranges work on all kinds of random arm cleanups..
PATCH 2/3 ARM/keystone: move the DMA offset handling under ifdef CONFIG_ARM_LPAE
In reply to Christoph HellwigThe DMA offset notifier can only be used if PHYS_OFFSET is at least KEYSTONE_HIGH_PHYS_START, which can't be represented by a 32-bit phys_addr_t. Currently the code compiles fine despite that, a pending change to the DMA offset handling would create a compiler warning for this case. Add an ifdef to not compile the code except for LPAE configs.
--- arch/arm/mach-keystone/keystone.c | 4 ++++ 1 file changed, 4 insertions(+)
arch/arm/mach-keystone/keystone.c +4 -0
@@ -24,6 +24,7 @@#include "keystone.h" +#ifdef CONFIG_ARM_LPAE static unsigned long keystone_dma_pfn_offset__read_mostly ; static int keystone_platform_notifier(struct notifier_block *nb,
@@ -48,14 +49,17 @@ static int keystone_platform_notifier(struct notifier_block *nb,static struct notifier_block platform_nb = { .notifier_call = keystone_platform_notifier, }; +#endif /* CONFIG_ARM_LPAE */ static void __init keystone_init (void ) { +#ifdef CONFIG_ARM_LPAE if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START ) { keystone_dma_pfn_offset =PFN_DOWN (KEYSTONE_HIGH_PHYS_START -KEYSTONE_LOW_PHYS_START );bus_register_notifier (&platform_bus_type, &platform_nb); } +#endif keystone_pm_runtime_init (); }
Signature · 1 line
2.28.0
Re: [PATCH 2/3] ARM/keystone: move the DMA offset handling under ifdef CONFIG_ARM_LPAE
Quoted 6 lines · Christoph Hellwig
The DMA offset notifier can only be used if
PHYS_OFFSETis at leastKEYSTONE_HIGH_PHYS_START, which can't be represented by a 32-bitphys_addr_t. Currently the code compiles fine despite that, a pending change to the DMA offset handling would create a compiler warning for this case. Add an ifdef to not compile the code except for LPAE configs.
Seems reasonable - once again I wonder whether this notifier is really needed any more since "dma-ranges" should now be handled properly for PCI devices as well, but that's not something to worry about in this series.
Quoted 36 lines
Signed-off-by Christoph Hellwig <[email protected]>--- arch/arm/mach-keystone/keystone.c | 4 ++++ 1 file changed, 4 insertions(+)arch/arm/mach-keystone/keystone.c +4
-0@@ -24,6 +24,7 @@#include "keystone.h" +#ifdef CONFIG_ARM_LPAE static unsigned long keystone_dma_pfn_offset__read_mostly ; static int keystone_platform_notifier(struct notifier_block *nb,@@ -48,14 +49,17 @@ static int keystone_platform_notifier(struct notifier_block *nb,static struct notifier_block platform_nb = { .notifier_call = keystone_platform_notifier, }; +#endif /* CONFIG_ARM_LPAE */ static void __init keystone_init (void ) { +#ifdef CONFIG_ARM_LPAE if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START ) { keystone_dma_pfn_offset =PFN_DOWN (KEYSTONE_HIGH_PHYS_START -KEYSTONE_LOW_PHYS_START );bus_register_notifier (&platform_bus_type, &platform_nb); } +#endif keystone_pm_runtime_init (); }
Russell King - ARM Linux admin
maintainerRe: [PATCH 2/3] ARM/keystone: move the DMA offset handling under ifdef CONFIG_ARM_LPAE
In reply to Christoph HellwigQuoted 6 lines · Christoph Hellwig
The DMA offset notifier can only be used if
PHYS_OFFSETis at leastKEYSTONE_HIGH_PHYS_START, which can't be represented by a 32-bitphys_addr_t. Currently the code compiles fine despite that, a pending change to the DMA offset handling would create a compiler warning for this case. Add an ifdef to not compile the code except for LPAE configs.
However, to have use of the high physical offset, LPAE needs to be enabled, which ensures that phys_addr_t is 64-bit.
I believe that DMA is non-coherent on this platform unless the high physical address is used. Or something like that.
Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
Robin Murphy
reviewerRe: [PATCH 2/3] ARM/keystone: move the DMA offset handling under ifdef CONFIG_ARM_LPAE
Quoted 8 lines · Russell King - ARM Linux admin
Quoted 6 lines · Christoph Hellwig
The DMA offset notifier can only be used if
PHYS_OFFSETis at leastKEYSTONE_HIGH_PHYS_START, which can't be represented by a 32-bitphys_addr_t. Currently the code compiles fine despite that, a pending change to the DMA offset handling would create a compiler warning for this case. Add an ifdef to not compile the code except for LPAE configs.
Quoted 5 lines
However, to have use of the high physical offset, LPAE needs to be enabled, which ensures that
phys_addr_tis 64-bit.I believe that DMA is non-coherent on this platform unless the high physical address is used. Or something like that.
Yeah, it's probably not a configuration that anyone would actually want to use in anger on Keystone itself, but as long as folks might have ARCH_KEYSTONE selected in their non-LPAE multiplatform config we should avoid build regressions. I did wonder if Keystone should simply have a hard dependency on LPAE, but there does appear to be some explicit support for running directly from the non-coherent low 2G alias :/
Robin.
Re: [PATCH 2/3] ARM/keystone: move the DMA offset handling under ifdef CONFIG_ARM_LPAE
In reply to Russell King - ARM Linux adminQuoted 8 lines · Russell King - ARM Linux admin
Quoted 6 lines · Christoph Hellwig
The DMA offset notifier can only be used if
PHYS_OFFSETis at leastKEYSTONE_HIGH_PHYS_START, which can't be represented by a 32-bitphys_addr_t. Currently the code compiles fine despite that, a pending change to the DMA offset handling would create a compiler warning for this case. Add an ifdef to not compile the code except for LPAE configs.
Quoted 6 lines
However, to have use of the high physical offset, LPAE needs to be enabled, which ensures that
phys_addr_tis 64-bit.I believe that DMA is non-coherent on this platform unless the high physical address is used. Or something like that.
Exactly. Higher address ranges needs to be used for DMA coherency.
Regards,
Santosh
PATCH 3/3 dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
In reply to Christoph HellwigFrom: Jim Quinlan <[email protected]>
The new field 'dma_range_map' in struct device is used to facilitate the use of single or multiple offsets between mapping regions of cpu addrs and dma addrs. It subsumes the role of "dev->dma_pfn_offset" which was only capable of holding a single uniform offset and had no region bounds checking.
The function of_dma_get_range() has been modified so that it takes a single argument -- the device node -- and returns a map, NULL, or an error code. The map is an array that holds the information regarding the DMA regions. Each range entry contains the address offset, the cpu_start address, the dma_start address, and the size of the region.
of_dma_configure() is the typical manner to set range offsets but there are a number of ad hoc assignments to "dev->dma_pfn_offset" in the kernel driver code. These cases now invoke the function dma_attach_offset_range(dev, cpu_addr, dma_addr, size).
