| Message ID | 20200526191303.1492-10-james.quinlan@broadcom.com |
|---|---|
| State | Changes Requested, archived |
| Headers | show |
| Series | PCI: brcmstb: enable PCIe for STB chips | expand |
| Context | Check | Description |
|---|---|---|
| robh/checkpatch | warning | total: 0 errors, 2 warnings, 221 lines checked |
On Tue, May 26, 2020 at 03:12:48PM -0400, Jim Quinlan wrote: > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > similar to 'dma_pfn_offset' except that the offset chosen depends on the > cpu or dma address involved. > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > --- > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > drivers/usb/core/message.c | 3 ++ > drivers/usb/core/usb.c | 3 ++ > include/linux/device.h | 10 +++++- > include/linux/dma-direct.h | 10 ++++-- > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > kernel/dma/Kconfig | 13 ++++++++ > 7 files changed, 144 insertions(+), 6 deletions(-) > > diff --git a/drivers/of/address.c b/drivers/of/address.c > index 96d8cfb14a60..a01afffcde7d 100644 > --- a/drivers/of/address.c > +++ b/drivers/of/address.c > @@ -918,6 +918,47 @@ void __iomem *of_io_request_and_map(struct device_node *np, int index, > } > EXPORT_SYMBOL(of_io_request_and_map); > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > +static int attach_dma_pfn_offset_map(struct device *dev, > + struct device_node *node, int num_ranges) > +{ > + struct of_range_parser parser; > + struct of_range range; > + size_t r_size = (num_ranges + 1) > + * sizeof(struct dma_pfn_offset_region); > + struct dma_pfn_offset_region *r; > + > + r = devm_kzalloc(dev, r_size, GFP_KERNEL); devm_?! Looking at r_size it should be rather kcalloc(). > + if (!r) > + return -ENOMEM; > + dev->dma_pfn_offset_map = r; > + of_dma_range_parser_init(&parser, node); > + > + /* > + * Record all info for DMA ranges array. We could > + * just use the of_range struct, but if we did that it > + * would require more calculations for phys_to_dma and > + * dma_to_phys conversions. > + */ > + for_each_of_range(&parser, &range) { > + r->cpu_beg = range.cpu_addr; > + r->cpu_end = r->cpu_beg + range.size; > + r->dma_beg = range.bus_addr; > + r->dma_end = r->dma_beg + range.size; > + r->pfn_offset = PFN_DOWN(range.cpu_addr) > + - PFN_DOWN(range.bus_addr); > + r++; > + } > + return 0; > +} > +#else > +static int attach_dma_pfn_offset_map(struct device *dev, > + struct device_node *node, int num_ranges) > +{ > + return 0; > +} > +#endif > + > /** > * of_dma_get_range - Get DMA range info > * @dev: device pointer; only needed for a corner case. > @@ -947,6 +988,8 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > struct of_range_parser parser; > struct of_range range; > u64 dma_start = U64_MAX, dma_end = 0, dma_offset = 0; > + bool dma_multi_pfn_offset = false; > + int num_ranges = 0; > > while (node) { > ranges = of_get_property(node, "dma-ranges", &len); > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > range.bus_addr, range.cpu_addr, range.size); > > + num_ranges++; > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > + pr_warn("Can't handle multiple dma-ranges with different offsets on node(%pOF)\n", node); > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > + /* > + * Don't error out as we'd break some existing > + * DTs that are using configs w/o > + * CONFIG_DMA_PFN_OFFSET_MAP set. > + */ > + continue; > + } > + dma_multi_pfn_offset = true; > } > dma_offset = range.cpu_addr - range.bus_addr; > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > dma_end = range.bus_addr + range.size; > } > > + if (dma_multi_pfn_offset) { > + dma_offset = 0; > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > + if (ret) > + return ret; > + } > + > if (dma_start >= dma_end) { > ret = -EINVAL; > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > index 6197938dcc2d..aaa3e58f5eb4 100644 > --- a/drivers/usb/core/message.c > +++ b/drivers/usb/core/message.c > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int configuration) > */ > intf->dev.dma_mask = dev->dev.dma_mask; > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > +#endif > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > intf->minor = -1; > device_initialize(&intf->dev); > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > index f16c26dc079d..d2ed4d90e56e 100644 > --- a/drivers/usb/core/usb.c > +++ b/drivers/usb/core/usb.c > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent, > */ > dev->dev.dma_mask = bus->sysdev->dma_mask; > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > +#endif > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > dev->state = USB_STATE_ATTACHED; > dev->lpm_disable_count = 1; > diff --git a/include/linux/device.h b/include/linux/device.h > index ac8e37cd716a..67a240ad4fc5 100644 > --- a/include/linux/device.h > +++ b/include/linux/device.h > @@ -493,6 +493,8 @@ struct dev_links_info { > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are multiple > + * pfn offsets for multiple dma-ranges. > * @dma_parms: A low level driver may set these to teach IOMMU code about > * segment limitations. > * @dma_pools: Dma pools (if dma'ble device). > @@ -578,7 +580,13 @@ struct device { > allocations such descriptors. */ > u64 bus_dma_limit; /* upstream dma constraint */ > unsigned long dma_pfn_offset; > - > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > + /* Like dma_pfn_offset, but for > + * the unlikely case of multiple > + * offsets. If non-null, dma_pfn_offset > + * will be set to 0. */ A bit harder to read comment indented too much and located after the declared variable. > +#endif > struct device_dma_parameters *dma_parms; > > struct list_head dma_pools; /* dma pools (if dma'ble) */ > diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h > index 24b8684aa21d..03110a57eabc 100644 > --- a/include/linux/dma-direct.h > +++ b/include/linux/dma-direct.h > @@ -14,15 +14,21 @@ extern unsigned int zone_dma_bits; > static inline dma_addr_t __phys_to_dma(struct device *dev, phys_addr_t paddr) > { > dma_addr_t dev_addr = (dma_addr_t)paddr; > + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ > + unsigned long dma_pfn_offset = dev->dma_pfn_offset > + + dma_pfn_offset_from_phys_addr(dev, paddr); > > - return dev_addr - ((dma_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); > + return dev_addr - ((dma_addr_t)dma_pfn_offset << PAGE_SHIFT); > } > > static inline phys_addr_t __dma_to_phys(struct device *dev, dma_addr_t dev_addr) > { > phys_addr_t paddr = (phys_addr_t)dev_addr; > + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ > + unsigned long dma_pfn_offset = dev->dma_pfn_offset > + + dma_pfn_offset_from_dma_addr(dev, paddr); > > - return paddr + ((phys_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); > + return paddr + ((phys_addr_t)dma_pfn_offset << PAGE_SHIFT); > } > #endif /* !CONFIG_ARCH_HAS_PHYS_TO_DMA */ > > diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h > index 330ad58fbf4d..91940bba2229 100644 > --- a/include/linux/dma-mapping.h > +++ b/include/linux/dma-mapping.h > @@ -256,6 +256,52 @@ static inline void dma_direct_sync_sg_for_cpu(struct device *dev, > size_t dma_direct_max_mapping_size(struct device *dev); > > #ifdef CONFIG_HAS_DMA > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > +struct dma_pfn_offset_region { > + phys_addr_t cpu_beg; > + phys_addr_t cpu_end; > + dma_addr_t dma_beg; > + dma_addr_t dma_end; Perhaps s,beg,start, in above names > + unsigned long pfn_offset; > +}; > + > +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, > + dma_addr_t dma_addr) > +{ > + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; > + if (m) > + for (; m->cpu_end; m++) Why not simple while (m) { ... } ? > + if (dma_addr >= m->dma_beg && dma_addr < m->dma_end) > + return m->pfn_offset; > + return 0; > +} > + > +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, > + phys_addr_t paddr) > +{ > + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; > + > + if (m) > + for (; m->cpu_end; m++) Ditto. > + if (paddr >= m->cpu_beg && paddr < m->cpu_end) > + return m->pfn_offset; > + return 0; > +} > +#else /* CONFIG_DMA_PFN_OFFSET_MAP */ > +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, > + dma_addr_t dma_addr) > +{ > + return 0; > +} > + > +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, > + phys_addr_t paddr) > +{ > + return 0; > +} > +#endif /* CONFIG_DMA_PFN_OFFSET_MAP */ > + > #include <asm/dma-mapping.h> > > static inline const struct dma_map_ops *get_dma_ops(struct device *dev) > diff --git a/kernel/dma/Kconfig b/kernel/dma/Kconfig > index 4c103a24e380..ceb7e5e8f501 100644 > --- a/kernel/dma/Kconfig > +++ b/kernel/dma/Kconfig > @@ -195,3 +195,16 @@ config DMA_API_DEBUG_SG > is technically out-of-spec. > > If unsure, say N. > + > +config DMA_PFN_OFFSET_MAP > + bool "Uses a DMA range map to calculate PFN offset" > + depends on PCIE_BRCMSTB > + default n Redundant. > + help > + Some devices have a dma-range that gets converted to > + a dev->dma_pfn_offset value. This option is for the > + atypical case of there being multiple dma-ranges requiring > + multiple pfn offsets, which are selected from when > + converting to phys to dma and vice versa. > + > + If unsure, say N. > -- > 2.17.1 >
Hello Andy, On Tue, May 26, 2020 at 4:54 PM Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote: > > On Tue, May 26, 2020 at 03:12:48PM -0400, Jim Quinlan wrote: > > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > > similar to 'dma_pfn_offset' except that the offset chosen depends on the > > cpu or dma address involved. > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > > --- > > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > > drivers/usb/core/message.c | 3 ++ > > drivers/usb/core/usb.c | 3 ++ > > include/linux/device.h | 10 +++++- > > include/linux/dma-direct.h | 10 ++++-- > > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > > kernel/dma/Kconfig | 13 ++++++++ > > 7 files changed, 144 insertions(+), 6 deletions(-) > > > > diff --git a/drivers/of/address.c b/drivers/of/address.c > > index 96d8cfb14a60..a01afffcde7d 100644 > > --- a/drivers/of/address.c > > +++ b/drivers/of/address.c > > @@ -918,6 +918,47 @@ void __iomem *of_io_request_and_map(struct device_node *np, int index, > > } > > EXPORT_SYMBOL(of_io_request_and_map); > > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > +static int attach_dma_pfn_offset_map(struct device *dev, > > + struct device_node *node, int num_ranges) > > +{ > > + struct of_range_parser parser; > > + struct of_range range; > > + size_t r_size = (num_ranges + 1) > > + * sizeof(struct dma_pfn_offset_region); > > + struct dma_pfn_offset_region *r; > > + > > > + r = devm_kzalloc(dev, r_size, GFP_KERNEL); > > devm_?! Yes, otherwise if the device gets unbound/bound repeatedly then there would be a memory leak. > > > Looking at r_size it should be rather kcalloc(). Yep. > > > > + if (!r) > > + return -ENOMEM; > > + dev->dma_pfn_offset_map = r; > > + of_dma_range_parser_init(&parser, node); > > + > > + /* > > + * Record all info for DMA ranges array. We could > > + * just use the of_range struct, but if we did that it > > + * would require more calculations for phys_to_dma and > > + * dma_to_phys conversions. > > + */ > > + for_each_of_range(&parser, &range) { > > + r->cpu_beg = range.cpu_addr; > > + r->cpu_end = r->cpu_beg + range.size; > > + r->dma_beg = range.bus_addr; > > + r->dma_end = r->dma_beg + range.size; > > + r->pfn_offset = PFN_DOWN(range.cpu_addr) > > + - PFN_DOWN(range.bus_addr); > > + r++; > > + } > > + return 0; > > +} > > +#else > > +static int attach_dma_pfn_offset_map(struct device *dev, > > + struct device_node *node, int num_ranges) > > +{ > > + return 0; > > +} > > +#endif > > + > > /** > > * of_dma_get_range - Get DMA range info > > * @dev: device pointer; only needed for a corner case. > > @@ -947,6 +988,8 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > > struct of_range_parser