--- arch/arm/include/asm/dma-direct.h | 9 +-- arch/arm/mach-keystone/keystone.c | 17 ++-- arch/sh/drivers/pci/pcie-sh7786.c | 9 ++- arch/x86/pci/sta2x11-fixup.c | 6 +- drivers/acpi/arm64/iort.c | 6 +- drivers/base/core.c | 2 + drivers/gpu/drm/sun4i/sun4i_backend.c | 8 +- drivers/iommu/io-pgtable-arm.c | 2 +- .../platform/sunxi/sun4i-csi/sun4i_csi.c | 9 ++- .../platform/sunxi/sun6i-csi/sun6i_csi.c | 11 ++- drivers/of/address.c | 73 ++++++++--------- drivers/of/device.c | 44 ++++++---- drivers/of/of_private.h | 11 +-- drivers/of/unittest.c | 34 +++++--- drivers/remoteproc/remoteproc_core.c | 4 +- .../staging/media/sunxi/cedrus/cedrus_hw.c | 10 ++- drivers/usb/core/message.c | 5 +- drivers/usb/core/usb.c | 3 +- include/linux/device.h | 4 +- include/linux/dma-direct.h | 52 ++++++++++-- include/linux/dma-mapping.h | 19 ++++- kernel/dma/coherent.c | 7 +- kernel/dma/direct.c | 81 ++++++++++++++++++- 23 files changed, 303 insertions(+), 123 deletions(-)
arch/arm/include/asm/dma-direct.h +4 -5
@@ -16,8 +16,8 @@#ifndef __arch_pfn_to_dmastatic inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) { -if (dev) - pfn -= dev->dma_pfn_offset; +if (dev && dev->dma_range_map) + pfn =PFN_DOWN (translate_phys_to_dma (dev,PFN_PHYS (pfn)));return (dma_addr_t )__pfn_to_bus (pfn); }
@@ -25,9 +25,8 @@ static inline unsigned long dma_to_pfn(struct device *dev, dma_addr_t addr) {unsigned long pfn =__bus_to_pfn (addr); -if (dev) - pfn += dev->dma_pfn_offset; - +if (dev && dev->dma_range_map) + pfn =PFN_DOWN (translate_dma_to_phys (dev,PFN_PHYS (pfn)));return pfn; }
arch/arm/mach-keystone/keystone.c +8 -9
@@ -8,6 +8,7 @@*/ #include <linux/io.h> #include <linux/of.h> +#include <linux/dma-mapping.h> #include <linux/init.h> #include <linux/of_platform.h> #include <linux/of_address.h>
@@ -25,8 +26,6 @@#include "keystone.h" #ifdef CONFIG_ARM_LPAE -static unsigned long keystone_dma_pfn_offset__read_mostly ; -static int keystone_platform_notifier (struct notifier_block *nb,unsigned long event,void *data) {
@@ -39,9 +38,12 @@ static int keystone_platform_notifier(struct notifier_block *nb,return NOTIFY_BAD ;if (!dev->of_node) { - dev->dma_pfn_offset = keystone_dma_pfn_offset; -dev_err (dev,"set dma_pfn_offset%08lx\n" , - dev->dma_pfn_offset); +int ret =dma_direct_set_offset (dev,KEYSTONE_HIGH_PHYS_START , +KEYSTONE_LOW_PHYS_START , +KEYSTONE_HIGH_PHYS_SIZE ); +dev_err (dev,"set dma_offset%08llx%s\n" , +KEYSTONE_HIGH_PHYS_START -KEYSTONE_LOW_PHYS_START , + ret ?" failed" :"" ); }return NOTIFY_OK ; }
@@ -54,11 +56,8 @@ static struct notifier_block platform_nb = {static void __init keystone_init (void ) {#ifdef CONFIG_ARM_LPAE -if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START ) { - keystone_dma_pfn_offset =PFN_DOWN (KEYSTONE_HIGH_PHYS_START - -KEYSTONE_LOW_PHYS_START ); +if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START )bus_register_notifier (&platform_bus_type, &platform_nb); - }#endif keystone_pm_runtime_init (); }
arch/sh/drivers/pci/pcie-sh7786.c +5 -4
@@ -12,6 +12,7 @@#include <linux/io.h> #include <linux/async.h> #include <linux/delay.h> +#include <linux/dma-mapping.h> #include <linux/slab.h> #include <linux/clk.h> #include <linux/sh_clk.h>
@@ -31,6 +32,8 @@ struct sh7786_pcie_port {static struct sh7786_pcie_port *sh7786_pcie_ports;static unsigned int nr_ports;static unsigned long dma_pfn_offset; +size_t memsize; +u64 memstart; static struct sh7786_pcie_hwops { int (*core_init)(void);
@@ -301,7 +304,6 @@ static int __init pcie_init(struct sh7786_pcie_port *port)struct pci_channel *chan = port->hose;unsigned int data;phys_addr_t memstart, memend; -size_t memsize;int ret, i, win;/* Begin initialization */
@@ -368,8 +370,6 @@ static int __init pcie_init(struct sh7786_pcie_port *port) memstart =ALIGN_DOWN (memstart, memsize); memsize =roundup_pow_of_two (memend - memstart); - dma_pfn_offset = memstart >>PAGE_SHIFT ; -/* * If there's more than 512MB of memory, we need to roll over to * LAR1/LAMR1.
@@ -487,7 +487,8 @@ int pcibios_map_platform_irq(const struct pci_dev *pdev, u8 slot, u8 pin)void pcibios_bus_add_device (struct pci_dev *pdev) { - pdev->dev.dma_pfn_offset = dma_pfn_offset; +dma_direct_set_offset (&pdev->dev,__pa (memory_start), +__pa (memory_start) - memstart, memsize); }static int __init sh7786_pcie_core_init (void )
arch/x86/pci/sta2x11-fixup.c +4 -2
@@ -133,7 +133,7 @@ static void sta2x11_map_ep(struct pci_dev *pdev)struct sta2x11_instance *instance =sta2x11_pdev_to_instance (pdev);struct device *dev = &pdev->dev;u32 amba_base, max_amba_addr; -int i; +int i, ret;if (!instance)return ;
@@ -141,7 +141,9 @@ static void sta2x11_map_ep(struct pci_dev *pdev)pci_read_config_dword (pdev,AHB_BASE (0 ), &amba_base); max_amba_addr = amba_base +STA2X11_AMBA_SIZE -1 ; - dev->dma_pfn_offset =PFN_DOWN (-amba_base); + ret =dma_direct_set_offset (dev,0 , amba_base,STA2X11_AMBA_SIZE ); +if (ret) +dev_err (dev,"sta2x11: could not set DMA offset\n" ); dev->bus_dma_limit = max_amba_addr;pci_set_consistent_dma_mask (pdev, max_amba_addr);
drivers/acpi/arm64/iort.c +4 -2
@@ -18,6 +18,7 @@#include <linux/pci.h> #include <linux/platform_device.h> #include <linux/slab.h> +#include <linux/dma-mapping.h> #define IORT_TYPE_MASK (type) (1 << (type))#define IORT_MSI_TYPE (1 << ACPI_IORT_NODE_ITS_GROUP)
@@ -1184,8 +1185,9 @@ void iort_dma_setup(struct device *dev, u64 *dma_addr, u64 *dma_size) *dma_addr = dmaaddr; *dma_size = size; - dev->dma_pfn_offset =PFN_DOWN (offset); -dev_dbg (dev,"dma_pfn_offset(%#08llx)\n" , offset); + ret =dma_direct_set_offset (dev, dmaaddr + offset, dmaaddr, size); + +dev_dbg (dev,"dma_offset(%#08llx)%s\n" , offset, ret ?" failed!" :"" ); } static void __init acpi_iort_register_irq(int hwirq, const char *name,
drivers/base/core.c +2 -0
@@ -1792,6 +1792,8 @@ static void device_release(struct kobject *kobj)*/ devres_release_all (dev); +kfree (dev->dma_range_map); +if (dev->release) dev->release (dev);else if (dev->type && dev->type->release)
drivers/gpu/drm/sun4i/sun4i_backend.c +7 -1
@@ -11,6 +11,7 @@#include <linux/module.h> #include <linux/of_device.h> #include <linux/of_graph.h> +#include <linux/dma-mapping.h> #include <linux/platform_device.h> #include <linux/reset.h>
@@ -811,8 +812,13 @@ static int sun4i_backend_bind(struct device *dev, struct device *master,* because of an old DT, we need to set the DMA offset by hand * on our device since the RAM mapping is at 0 for the DMA bus, * unlike the CPU. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. */ - drm->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; + ret =dma_direct_set_offset (drm->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; } backend->engine.node = dev->of_node;
drivers/iommu/io-pgtable-arm.c +1 -1
@@ -751,7 +751,7 @@ arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg)if (cfg->oas >ARM_LPAE_MAX_ADDR_BITS )return NULL ; -if (!selftest_running && cfg->iommu_dev->dma_pfn_offset) { +if (!selftest_running && cfg->iommu_dev->dma_range_map) {dev_err (cfg->iommu_dev,"Cannot accommodate DMA offset for IOMMU page tables\n" );return NULL ; }
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c +8 -1
@@ -7,6 +7,7 @@*/ #include <linux/clk.h> +#include <linux/dma-mapping.h> #include <linux/interrupt.h> #include <linux/module.h> #include <linux/mutex.h>
@@ -182,8 +183,14 @@ static int sun4i_csi_probe(struct platform_device *pdev)if (ret)return ret; }else { +/* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. +*/ #ifdef PHYS_PFN_OFFSET - csi->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; + ret =dma_direct_set_offset (csi->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret;#endif }
drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c +9 -2
@@ -899,8 +899,15 @@ static int sun6i_csi_probe(struct platform_device *pdev)return -ENOMEM ; sdev->dev = &pdev->dev; -/* The DMA bus has the memory mapped at 0 */ - sdev->dev->dma_pfn_offset =PHYS_OFFSET >>PAGE_SHIFT ; +/* +* The DMA bus has the memory mapped at 0. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. +*/ + ret =dma_direct_set_offset (sdev->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; ret =sun6i_csi_resource_request (sdev, pdev);if (ret)
drivers/of/address.c +32 -41
120 lines not on this page · raw message
drivers/of/device.c +27 -17
101 lines not on this page · raw message
drivers/of/of_private.h +6 -5
@@ -157,12 +157,13 @@ extern void __of_sysfs_remove_bin_file(struct device_node *np,extern int of_bus_n_addr_cells (struct device_node *np);extern int of_bus_n_size_cells (struct device_node *np); -#ifdef CONFIG_OF_ADDRESS -extern int of_dma_get_range (struct device_node *np,u64 *dma_addr, -u64 *paddr,u64 *size); +struct bus_dma_region ; +#if defined(CONFIG_OF_ADDRESS ) && defined(CONFIG_HAS_DMA ) +int of_dma_get_range (struct device_node *np, +const struct bus_dma_region **map);#else -static inline int of_dma_get_range (struct device_node *np,u64 *dma_addr, -u64 *paddr,u64 *size) +static inline int of_dma_get_range (struct device_node *np, +const struct bus_dma_region **map) {return -ENODEV ; }