parser; > > struct of_range range; > > u64 dma_start = U64_MAX, dma_end = 0, dma_offset = 0; > > + bool dma_multi_pfn_offset = false; > > + int num_ranges = 0; > > > > while (node) { > > ranges = of_get_property(node, "dma-ranges", &len); > > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > > range.bus_addr, range.cpu_addr, range.size); > > > > + num_ranges++; > > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > > + pr_warn("Can't handle multiple dma-ranges with different offsets on node(%pOF)\n", node); > > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > > + /* > > + * Don't error out as we'd break some existing > > + * DTs that are using configs w/o > > + * CONFIG_DMA_PFN_OFFSET_MAP set. > > + */ > > + continue; > > + } > > + dma_multi_pfn_offset = true; > > } > > dma_offset = range.cpu_addr - range.bus_addr; > > > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, > > dma_end = range.bus_addr + range.size; > > } > > > > + if (dma_multi_pfn_offset) { > > + dma_offset = 0; > > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > > + if (ret) > > + return ret; > > + } > > + > > if (dma_start >= dma_end) { > > ret = -EINVAL; > > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > > index 6197938dcc2d..aaa3e58f5eb4 100644 > > --- a/drivers/usb/core/message.c > > +++ b/drivers/usb/core/message.c > > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int configuration) > > */ > > intf->dev.dma_mask = dev->dev.dma_mask; > > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > > +#endif > > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > > intf->minor = -1; > > device_initialize(&intf->dev); > > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > > index f16c26dc079d..d2ed4d90e56e 100644 > > --- a/drivers/usb/core/usb.c > > +++ b/drivers/usb/core/usb.c > > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent, > > */ > > dev->dev.dma_mask = bus->sysdev->dma_mask; > > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > > +#endif > > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > > dev->state = USB_STATE_ATTACHED; > > dev->lpm_disable_count = 1; > > diff --git a/include/linux/device.h b/include/linux/device.h > > index ac8e37cd716a..67a240ad4fc5 100644 > > --- a/include/linux/device.h > > +++ b/include/linux/device.h > > @@ -493,6 +493,8 @@ struct dev_links_info { > > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are multiple > > + * pfn offsets for multiple dma-ranges. > > * @dma_parms: A low level driver may set these to teach IOMMU code about > > * segment limitations. > > * @dma_pools: Dma pools (if dma'ble device). > > @@ -578,7 +580,13 @@ struct device { > > allocations such descriptors. */ > > u64 bus_dma_limit; /* upstream dma constraint */ > > unsigned long dma_pfn_offset; > > - > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > > > + /* Like dma_pfn_offset, but for > > + * the unlikely case of multiple > > + * offsets. If non-null, dma_pfn_offset > > + * will be set to 0. */ > > A bit harder to read comment indented too much and located after the declared variable. Okay, will change. I was trying to keep the comment style of the other variables. > > +#endif > > struct device_dma_parameters *dma_parms; > > > > struct list_head dma_pools; /* dma pools (if dma'ble) */ > > diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h > > index 24b8684aa21d..03110a57eabc 100644 > > --- a/include/linux/dma-direct.h > > +++ b/include/linux/dma-direct.h > > @@ -14,15 +14,21 @@ extern unsigned int zone_dma_bits; > > static inline dma_addr_t __phys_to_dma(struct device *dev, phys_addr_t paddr) > > { > > dma_addr_t dev_addr = (dma_addr_t)paddr; > > + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ > > + unsigned long dma_pfn_offset = dev->dma_pfn_offset > > + + dma_pfn_offset_from_phys_addr(dev, paddr); > > > > - return dev_addr - ((dma_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); > > + return dev_addr - ((dma_addr_t)dma_pfn_offset << PAGE_SHIFT); > > } > > > > static