drivers/of/unittest.c +24 -10
@@ -7,6 +7,7 @@#include <linux/memblock.h> #include <linux/clk.h> +#include <linux/dma-direct.h> /* to test phys_to_dma/dma_to_phys */ #include <linux/err.h> #include <linux/errno.h> #include <linux/hashtable.h>
@@ -869,10 +870,11 @@ static void __init of_unittest_changeset(void) }static void __init of_unittest_dma_ranges_one (const char *path, -u64 expect_dma_addr,u64 expect_paddr,u64 expect_size) +u64 expect_dma_addr,u64 expect_paddr) { +#ifdef CONFIG_HAS_DMA struct device_node *np; -u64 dma_addr, paddr, size; +const struct bus_dma_region *map =NULL ;int rc; np =of_find_node_by_path (path);
@@ -881,28 +883,40 @@ static void __init of_unittest_dma_ranges_one(const char *path,return ; } - rc =of_dma_get_range (np, &dma_addr, &paddr, &size); + rc =of_dma_get_range (np, &map);unittest (!rc,"of_dma_get_range failed on node %pOF rc=%i\n" , np, rc); +if (!rc) { -unittest (size = expect_size, -"of_dma_get_range wrong size on node %pOF size=%llx\n" , np, size); +phys_addr_t paddr; +dma_addr_t dma_addr; +struct device dev_bogus; + + dev_bogus.dma_range_map = map; + paddr =dma_to_phys (&dev_bogus, expect_dma_addr); + dma_addr =phys_to_dma (&dev_bogus, expect_paddr); +unittest (paddr = expect_paddr, -"of_dma_get_range wrong phys addr (%llx) on node %pOF" , paddr, np); +"of_dma_get_range: wrong phys addr %pap (expecting %llx) on node %pOF\n" , + &paddr, expect_paddr, np);unittest (dma_addr = expect_dma_addr, -"of_dma_get_range wrong DMA addr (%llx) on node %pOF" , dma_addr, np); +"of_dma_get_range: wrong DMA addr %pad (expecting %llx) on node %pOF\n" , + &dma_addr, expect_dma_addr, np); + +kfree (map); }of_node_put (np); +#endif }static void __init of_unittest_parse_dma_ranges (void ) {of_unittest_dma_ranges_one ("/testcase-data/address-tests/device@70000000" , -0x0 ,0x20000000 ,0x40000000 ); +0x0 ,0x20000000 );of_unittest_dma_ranges_one ("/testcase-data/address-tests/bus@80000000/device@1000" , -0x100000000 ,0x20000000 ,0x2000000000 ); +0x100000000 ,0x20000000 );of_unittest_dma_ranges_one ("/testcase-data/address-tests/pci@90000000" , -0x80000000 ,0x20000000 ,0x10000000 ); +0x80000000 ,0x20000000 ); }static void __init of_unittest_pci_dma_ranges (void )
drivers/remoteproc/remoteproc_core.c +3 -1
@@ -529,7 +529,9 @@ static int rproc_handle_vdev(struct rproc *rproc, struct fw_rsc_vdev *rsc,/* Initialise vdev subdevice */ snprintf (name,sizeof (name),"vdev%dbuffer" , rvdev->index); rvdev->dev.parent = &rproc->dev; - rvdev->dev.dma_pfn_offset = rproc->dev.parent->dma_pfn_offset; + ret =dma_direct_copy_range_map (&rvdev->dev, rproc->dev.parent); +if (ret) +return ret; rvdev->dev.release = rproc_rvdev_release;dev_set_name (&rvdev->dev,"%s#%s" ,dev_name (rvdev->dev.parent), name);dev_set_drvdata (&rvdev->dev, rvdev);
drivers/staging/media/sunxi/cedrus/cedrus_hw.c +8 -2
@@ -227,11 +227,17 @@ int cedrus_hw_probe(struct cedrus_dev *dev)* the RAM offset to the physcal addresses. * * This information will eventually be obtained from device-tree. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. */ #ifdef PHYS_PFN_OFFSET -if (!(variant->quirks &CEDRUS_QUIRK_NO_DMA_OFFSET )) - dev->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; +if (!(variant->quirks &CEDRUS_QUIRK_NO_DMA_OFFSET )) { + ret =dma_direct_set_offset (dev->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; + }#endif ret =of_reserved_mem_device_init (dev->dev);
drivers/usb/core/message.c +3 -2
@@ -1956,10 +1956,11 @@ int usb_set_configuration(struct usb_device *dev, int configuration) intf->dev.groups = usb_interface_groups;/* * Please refer to usb_alloc_dev() to see why we set -* dma_mask and dma_pfn_offset. +* dma_mask and dma_range_map. */ intf->dev.dma_mask = dev->dev.dma_mask; - intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; +if (dma_direct_copy_range_map (&intf->dev, &dev->dev)) +dev_err (&dev->dev,"failed to copy DMA map\n" );INIT_WORK (&intf->reset_ws, __usb_queue_reset_device); intf->minor =-1 ;device_initialize (&intf->dev);
drivers/usb/core/usb.c +2 -1
@@ -610,7 +610,8 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,* mask for the entire HCD, so don't do that. */ dev->dev.dma_mask = bus->sysdev->dma_mask; - dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; +if (dma_direct_copy_range_map (&dev->dev, bus->sysdev)) +dev_err (&dev->dev,"failed to copy DMA map\n" );set_dev_node (&dev->dev,dev_to_node (bus->sysdev)); dev->state =USB_STATE_ATTACHED ; dev->lpm_disable_count =1 ;
include/linux/device.h +2 -2
@@ -466,7 +466,7 @@ struct dev_links_info {* such descriptors. * @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_range_map: map for DMA memory ranges relative to that of RAM * @dma_parms: A low level driver may set these to teach IOMMU code about * segment limitations. * @dma_pools: Dma pools (if dma'ble device).
@@ -561,7 +561,7 @@ struct device {64 bit addresses for consistent allocations such descriptors. */ u64 bus_dma_limit;/* upstream dma constraint */ -unsigned long dma_pfn_offset; +const struct bus_dma_region *dma_range_map;struct device_dma_parameters *dma_parms;
include/linux/dma-direct.h +44 -8
72 lines not on this page · raw message
include/linux/dma-mapping.h +18 -1
@@ -730,4 +730,21 @@ static inline int dma_mmap_wc(struct device *dev,#define dma_unmap_len_set (PTR ,LEN_NAME ,VAL ) do { } while (0)#endif -#endif +/* +* These interfaces deal with the offset ranges for the direct mapping. +* Drivers really should not use them, but we have a few legacy cases +* that need them left. +*/ +int dma_direct_set_offset (struct device *dev,phys_addr_t cpu_start, +dma_addr_t dma_start,u64 size); +#ifdef CONFIG_HAS_DMA +int dma_direct_copy_range_map (struct device *to,struct device *from); +#else +static inline int dma_direct_copy_range_map (struct device *to, +struct device *from) +{ +return 0 ; +} +#endif /* CONFIG_HAS_DMA */ + +#endif /* _LINUX_DMA_MAPPING_H */
kernel/dma/coherent.c +3 -4
@@ -7,7 +7,7 @@#include <linux/slab.h> #include <linux/kernel.h> #include <linux/module.h> -#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> struct dma_coherent_mem {void *virt_base;
@@ -32,9 +32,8 @@ static inline dma_addr_t dma_get_device_base(struct device *dev, struct dma_coherent_mem * mem) {if (mem->use_dev_dma_pfn_offset) -return (mem->pfn_base - dev->dma_pfn_offset) <<PAGE_SHIFT ; -else -return mem->device_base; +return phys_to_dma (dev,PFN_PHYS (mem->pfn_base)); +return mem->device_base; } static int dma_init_coherent_memory(phys_addr_t phys_addr,
kernel/dma/direct.c +78 -1
94 lines not on this page · raw message
Signature · 1 line
2.28.0
Re: [PATCH 3/3] dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
Quoted 14 lines · Christoph Hellwig
From: Jim Quinlan <[email protected]>
The new field '
dma_range_map' in struct device is used to facilitate the use of single or multiple offsets between mapping regions of cpu addrs and dma addrs. It subsumes the role of "dev->dma_pfn_offset" which was only capable of holding a single uniform offset and had no region bounds checking.The function
of_dma_get_range()has been modified so that it takes a single argument -- the device node -- and returns a map, NULL, or an error code. The map is an array that holds the information regarding the DMA regions. Each range entry contains the address offset, thecpu_startaddress, thedma_startaddress, and the size of the region.
Quoted 4 lines
of_dma_configure()is the typical manner to set range offsets but there are a number of ad hoc assignments to "dev->dma_pfn_offset" in the kernel driver code. These cases now invoke the functiondma_attach_offset_range(dev, cpu_addr, dma_addr, size).
So if an error happens, we don't do anything?
ice_init(dev->dev);
Quoted 5 lines
drivers/usb/core/message.c +0
-0
drivers/usb/core/message.c int usb_set_configuration(struct usb_device *dev, int configuration) · line 1963
Quoted 10 lines
drivers/usb/core/message.c +3
-2@@ -1956,10 +1956,11 @@ int usb_set_configuration(struct usb_device *dev, int configuration) intf->dev.groups = usb_interface_groups;/* * Please refer to usb_alloc_dev() to see why we set -* dma_mask and dma_pfn_offset. +* dma_mask and dma_range_map. */ intf->dev.dma_mask = dev->dev.dma_mask; - intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; +if (dma_direct_copy_range_map (&intf->dev, &dev->dev)) +dev_err (&dev->dev,"failed to copy DMA map\n" );
We tell the user, but then just keep on running? Is there anything that we can do here?
If not, why not have dma_direct_copy_range_map() print out the error?
thanks,
greg k-h
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH 3/3] dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
Quoted 13 lines · Greg KH
Review
Quoted 9 lines
(quoted diff) +3
-2/* * Please refer to usb_alloc_dev() to see why we set - * dma_mask and dma_pfn_offset. + * dma_mask and dma_range_map. */ intf->dev.dma_mask = dev->dev.dma_mask; - intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; + if (dma_direct_copy_range_map(&intf->dev, &dev->dev)) + dev_err(&dev->dev, "failed to copy DMA map\n");We tell the user, but then just keep on running? Is there anything that we can do here?
Quoted 1 line
If not, why not have
dma_direct_copy_range_map()print out the error?
At least for USB I'm pretty sure this isn't required at all. I've been running with the patch below on my desktop for two days now trying all the usb toys I have (in addition to grepping for obvious abuses in the drivers). remoteproc is a different story, but the DMA handling seems there is sketchy to start with..
---
From 8bae3e6833f2ca431dcfcbc8f9cced7d5e972a01 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <[email protected]>
Date: Wed, 9 Sep 2020 08:28:59 +0200
Subject: usb: don't inherity DMA properties for USB devices
As the comment in usb_alloc_dev correctly states, drivers can't use the DMA API on usb device, and at least calling dma_set_mask on them is highly dangerous. Unlike what the comment states upper level drivers also can't really use the presence of a dma mask to check for DMA support, as the dma_mask is set by default for most busses.
Remove the copying over of DMA information, and remove the now unused dma_direct_copy_range_map export.
--- drivers/usb/core/message.c | 7 ------- drivers/usb/core/usb.c | 13 ------------- kernel/dma/direct.c | 1 - 3 files changed, 21 deletions(-)
drivers/usb/core/message.c +0 -7
@@ -1954,13 +1954,6 @@ int usb_set_configuration(struct usb_device *dev, int configuration) intf->dev.bus = &usb_bus_type; intf->dev.type = &usb_if_device_type; intf->dev.groups = usb_interface_groups; -/* -* Please refer to usb_alloc_dev() to see why we set -* dma_mask and dma_range_map. -*/ - intf->dev.dma_mask = dev->dev.dma_mask; -if (dma_direct_copy_range_map (&intf->dev, &dev->dev)) -dev_err (&dev->dev,"failed to copy DMA map\n" );INIT_WORK (&intf->reset_ws, __usb_queue_reset_device); intf->minor =-1 ;device_initialize (&intf->dev);
drivers/usb/core/usb.c +0 -13
@@ -599,19 +599,6 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent, dev->dev.bus = &usb_bus_type; dev->dev.type = &usb_device_type; dev->dev.groups = usb_device_groups; -/* -* Fake a dma_mask/offset for the USB device: -* We cannot really use the dma-mapping API (dma_alloc_* and -* dma_map_*) for USB devices but instead need to use -* usb_alloc_coherent and pass data in 'urb's, but some subsystems -* manually look into the mask/offset pair to determine whether -* they need bounce buffers. -* Note: calling dma_set_mask() on a USB device would set the -* mask for the entire HCD, so don't do that. -*/ - dev->dev.dma_mask = bus->sysdev->dma_mask; -if (dma_direct_copy_range_map (&dev->dev, bus->sysdev)) -dev_err (&dev->dev,"failed to copy DMA map\n" );set_dev_node (&dev->dev,dev_to_node (bus->sysdev)); dev->state =USB_STATE_ATTACHED ; dev->lpm_disable_count =1 ;
kernel/dma/direct.c +0 -1
@@ -552,4 +552,3 @@ int dma_direct_copy_range_map(struct device *to, struct device *from) to->dma_range_map = new_map;return 0 ; } -EXPORT_SYMBOL_GPL (dma_direct_copy_range_map);
Signature · 1 line
2.28.0
Re: [PATCH 3/3] dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
Quoted 91 lines · Christoph Hellwig
Quoted 13 lines · Greg KH
Review
Quoted 9 lines
(quoted diff) +3
-2/* * Please refer to usb_alloc_dev() to see why we set - * dma_mask and dma_pfn_offset. + * dma_mask and dma_range_map. */ intf->dev.dma_mask = dev->dev.dma_mask; - intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; + if (dma_direct_copy_range_map(&intf->dev, &dev->dev)) + dev_err(&dev->dev, "failed to copy DMA map\n");We tell the user, but then just keep on running? Is there anything that we can do here?
Quoted 1 line
If not, why not have
dma_direct_copy_range_map()print out the error?At least for USB I'm pretty sure this isn't required at all. I've been running with the patch below on my desktop for two days now trying all the usb toys I have (in addition to grepping for obvious abuses in the drivers). remoteproc is a different story, but the DMA handling seems there is sketchy to start with..
---
Quoted 1 line
From
8bae3e6833f2ca431dcfcbc8f9cced7d5e972a01Mon Sep 17 00:00:00 2001From: Christoph Hellwig <[email protected]>
Date: Wed, 9 Sep 2020 08:28:59 +0200
Subject: usb: don't inherity DMA properties for USB devicesAs the comment in
usb_alloc_devcorrectly states, drivers can't use the DMA API on usb device, and at least callingdma_set_maskon them is highly dangerous. Unlike what the comment states upper level drivers also can't really use the presence of a dma mask to check for DMA support, as thedma_maskis set by default for most busses.Remove the copying over of DMA information, and remove the now unused
dma_direct_copy_range_mapexport.Signed-off-by Christoph Hellwig <[email protected]>--- drivers/usb/core/message.c | 7 ------- drivers/usb/core/usb.c | 13 ------------- kernel/dma/direct.c | 1 - 3 files changed, 21 deletions(-)drivers/usb/core/message.c +0
-7@@ -1954,13 +1954,6 @@ int usb_set_configuration(struct usb_device *dev, int configuration) intf->dev.bus = &usb_bus_type; intf->dev.type = &usb_if_device_type; intf->dev.groups = usb_interface_groups; -/* -* Please refer to usb_alloc_dev() to see why we set -* dma_mask and dma_range_map. -*/ - intf->dev.dma_mask = dev->dev.dma_mask; -if (dma_direct_copy_range_map (&intf->dev, &dev->dev)) -dev_err (&dev->dev,"failed to copy DMA map\n" );INIT_WORK (&intf->reset_ws, __usb_queue_reset_device); intf->minor =-1 ;device_initialize (&intf->dev);drivers/usb/core/usb.c +0
-13@@ -599,19 +599,6 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent, dev->dev.bus = &usb_bus_type; dev->dev.type = &usb_device_type; dev->dev.groups = usb_device_groups; -/* -* Fake a dma_mask/offset for the USB device: -* We cannot really use the dma-mapping API (dma_alloc_* and -* dma_map_*) for USB devices but instead need to use -* usb_alloc_coherent and pass data in 'urb's, but some subsystems -* manually look into the mask/offset pair to determine whether -* they need bounce buffers. -* Note: calling dma_set_mask() on a USB device would set the -* mask for the entire HCD, so don't do that. -*/ - dev->dev.dma_mask = bus->sysdev->dma_mask; -if (dma_direct_copy_range_map (&dev->dev, bus->sysdev)) -dev_err (&dev->dev,"failed to copy DMA map\n" );set_dev_node (&dev->dev,dev_to_node (bus->sysdev)); dev->state =USB_STATE_ATTACHED ; dev->lpm_disable_count =1 ;kernel/dma/direct.c +0
-0
kernel/dma/direct.c int dma_direct_copy_range_map(struct device *to, struct device *from) · line 555
Quoted 4 lines
kernel/dma/direct.c +0
-1@@ -552,4 +552,3 @@ int dma_direct_copy_range_map(struct device *to, struct device *from) to->dma_range_map = new_map;return 0 ; } -EXPORT_SYMBOL_GPL (dma_direct_copy_range_map);
If you think this is safe to do, great, but for some reason I thought host controllers wanted this information, and that the scsi layer was the offending layer that also wanted this type of thing. But it's been a really long time so I don't remember for sure, sorry.
thanks,
greg k-h
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [PATCH 3/3] dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
In reply to Christoph Hellwig(apologies to Jim - I did look through one of the previous versions since I last commented and thought it looked OK, but never actually replied as such)
Quoted 14 lines · Christoph Hellwig
From: Jim Quinlan <[email protected]>
The new field '
dma_range_map' in struct device is used to facilitate the use of single or multiple offsets between mapping regions of cpu addrs and dma addrs. It subsumes the role of "dev->dma_pfn_offset" which was only capable of holding a single uniform offset and had no region bounds checking.The function
of_dma_get_range()has been modified so that it takes a single argument -- the device node -- and returns a map, NULL, or an error code. The map is an array that holds the information regarding the DMA regions. Each range entry contains the address offset, thecpu_startaddress, thedma_startaddress, and the size of the region.
Quoted 4 lines
of_dma_configure()is the typical manner to set range offsets but there are a number of ad hoc assignments to "dev->dma_pfn_offset" in the kernel driver code. These cases now invoke the functiondma_attach_offset_range(dev, cpu_addr, dma_addr, size).
This is now called dma_direct_set_offset(), right?