inline phys_addr_t __dma_to_phys(struct device *dev, dma_addr_t dev_addr) > > { > > phys_addr_t paddr = (phys_addr_t)dev_addr; > > + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ > > + unsigned long dma_pfn_offset = dev->dma_pfn_offset > > + + dma_pfn_offset_from_dma_addr(dev, paddr); > > > > - return paddr + ((phys_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); > > + return paddr + ((phys_addr_t)dma_pfn_offset << PAGE_SHIFT); > > } > > #endif /* !CONFIG_ARCH_HAS_PHYS_TO_DMA */ > > > > diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h > > index 330ad58fbf4d..91940bba2229 100644 > > --- a/include/linux/dma-mapping.h > > +++ b/include/linux/dma-mapping.h > > @@ -256,6 +256,52 @@ static inline void dma_direct_sync_sg_for_cpu(struct device *dev, > > size_t dma_direct_max_mapping_size(struct device *dev); > > > > #ifdef CONFIG_HAS_DMA > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > +struct dma_pfn_offset_region { > > > + phys_addr_t cpu_beg; > > + phys_addr_t cpu_end; > > + dma_addr_t dma_beg; > > + dma_addr_t dma_end; > > Perhaps > s,beg,start, > in above names > Okay. > > > + unsigned long pfn_offset; > > +}; > > + > > +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, > > + dma_addr_t dma_addr) > > +{ > > + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; > > > + if (m) > > + for (; m->cpu_end; m++) > > Why not simple > > while (m) { > ... > } > > ? > That won't work; 'm' is either null or a valid pointer to an array which has an additional entry that is 0-filled.. If non-null, 'm' will never turn into NULL via 'm++' and the while loop will not terminate. > > > > + if (dma_addr >= m->dma_beg && dma_addr < m->dma_end) > > + return m->pfn_offset; > > + return 0; > > +} > > + > > +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, > > + phys_addr_t paddr) > > +{ > > + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; > > + > > > + if (m) > > + for (; m->cpu_end; m++) > > Ditto. > > > + if (paddr >= m->cpu_beg && paddr < m->cpu_end) > > + return m->pfn_offset; > > + return 0; > > +} > > +#else /* CONFIG_DMA_PFN_OFFSET_MAP */ > > +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, > > + dma_addr_t dma_addr) > > +{ > > + return 0; > > +} > > + > > +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, > > + phys_addr_t paddr) > > +{ > > + return 0; > > +} > > +#endif /* CONFIG_DMA_PFN_OFFSET_MAP */ > > + > > #include <asm/dma-mapping.h> > > > > static inline const struct dma_map_ops *get_dma_ops(struct device *dev) > > diff --git a/kernel/dma/Kconfig b/kernel/dma/Kconfig > > index 4c103a24e380..ceb7e5e8f501 100644 > > --- a/kernel/dma/Kconfig > > +++ b/kernel/dma/Kconfig > > @@ -195,3 +195,16 @@ config DMA_API_DEBUG_SG > > is technically out-of-spec. > > > > If unsure, say N. > > + > > +config DMA_PFN_OFFSET_MAP > > + bool "Uses a DMA range map to calculate PFN offset" > > + depends on PCIE_BRCMSTB > > > + default n > > Redundant. Okay. > > > + help > > + Some devices have a dma-range that gets converted to > > + a dev->dma_pfn_offset value. This option is for the > > + atypical case of there being multiple dma-ranges requiring > > + multiple pfn offsets, which are selected from when > > + converting to phys to dma and vice versa. > > + > > + If unsure, say N. > > -- > > 2.17.1 > > > > -- > With Best Regards, > Andy Shevchenko Thanks! Jim Quinlan
Hi Jim, one thing comes to mind, there is a small test suite in drivers/of/unittest.c (specifically of_unittest_pci_dma_ranges()) you could extend it to include your use cases. On Tue, 2020-05-26 at 15:12 -0400, Jim Quinlan wrote: > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > similar to 'dma_pfn_offset' except that the offset chosen depends on the > cpu or dma address involved. > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > --- > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > drivers/usb/core/message.c | 3 ++ > drivers/usb/core/usb.c | 3 ++ > include/linux/device.h | 10 +++++- > include/linux/dma-direct.h | 10 ++++-- > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > kernel/dma/Kconfig | 13 ++++++++ > 7 files changed, 144 insertions(+), 6 deletions(-) > [...] > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct > device_node *np, u64 *dma_addr, > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > range.bus_addr, range.cpu_addr, range.size); > > + num_ranges++; > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > + pr_warn("Can't handle multiple dma-ranges with > different offsets on node(%pOF)\n", node); > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > + /* > + * Don't error out as we'd break some existing > + * DTs that are using configs w/o > + * CONFIG_DMA_PFN_OFFSET_MAP set. > + */ > + continue; dev->bus_dma_limit is set in of_dma_configure(), this function's caller, based on dma_start's value (set after this continue). So you'd be effectively setting the dev->bus_dma_limit to whatever we get from the first dma-range. This can be troublesome depending on how the dma-ranges are setup, for example if the first dma-range doesn't include the CMA area, in arm64 generally set as high as possible in ZONE_DMA32, that would render it useless for dma/{direct/swiotlb}. Again depending on the bus_dma_limit value, if smaller than ZONE_DMA you'd be unable to allocate any DMA memory. IMO, a solution to this calls for a revamp of dma-direct's dma_capable(): match the target DMA memory area with each dma-range we have to see if it fits. > + } > + dma_multi_pfn_offset = true; > } > dma_offset = range.cpu_addr - range.bus_addr; > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct > device_node *np, u64 *dma_addr, > dma_end = range.bus_addr + range.size; > } > > + if (dma_multi_pfn_offset) { > + dma_offset = 0; > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > + if (ret) > + return ret; > + } > + > if (dma_start >= dma_end) { > ret = -EINVAL; > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > index 6197938dcc2d..aaa3e58f5eb4 100644 > --- a/drivers/usb/core/message.c > +++ b/drivers/usb/core/message.c > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int > configuration) > */ > intf->dev.dma_mask = dev->dev.dma_mask; > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > +#endif Thanks for looking at this, that said, I see more instances of drivers changing dma_pfn_offset outside of the core code. Why not doing this there too? Also, are we 100% sure that dev->dev.dma_pfn_offset isn't going to be freed before we're done using intf->dev? Maybe it's safer to copy the ranges? > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > intf->minor = -1; > device_initialize(&intf->dev); > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > index f16c26dc079d..d2ed4d90e56e 100644 > --- a/drivers/usb/core/usb.c > +++ b/drivers/usb/core/usb.c > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device > *parent, > */ > dev->dev.dma_mask = bus->sysdev->dma_mask; > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > +#endif > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > dev->state = USB_STATE_ATTACHED; > dev->lpm_disable_count = 1; > diff --git a/include/linux/device.h b/include/linux/device.h > index ac8e37cd716a..67a240ad4fc5 100644 > --- a/include/linux/device.h > +++ b/include/linux/device.h > @@ -493,6 +493,8 @@ struct dev_links_info { > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are > multiple > + * pfn offsets for multiple dma-ranges. > * @dma_parms: A low level driver may set these to teach IOMMU code > about > * segment limitations. > * @dma_pools: Dma pools (if dma'ble device). > @@ -578,7 +580,13 @@ struct device { > allocations such descriptors. */ > u64 bus_dma_limit; /* upstream dma constraint */ > unsigned long dma_pfn_offset; > - > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > + /* Like dma_pfn_offset, but for > + * the unlikely case of multiple > + * offsets. If non-null, dma_pfn_offset > + * will be set to 0. */ > +#endif