Quoted 81 lines
Signed-off-by Jim Quinlan <[email protected]>[hch: various interface cleanups]Signed-off-by Christoph Hellwig <[email protected]>Tested-by Nathan Chancellor <[email protected]>--- arch/arm/include/asm/dma-direct.h | 9 +-- arch/arm/mach-keystone/keystone.c | 17 ++-- arch/sh/drivers/pci/pcie-sh7786.c | 9 ++- arch/x86/pci/sta2x11-fixup.c | 6 +- drivers/acpi/arm64/iort.c | 6 +- drivers/base/core.c | 2 + drivers/gpu/drm/sun4i/sun4i_backend.c | 8 +- drivers/iommu/io-pgtable-arm.c | 2 +- .../platform/sunxi/sun4i-csi/sun4i_csi.c | 9 ++- .../platform/sunxi/sun6i-csi/sun6i_csi.c | 11 ++- drivers/of/address.c | 73 ++++++++--------- drivers/of/device.c | 44 ++++++---- drivers/of/of_private.h | 11 +-- drivers/of/unittest.c | 34 +++++--- drivers/remoteproc/remoteproc_core.c | 4 +- .../staging/media/sunxi/cedrus/cedrus_hw.c | 10 ++- drivers/usb/core/message.c | 5 +- drivers/usb/core/usb.c | 3 +- include/linux/device.h | 4 +- include/linux/dma-direct.h | 52 ++++++++++-- include/linux/dma-mapping.h | 19 ++++- kernel/dma/coherent.c | 7 +- kernel/dma/direct.c | 81 ++++++++++++++++++- 23 files changed, 303 insertions(+), 123 deletions(-)arch/arm/include/asm/dma-direct.h +4
-5@@ -16,8 +16,8 @@#ifndef __arch_pfn_to_dmastatic inline dma_addr_t pfn_to_dma (struct device *dev,unsigned long pfn) { -if (dev) - pfn -= dev->dma_pfn_offset; +if (dev && dev->dma_range_map) + pfn =PFN_DOWN (translate_phys_to_dma (dev,PFN_PHYS (pfn)));return (dma_addr_t )__pfn_to_bus (pfn); }@@ -25,9 +25,8 @@ static inline unsigned long dma_to_pfn(struct device *dev, dma_addr_t addr) {unsigned long pfn =__bus_to_pfn (addr); -if (dev) - pfn += dev->dma_pfn_offset; - +if (dev && dev->dma_range_map) + pfn =PFN_DOWN (translate_dma_to_phys (dev,PFN_PHYS (pfn)));return pfn; }arch/arm/mach-keystone/keystone.c +1
-2@@ -8,6 +8,7 @@*/ #include <linux/io.h> #include <linux/of.h> +#include <linux/dma-mapping.h> #include <linux/init.h> #include <linux/of_platform.h> #include <linux/of_address.h> @@ -25,8 +26,6 @@#include "keystone.h" #ifdef CONFIG_ARM_LPAE -static unsigned long keystone_dma_pfn_offset__read_mostly ; -static int keystone_platform_notifier (struct notifier_block *nb,unsigned long event,void *data) {@@ -39,9 +38,12 @@ static int keystone_platform_notifier(struct notifier_block *nb,return NOTIFY_BAD ;
arch/arm/mach-keystone/keystone.c static int keystone_platform_notifier(struct notifier_block *nb, · line 46
Quoted 10 lines
arch/arm/mach-keystone/keystone.c +6
-3if (!dev->of_node) { - dev->dma_pfn_offset = keystone_dma_pfn_offset; -dev_err (dev,"set dma_pfn_offset%08lx\n" , - dev->dma_pfn_offset); +int ret =dma_direct_set_offset (dev,KEYSTONE_HIGH_PHYS_START , +KEYSTONE_LOW_PHYS_START , +KEYSTONE_HIGH_PHYS_SIZE ); +dev_err (dev,"set dma_offset%08llx%s\n" , +KEYSTONE_HIGH_PHYS_START -KEYSTONE_LOW_PHYS_START , + ret ?" failed" :"" );
FWIW I've already been thinking of some optimisations which would have the happy side-effect of removing many of these allocation failure scenarios, but at this point I reckon it's more practical to just get the current implementation landed and working.
Quoted 138 lines
arch/arm/mach-keystone/keystone.c +1
-4}return NOTIFY_OK ; }@@ -54,11 +56,8 @@ static struct notifier_block platform_nb = {static void __init keystone_init (void ) {#ifdef CONFIG_ARM_LPAE -if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START ) { - keystone_dma_pfn_offset =PFN_DOWN (KEYSTONE_HIGH_PHYS_START - -KEYSTONE_LOW_PHYS_START ); +if (PHYS_OFFSET >=KEYSTONE_HIGH_PHYS_START )bus_register_notifier (&platform_bus_type, &platform_nb); - }#endif keystone_pm_runtime_init (); }arch/sh/drivers/pci/pcie-sh7786.c +5
-4@@ -12,6 +12,7 @@#include <linux/io.h> #include <linux/async.h> #include <linux/delay.h> +#include <linux/dma-mapping.h> #include <linux/slab.h> #include <linux/clk.h> #include <linux/sh_clk.h> @@ -31,6 +32,8 @@ struct sh7786_pcie_port {static struct sh7786_pcie_port *sh7786_pcie_ports;static unsigned int nr_ports;static unsigned long dma_pfn_offset; +size_t memsize; +u64 memstart; static struct sh7786_pcie_hwops { int (*core_init)(void);@@ -301,7 +304,6 @@ static int __init pcie_init(struct sh7786_pcie_port *port)struct pci_channel *chan = port->hose;unsigned int data;phys_addr_t memstart, memend; -size_t memsize;int ret, i, win;/* Begin initialization */ @@ -368,8 +370,6 @@ static int __init pcie_init(struct sh7786_pcie_port *port) memstart =ALIGN_DOWN (memstart, memsize); memsize =roundup_pow_of_two (memend - memstart); - dma_pfn_offset = memstart >>PAGE_SHIFT ; -/* * If there's more than 512MB of memory, we need to roll over to * LAR1/LAMR1. @@ -487,7 +487,8 @@ int pcibios_map_platform_irq(const struct pci_dev *pdev, u8 slot, u8 pin)void pcibios_bus_add_device (struct pci_dev *pdev) { - pdev->dev.dma_pfn_offset = dma_pfn_offset; +dma_direct_set_offset (&pdev->dev,__pa (memory_start), +__pa (memory_start) - memstart, memsize); }static int __init sh7786_pcie_core_init (void )arch/x86/pci/sta2x11-fixup.c +4
-2@@ -133,7 +133,7 @@ static void sta2x11_map_ep(struct pci_dev *pdev)struct sta2x11_instance *instance =sta2x11_pdev_to_instance (pdev);struct device *dev = &pdev->dev;u32 amba_base, max_amba_addr; -int i; +int i, ret;if (!instance)return ;@@ -141,7 +141,9 @@ static void sta2x11_map_ep(struct pci_dev *pdev)pci_read_config_dword (pdev,AHB_BASE (0 ), &amba_base); max_amba_addr = amba_base +STA2X11_AMBA_SIZE -1 ; - dev->dma_pfn_offset =PFN_DOWN (-amba_base); + ret =dma_direct_set_offset (dev,0 , amba_base,STA2X11_AMBA_SIZE ); +if (ret) +dev_err (dev,"sta2x11: could not set DMA offset\n" ); dev->bus_dma_limit = max_amba_addr;pci_set_consistent_dma_mask (pdev, max_amba_addr);drivers/acpi/arm64/iort.c +4
-2@@ -18,6 +18,7 @@#include <linux/pci.h> #include <linux/platform_device.h> #include <linux/slab.h> +#include <linux/dma-mapping.h> #define IORT_TYPE_MASK (type) (1 << (type))#define IORT_MSI_TYPE (1 << ACPI_IORT_NODE_ITS_GROUP)@@ -1184,8 +1185,9 @@ void iort_dma_setup(struct device *dev, u64 *dma_addr, u64 *dma_size) *dma_addr = dmaaddr; *dma_size = size; - dev->dma_pfn_offset =PFN_DOWN (offset); -dev_dbg (dev,"dma_pfn_offset(%#08llx)\n" , offset); + ret =dma_direct_set_offset (dev, dmaaddr + offset, dmaaddr, size); + +dev_dbg (dev,"dma_offset(%#08llx)%s\n" , offset, ret ?" failed!" :"" ); } static void __init acpi_iort_register_irq(int hwirq, const char *name,drivers/base/core.c +2
-0@@ -1792,6 +1792,8 @@ static void device_release(struct kobject *kobj)*/ devres_release_all (dev); +kfree (dev->dma_range_map); +if (dev->release) dev->release (dev);else if (dev->type && dev->type->release)drivers/gpu/drm/sun4i/sun4i_backend.c +1
-0@@ -11,6 +11,7 @@#include <linux/module.h> #include <linux/of_device.h> #include <linux/of_graph.h> +#include <linux/dma-mapping.h> #include <linux/platform_device.h> #include <linux/reset.h>
drivers/gpu/drm/sun4i/sun4i_backend.c static int sun4i_backend_bind(struct device *dev, struct device *master, · line 817
Quoted 6 lines
drivers/gpu/drm/sun4i/sun4i_backend.c +3
-0@@ -811,8 +812,13 @@ static int sun4i_backend_bind(struct device *dev, struct device *master,* because of an old DT, we need to set the DMA offset by hand * on our device since the RAM mapping is at 0 for the DMA bus, * unlike the CPU. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree.
As the context implies, this has actually grown a proper DT description of the funky interconnect layout (see 564d6fd611f9 and the linked patch series), and this is just an ugly fallback path to prevent regressions with old DTBs that are already out there. So unless you can fire up the time machine to fix those, this extra comment is really just beating a dead horse :(
Quoted 34 lines
drivers/gpu/drm/sun4i/sun4i_backend.c +3
-1*/ - drm->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; + ret =dma_direct_set_offset (drm->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; } backend->engine.node = dev->of_node;drivers/iommu/io-pgtable-arm.c +1
-1@@ -751,7 +751,7 @@ arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg)if (cfg->oas >ARM_LPAE_MAX_ADDR_BITS )return NULL ; -if (!selftest_running && cfg->iommu_dev->dma_pfn_offset) { +if (!selftest_running && cfg->iommu_dev->dma_range_map) {dev_err (cfg->iommu_dev,"Cannot accommodate DMA offset for IOMMU page tables\n" );return NULL ; }drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c +1
-0@@ -7,6 +7,7 @@*/ #include <linux/clk.h> +#include <linux/dma-mapping.h> #include <linux/interrupt.h> #include <linux/module.h> #include <linux/mutex.h>
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c static int sun4i_csi_probe(struct platform_device *pdev) · line 189
Quoted 7 lines
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c +4
-0@@ -182,8 +183,14 @@ static int sun4i_csi_probe(struct platform_device *pdev)if (ret)return ret; }else { +/* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. +*/
Again, the top half of that "else" does already handle the proper DT case and document this as a fallback :(
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c static int sun4i_csi_probe(struct platform_device *pdev) · line 190
Quoted 1 line
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c +0
-0#ifdef PHYS_PFN_OFFSET
Strictly that probably deserves updating too, but PHYS_PFN_OFFSET implies PHYS_OFFSET so there should be no functional issue, just semantics.
Quoted 13 lines
drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c +3
-1- csi->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; + ret =dma_direct_set_offset (csi->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret;#endif }drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c +0
-0@@ -899,8 +899,15 @@ static int sun6i_csi_probe(struct platform_device *pdev)return -ENOMEM ;
drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c static int sun6i_csi_probe(struct platform_device *pdev) · line 907
Quoted 10 lines
drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c +6
-2sdev->dev = &pdev->dev; -/* The DMA bus has the memory mapped at 0 */ - sdev->dev->dma_pfn_offset =PHYS_OFFSET >>PAGE_SHIFT ; +/* +* The DMA bus has the memory mapped at 0. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. +*/
Apparently this one *does* want updating to use the MBUS interconnect bindings - as does the cedrus instance below - so no complaint there :)
Other than those few trivial nitpicks, I think I've stared at this patch enough times now that even if there are any bugs left I'm not going to see them... let's try to get it thrown at linux-next and everyone's CI systems :)
Cheers,
Robin.
Quoted 666 lines
drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c +3
-0+ ret =dma_direct_set_offset (sdev->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; ret =sun6i_csi_resource_request (sdev, pdev);if (ret)drivers/of/address.c +32
-41@@ -13,6 +13,7 @@#include <linux/sizes.h> #include <linux/slab.h> #include <linux/string.h> +#include <linux/dma-direct.h> /* for bus_dma_region */ #include "of_private.h" @@ -937,33 +938,33 @@ void __iomem *of_io_request_and_map(struct device_node *np, int index, }EXPORT_SYMBOL (of_io_request_and_map); +#ifdef CONFIG_HAS_DMA /** -* of_dma_get_range - Get DMA range info +* of_dma_get_range - Get DMA range info and put it into a map array * @np: device node to get DMA range info -* @dma_addr: pointer to store initial DMA address of DMA range -* @paddr: pointer to store initial CPU address of DMA range -* @size: pointer to store size of DMA range +* @map: dma range structure to return * * Look in bottom up direction for the first "dma-ranges" property -* and parse it. -* dma-ranges format: +* and parse it. Put the information into a DMA offset map array. +* +* dma-ranges format: * DMA addr (dma_addr) : naddr cells * CPU addr (phys_addr_t) : pna cells * size : nsize cells * -* It returns -ENODEV if "dma-ranges" property was not found -* for this device in DT. +* It returns -ENODEV if "dma-ranges" property was not found for this +* device in the DT. */ -int of_dma_get_range (struct device_node *np,u64 *dma_addr,u64 *paddr,u64 *size) +int of_dma_get_range (struct device_node *np,const struct bus_dma_region **map) {struct device_node *node =of_node_get (np);const __be32 *ranges =NULL ; -int len; -int ret =0 ;bool found_dma_ranges = false;struct of_range_parser parser;struct of_range range; -u64 dma_start =U64_MAX , dma_end =0 , dma_offset =0 ; +struct bus_dma_region *r; +int len, num_ranges =0 ; +int ret =0 ;while (node) { ranges =of_get_property (node,"dma-ranges" , &len);@@ -989,49 +990,39 @@ int of_dma_get_range(struct device_node *np, u64 *dma_addr, u64 *paddr, u64 *siz }of_dma_range_parser_init (&parser, node); +for_each_of_range (&parser, &range) + num_ranges++; + + r =kcalloc (num_ranges +1 ,sizeof (*r),GFP_KERNEL ); +if (!r) { + ret = -ENOMEM ; + goto out; + } +/* +* Record all info in the generic DMA ranges array for struct device. +*/ + *map = r; +of_dma_range_parser_init (&parser, node);for_each_of_range (&parser, &range) {pr_debug ("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n" , range.bus_addr, range.cpu_addr, range.size); - -if (dma_offset && range.cpu_addr - range.bus_addr != dma_offset) { -pr_warn ("Can't handle multiple dma-ranges with different offsets on node(%pOF)\n" , node); -/* Don't error out as we'd break some existing DTs */ -continue ; - }if (range.cpu_addr =OF_BAD_ADDR ) {pr_err ("translation of DMA address(%llx) to CPU address failed node(%pOF)\n" , range.bus_addr, node);continue ; } - dma_offset = range.cpu_addr - range.bus_addr; - -/* Take lower and upper limits */ -if (range.bus_addr < dma_start) - dma_start = range.bus_addr; -if (range.bus_addr + range.size > dma_end) - dma_end = range.bus_addr + range.size; - } - -if (dma_start >= dma_end) { - ret = -EINVAL ; -pr_debug ("Invalid DMA ranges configuration on node(%pOF)\n" , - node); - goto out; + r->cpu_start = range.cpu_addr; + r->dma_start = range.bus_addr; + r->size = range.size; + r->offset = range.cpu_addr - range.bus_addr; + r++; } - - *dma_addr = dma_start; - *size = dma_end - dma_start; - *paddr = dma_start + dma_offset; - -pr_debug ("final: dma_addr(%llx) cpu_addr(%llx) size(%llx)\n" , - *dma_addr, *paddr, *size); - out:of_node_put (node); -return ret; } +#endif /* CONFIG_HAS_DMA */ /** * of_dma_is_coherent - Check if device is coherent drivers/of/device.c +27
-17@@ -5,7 +5,7 @@#include <linux/of_device.h> #include <linux/of_address.h> #include <linux/of_iommu.h> -#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> /* for bus_dma_region */ #include <linux/init.h> #include <linux/module.h> #include <linux/mod_devicetable.h> @@ -90,14 +90,14 @@ int of_device_add(struct platform_device *ofdev)int of_dma_configure_id (struct device *dev,struct device_node *np,bool force_dma,const u32 *id) { -u64 dma_addr, paddr, size =0 ; -int ret; -bool coherent; -unsigned long offset;const struct iommu_ops *iommu; -u64 mask, end; +const struct bus_dma_region *map =NULL ; +dma_addr_t dma_start =0 ; +u64 mask, end, size =0 ; +bool coherent; +int ret; - ret =of_dma_get_range (np, &dma_addr, &paddr, &size); + ret =of_dma_get_range (np, &map);if (ret <0 ) {/* * For legacy reasons, we have to assume some devices need @@ -106,26 +106,35 @@ int of_dma_configure_id(struct device *dev, struct device_node *np,*/ if (!force_dma)return ret = -ENODEV ?0 : ret; - - dma_addr = offset =0 ; }else { - offset =PFN_DOWN (paddr - dma_addr); +const struct bus_dma_region *r = map; +dma_addr_t dma_end =0 ; + +/* Determine the overall bounds of all DMA regions */ +for (dma_start = ~(dma_addr_t )0 ; r->size; r++) { +/* Take lower and upper limits */ +if (r->dma_start < dma_start) + dma_start = r->dma_start; +if (r->dma_start + r->size > dma_end) + dma_end = r->dma_start + r->size; + } + size = dma_end - dma_start;/* * Add a work around to treat the size as mask + 1 in case * it is defined in DT as a mask. */ if (size &1 ) { -dev_warn (dev,"Invalid size 0x%llx for dma-range\n" , +dev_warn (dev,"Invalid size 0x%llx for dma-range(s)\n" , size); size = size +1 ; }if (!size) {dev_err (dev,"Adjusted size 0x%llx invalid\n" , size); +kfree (map);return -EINVAL ; } -dev_dbg (dev,"dma_pfn_offset(%#08lx)\n" , offset); }/* @@ -144,13 +153,11 @@ int of_dma_configure_id(struct device *dev, struct device_node *np,else if (!size) size =1ULL <<32 ; - dev->dma_pfn_offset = offset; -/* * Limit coherent and dma mask based on size and default mask * set by the driver. */ - end = dma_addr + size -1 ; + end = dma_start + size -1 ; mask =DMA_BIT_MASK (ilog2 (end) +1 ); dev->coherent_dma_mask &= mask; *dev->dma_mask &= mask;@@ -163,14 +170,17 @@ int of_dma_configure_id(struct device *dev, struct device_node *np, coherent ? " " : " not "); iommu =of_iommu_configure (dev, np, id); -if (PTR_ERR (iommu) = -EPROBE_DEFER ) +if (PTR_ERR (iommu) = -EPROBE_DEFER ) { +kfree (map);return -EPROBE_DEFER ; + }dev_dbg (dev,"device is%sbehind an iommu\n" , iommu ?" " :" not " ); -arch_setup_dma_ops (dev, dma_addr, size, iommu, coherent); +arch_setup_dma_ops (dev, dma_start, size, iommu, coherent); + dev->dma_range_map = map;return 0 ; }EXPORT_SYMBOL_GPL (of_dma_configure_id);drivers/of/of_private.h +6
-5@@ -157,12 +157,13 @@ extern void __of_sysfs_remove_bin_file(struct device_node *np,extern int of_bus_n_addr_cells (struct device_node *np);extern int of_bus_n_size_cells (struct device_node *np); -#ifdef CONFIG_OF_ADDRESS -extern int of_dma_get_range (struct device_node *np,u64 *dma_addr, -u64 *paddr,u64 *size); +struct bus_dma_region ; +#if defined(CONFIG_OF_ADDRESS ) && defined(CONFIG_HAS_DMA ) +int of_dma_get_range (struct device_node *np, +const struct bus_dma_region **map);#else -static inline int of_dma_get_range (struct device_node *np,u64 *dma_addr, -u64 *paddr,u64 *size) +static inline int of_dma_get_range (struct device_node *np, +const struct bus_dma_region **map) {return -ENODEV ; }drivers/of/unittest.c +24
-10@@ -7,6 +7,7 @@#include <linux/memblock.h> #include <linux/clk.h> +#include <linux/dma-direct.h> /* to test phys_to_dma/dma_to_phys */ #include <linux/err.h> #include <linux/errno.h> #include <linux/hashtable.h> @@ -869,10 +870,11 @@ static void __init of_unittest_changeset(void) }static void __init of_unittest_dma_ranges_one (const char *path, -u64 expect_dma_addr,u64 expect_paddr,u64 expect_size) +u64 expect_dma_addr,u64 expect_paddr) { +#ifdef CONFIG_HAS_DMA struct device_node *np; -u64 dma_addr, paddr, size; +const struct bus_dma_region *map =NULL ;int rc; np =of_find_node_by_path (path);@@ -881,28 +883,40 @@ static void __init of_unittest_dma_ranges_one(const char *path,return ; } - rc =of_dma_get_range (np, &dma_addr, &paddr, &size); + rc =of_dma_get_range (np, &map);unittest (!rc,"of_dma_get_range failed on node %pOF rc=%i\n" , np, rc); +if (!rc) { -unittest (size = expect_size, -"of_dma_get_range wrong size on node %pOF size=%llx\n" , np, size); +phys_addr_t paddr; +dma_addr_t dma_addr; +struct device dev_bogus; + + dev_bogus.dma_range_map = map; + paddr =dma_to_phys (&dev_bogus, expect_dma_addr); + dma_addr =phys_to_dma (&dev_bogus, expect_paddr); +unittest (paddr = expect_paddr, -"of_dma_get_range wrong phys addr (%llx) on node %pOF" , paddr, np); +"of_dma_get_range: wrong phys addr %pap (expecting %llx) on node %pOF\n" , + &paddr, expect_paddr, np);unittest (dma_addr = expect_dma_addr, -"of_dma_get_range wrong DMA addr (%llx) on node %pOF" , dma_addr, np); +"of_dma_get_range: wrong DMA addr %pad (expecting %llx) on node %pOF\n" , + &dma_addr, expect_dma_addr, np); + +kfree (map); }of_node_put (np); +#endif }static void __init of_unittest_parse_dma_ranges (void ) {of_unittest_dma_ranges_one ("/testcase-data/address-tests/device@70000000" , -0x0 ,0x20000000 ,0x40000000 ); +0x0 ,0x20000000 );of_unittest_dma_ranges_one ("/testcase-data/address-tests/bus@80000000/device@1000" , -0x100000000 ,0x20000000 ,0x2000000000 ); +0x100000000 ,0x20000000 );of_unittest_dma_ranges_one ("/testcase-data/address-tests/pci@90000000" , -0x80000000 ,0x20000000 ,0x10000000 ); +0x80000000 ,0x20000000 ); }static void __init of_unittest_pci_dma_ranges (void )drivers/remoteproc/remoteproc_core.c +3
-1@@ -529,7 +529,9 @@ static int rproc_handle_vdev(struct rproc *rproc, struct fw_rsc_vdev *rsc,/* Initialise vdev subdevice */ snprintf (name,sizeof (name),"vdev%dbuffer" , rvdev->index); rvdev->dev.parent = &rproc->dev; - rvdev->dev.dma_pfn_offset = rproc->dev.parent->dma_pfn_offset; + ret =dma_direct_copy_range_map (&rvdev->dev, rproc->dev.parent); +if (ret) +return ret; rvdev->dev.release = rproc_rvdev_release;dev_set_name (&rvdev->dev,"%s#%s" ,dev_name (rvdev->dev.parent), name);dev_set_drvdata (&rvdev->dev, rvdev);drivers/staging/media/sunxi/cedrus/cedrus_hw.c +8
-2@@ -227,11 +227,17 @@ int cedrus_hw_probe(struct cedrus_dev *dev)* the RAM offset to the physcal addresses. * * This information will eventually be obtained from device-tree. +* +* XXX(hch): this has no business in a driver and needs to move +* to the device tree. */ #ifdef PHYS_PFN_OFFSET -if (!(variant->quirks &CEDRUS_QUIRK_NO_DMA_OFFSET )) - dev->dev->dma_pfn_offset =PHYS_PFN_OFFSET ; +if (!(variant->quirks &CEDRUS_QUIRK_NO_DMA_OFFSET )) { + ret =dma_direct_set_offset (dev->dev,PHYS_OFFSET ,0 ,SZ_4G ); +if (ret) +return ret; + }#endif ret =of_reserved_mem_device_init (dev->dev);drivers/usb/core/message.c +3
-2@@ -1956,10 +1956,11 @@ int usb_set_configuration(struct usb_device *dev, int configuration) intf->dev.groups = usb_interface_groups;/* * Please refer to usb_alloc_dev() to see why we set -* dma_mask and dma_pfn_offset. +* dma_mask and dma_range_map. */ intf->dev.dma_mask = dev->dev.dma_mask; - intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; +if (dma_direct_copy_range_map (&intf->dev, &dev->dev)) +dev_err (&dev->dev,"failed to copy DMA map\n" );INIT_WORK (&intf->reset_ws, __usb_queue_reset_device); intf->minor =-1 ;device_initialize (&intf->dev);drivers/usb/core/usb.c +2
-1@@ -610,7 +610,8 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent,* mask for the entire HCD, so don't do that. */ dev->dev.dma_mask = bus->sysdev->dma_mask; - dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; +if (dma_direct_copy_range_map (&dev->dev, bus->sysdev)) +dev_err (&dev->dev,"failed to copy DMA map\n" );set_dev_node (&dev->dev,dev_to_node (bus->sysdev)); dev->state =USB_STATE_ATTACHED ; dev->lpm_disable_count =1 ;include/linux/device.h +2
-2@@ -466,7 +466,7 @@ struct dev_links_info {* such descriptors. * @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_range_map: map for DMA memory ranges relative to that of RAM * @dma_parms: A low level driver may set these to teach IOMMU code about * segment limitations. * @dma_pools: Dma pools (if dma'ble device). @@ -561,7 +561,7 @@ struct device {64 bit addresses for consistent allocations such descriptors. */ u64 bus_dma_limit;/* upstream dma constraint */ -unsigned long dma_pfn_offset; +const struct bus_dma_region *dma_range_map;struct device_dma_parameters *dma_parms;include/linux/dma-direct.h +44
-8@@ -14,21 +14,56 @@extern unsigned int zone_dma_bits; +/* +* Record the mapping of CPU physical to DMA addresses for a given region. +*/ +struct bus_dma_region { +phys_addr_t cpu_start; +dma_addr_t dma_start; +u64 size; +u64 offset; +}; + +static inline dma_addr_t translate_phys_to_dma (struct device *dev, +phys_addr_t paddr) +{ +const struct bus_dma_region *m; + +for (m = dev->dma_range_map; m->size; m++) +if (paddr >= m->cpu_start && paddr - m->cpu_start < m->size) +return (dma_addr_t )paddr - m->offset; + +/* make sure dma_capable fails when no translation is available */ +return DMA_MAPPING_ERROR ; +} + +static inline phys_addr_t translate_dma_to_phys (struct device *dev, +dma_addr_t dma_addr) +{ +const struct bus_dma_region *m; + +for (m = dev->dma_range_map; m->size; m++) +if (dma_addr >= m->dma_start && dma_addr - m->dma_start < m->size) +return (phys_addr_t )dma_addr + m->offset; + +return 0 ; +} +#ifdef CONFIG_ARCH_HAS_PHYS_TO_DMA #include <asm/dma-direct.h> #else static inline dma_addr_t __phys_to_dma (struct device *dev,phys_addr_t paddr) { -dma_addr_t dev_addr = (dma_addr_t )paddr; - -return dev_addr - ((dma_addr_t )dev->dma_pfn_offset <<PAGE_SHIFT ); +if (dev->dma_range_map) +return translate_phys_to_dma (dev, paddr); +return paddr; } -static inline phys_addr_t __dma_to_phys (struct device *dev,dma_addr_t dev_addr) +static inline phys_addr_t __dma_to_phys (struct device *dev,dma_addr_t dma_addr) { -phys_addr_t paddr = (phys_addr_t )dev_addr; - -return paddr + ((phys_addr_t )dev->dma_pfn_offset <<PAGE_SHIFT ); +if (dev->dma_range_map) +return translate_dma_to_phys (dev, dma_addr); +return dma_addr; }#endif /* !CONFIG_ARCH_HAS_PHYS_TO_DMA */ @@ -64,7 +99,8 @@ static inline bool dma_capable(struct device *dev, dma_addr_t addr, size_t size,if (!dev->dma_mask)return false; - +if (addr =DMA_MAPPING_ERROR ) +return false;if (is_ram && !IS_ENABLED (CONFIG_ARCH_DMA_ADDR_T_64BIT ) &&min (addr, end) <phys_to_dma (dev,PFN_PHYS (min_low_pfn)))return false;include/linux/dma-mapping.h +18
-1@@ -730,4 +730,21 @@ static inline int dma_mmap_wc(struct device *dev,#define dma_unmap_len_set (PTR ,LEN_NAME ,VAL ) do { } while (0)#endif -#endif +/* +* These interfaces deal with the offset ranges for the direct mapping. +* Drivers really should not use them, but we have a few legacy cases +* that need them left. +*/ +int dma_direct_set_offset (struct device *dev,phys_addr_t cpu_start, +dma_addr_t dma_start,u64 size); +#ifdef CONFIG_HAS_DMA +int dma_direct_copy_range_map (struct device *to,struct device *from); +#else +static inline int dma_direct_copy_range_map (struct device *to, +struct device *from) +{ +return 0 ; +} +#endif /* CONFIG_HAS_DMA */ + +#endif /* _LINUX_DMA_MAPPING_H */ kernel/dma/coherent.c +3
-4@@ -7,7 +7,7 @@#include <linux/slab.h> #include <linux/kernel.h> #include <linux/module.h> -#include <linux/dma-mapping.h> +#include <linux/dma-direct.h> struct dma_coherent_mem {void *virt_base;@@ -32,9 +32,8 @@ static inline dma_addr_t dma_get_device_base(struct device *dev, struct dma_coherent_mem * mem) {if (mem->use_dev_dma_pfn_offset) -return (mem->pfn_base - dev->dma_pfn_offset) <<PAGE_SHIFT ; -else -return mem->device_base; +return phys_to_dma (dev,PFN_PHYS (mem->pfn_base)); +return mem->device_base; } static int dma_init_coherent_memory(phys_addr_t phys_addr,kernel/dma/direct.c +78
-1@@ -13,6 +13,7 @@#include <linux/pfn.h> #include <linux/vmalloc.h> #include <linux/set_memory.h> +#include <linux/slab.h> /* * Most architectures use ZONE_DMA for the first 16 Megabytes, but some use it @@ -70,8 +71,12 @@ static gfp_t dma_direct_optimal_gfp_mask(struct device *dev, u64 dma_mask,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_limit); +dma_addr_t dma_addr =phys_to_dma_direct (dev, phys); + +if (dma_addr =DMA_MAPPING_ERROR ) +return false; + return dma_addr + size - 1 <> + min_not_zero(dev->coherent_dma_mask, dev->bus_dma_limit); }/* @@ -476,3 +481,75 @@ bool dma_direct_need_sync(struct device *dev, dma_addr_t dma_addr)return !dev_is_dma_coherent (dev) ||is_swiotlb_buffer (dma_to_phys (dev, dma_addr)); } + +/** +* dma_direct_set_offset - Assign scalar offset for a single DMA range. +* @dev: device pointer; needed to "own" the alloced memory. +* @cpu_start: beginning of memory region covered by this offset. +* @dma_start: beginning of DMA/PCI region covered by this offset. +* @size: size of the region. +* +* This is for the simple case of a uniform offset which cannot +* be discovered by "dma-ranges". +* +* It returns -ENOMEM if out of memory, -EINVAL if a map +* already exists, 0 otherwise. +* +* Note: any call to this from a driver is a bug. The mapping needs +* to be described by the device tree or other firmware interfaces. +*/ +int dma_direct_set_offset (struct device *dev,phys_addr_t cpu_start, +dma_addr_t dma_start,u64 size) +{ +struct bus_dma_region *map; +u64 offset = (u64 )cpu_start - (u64 )dma_start; + +if (dev->dma_range_map) { +dev_err (dev,"attempt to add DMA range to existing map\n" ); +return -EINVAL ; + } + +if (!offset) +return 0 ; + + map =kcalloc (2 ,sizeof (*map),GFP_KERNEL ); +if (!map) +return -ENOMEM ; + map[0 ].cpu_start = cpu_start; + map[0 ].dma_start = dma_start; + map[0 ].offset = offset; + map[0 ].size = size; + dev->dma_range_map = map; + +return 0 ; +} +EXPORT_SYMBOL_GPL (dma_direct_set_offset); + +/** +* dma_direct_copy_range_map - Copy the dma_range from one device to another +* @to: device to copy to +* @from: device to copy from +* +* Note: this should not be used in any new code. Drivers should never try +* to inherit DMA parameters from other devices but just use those other +* devices in the DMA mapping functions. +*/ +int dma_direct_copy_range_map (struct device *to,struct device *from) +{ +const struct bus_dma_region *map = from->dma_range_map, *new_map, *r; +int num_ranges =0 ; + +if (!map) +return 0 ; + +for (r = map; r->size; r++) + num_ranges++; + + new_map =kmemdup (map,array_size (num_ranges +1 ,sizeof (*map)), +GFP_KERNEL ); +if (!new_map) +return -ENOMEM ; + to->dma_range_map = new_map; +return 0 ; +} +EXPORT_SYMBOL_GPL (dma_direct_copy_range_map);
Re: [PATCH 3/3] dma-mapping: introduce DMA range map, supplanting dma_pfn_offset
Quoted 19 lines · Robin Murphy
(apologies to Jim - I did look through one of the previous versions since I last commented and thought it looked OK, but never actually replied as such)
Quoted 14 lines · Christoph Hellwig
From: Jim Quinlan <[email protected]>
The new field '
dma_range_map' in struct device is used to facilitate the use of single or multiple offsets between mapping regions of cpu addrs and dma addrs. It subsumes the role of "dev->dma_pfn_offset" which was only capable of holding a single uniform offset and had no region bounds checking.The function
of_dma_get_range()has been modified so that it takes a single argument -- the device node -- and returns a map, NULL, or an error code. The map is an array that holds the information regarding the DMA regions. Each range entry contains the address offset, thecpu_startaddress, thedma_startaddress, and the size of the region.
Quoted 6 lines
Quoted 4 lines
of_dma_configure()is the typical manner to set range offsets but there are a number of ad hoc assignments to "dev->dma_pfn_offset" in the kernel driver code. These cases now invoke the functiondma_attach_offset_range(dev, cpu_addr, dma_addr, size).This is now called
dma_direct_set_offset(), right?
Yes.
Quoted 11 lines
arch/arm/mach-keystone/keystone.c static int keystone_platform_notifier(struct notifier_block *nb, · line 46
Quoted 6 lines
arch/arm/mach-keystone/keystone.c +6
-0+int ret =dma_direct_set_offset (dev,KEYSTONE_HIGH_PHYS_START , +KEYSTONE_LOW_PHYS_START , +KEYSTONE_HIGH_PHYS_SIZE ); +dev_err (dev,"set dma_offset%08llx%s\n" , +KEYSTONE_HIGH_PHYS_START -KEYSTONE_LOW_PHYS_START , + ret ?" failed" :"" );FWIW I've already been thinking of some optimisations which would have the happy side-effect of removing many of these allocation failure scenarios, but at this point I reckon it's more practical to just get the current implementation landed and working.
Given that no one deals or can easily deal with these failures we should probably take care of that. IMHO we could just allocate a single static range and point all the devices to it, what would you think of that?
Quoted 7 lines
Quoted 7 lines
@@ -811,8 +812,13 @@ static int sun4i_backend_bind(struct device *dev,
struct device *master,
* because of an old DT, we need to set the DMA offset by hand
* on our device since the RAM mapping is at 0 for the DMA bus,
* unlike the CPU.(quoted diff) +2
-0+ * + * XXX(hch): this has no business in a driver and needs to move
Quoted 8 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ * to the device tree.As the context implies, this has actually grown a proper DT description of the funky interconnect layout (see
564d6fd611f9and the linked patch series), and this is just an ugly fallback path to prevent regressions with old DTBs that are already out there. So unless you can fire up the time machine to fix those, this extra comment is really just beating a dead horse :(
Well, I need to beat the dead horse to avoid having to export the function, which will really just bread new users. So at the minimum we'd need to move the setup to platform code that is always built in.