I'm still sad this doesn't fully replace dma_pfn_offset & bus_dma_limit. I feel the extra logic involved in incorporating this as default isn't going to be noticeable as far as performance is concerned to single dma-range users, and it'd make for a nicer DMA code. Also you'd force everyone to test their changes on the multi dma-ranges code path, as opposed to having this disabled 99.9% of the time (hence broken every so often). Note that I sympathize with the amount of work involved on improving that, so better wait to hear what more knowledgeable people have to say about this :) Regards, Nicolas
Hi Nicolas, On Wed, May 27, 2020 at 11:00 AM Nicolas Saenz Julienne <nsaenzjulienne@suse.de> wrote: > > Hi Jim, > one thing comes to mind, there is a small test suite in drivers/of/unittest.c > (specifically of_unittest_pci_dma_ranges()) you could extend it to include your > use cases. Sure, will check out. > > On Tue, 2020-05-26 at 15:12 -0400, Jim Quinlan wrote: > > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > > similar to 'dma_pfn_offset' except that the offset chosen depends on the > > cpu or dma address involved. > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > > --- > > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > > drivers/usb/core/message.c | 3 ++ > > drivers/usb/core/usb.c | 3 ++ > > include/linux/device.h | 10 +++++- > > include/linux/dma-direct.h | 10 ++++-- > > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > > kernel/dma/Kconfig | 13 ++++++++ > > 7 files changed, 144 insertions(+), 6 deletions(-) > > > > [...] > > > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct > > device_node *np, u64 *dma_addr, > > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > > range.bus_addr, range.cpu_addr, range.size); > > > > + num_ranges++; > > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > > + pr_warn("Can't handle multiple dma-ranges with > > different offsets on node(%pOF)\n", node); > > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > > + /* > > + * Don't error out as we'd break some existing > > + * DTs that are using configs w/o > > + * CONFIG_DMA_PFN_OFFSET_MAP set. > > + */ > > + continue; > > dev->bus_dma_limit is set in of_dma_configure(), this function's caller, based > on dma_start's value (set after this continue). So you'd be effectively setting > the dev->bus_dma_limit to whatever we get from the first dma-range. I'm not seeing that at all. On the evaluation of each dma-range, dma_start and dma_end are re-evaluated to be the lowest and highest bus values of the dma-ranges seen so far. After all dma-ranges are examined, dev->bus_dma_limit being set to the highest. In fact, the current code -- ie before my commits -- already does this for multiple dma-ranges as long as the cpu-bus offset is the same in the dma-ranges. > > This can be troublesome depending on how the dma-ranges are setup, for example > if the first dma-range doesn't include the CMA area, in arm64 generally set as > high as possible in ZONE_DMA32, that would render it useless for > dma/{direct/swiotlb}. Again depending on the bus_dma_limit value, if smaller > than ZONE_DMA you'd be unable to allocate any DMA memory. > > IMO, a solution to this calls for a revamp of dma-direct's dma_capable(): match > the target DMA memory area with each dma-range we have to see if it fits. > > > + } > > + dma_multi_pfn_offset = true; > > } > > dma_offset = range.cpu_addr - range.bus_addr; > > > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct > > device_node *np, u64 *dma_addr, > > dma_end = range.bus_addr + range.size; > > } > > > > + if (dma_multi_pfn_offset) { > > + dma_offset = 0; > > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > > + if (ret) > > + return ret; > > + } > > + > > if (dma_start >= dma_end) { > > ret = -EINVAL; > > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > > index 6197938dcc2d..aaa3e58f5eb4 100644 > > --- a/drivers/usb/core/message.c > > +++ b/drivers/usb/core/message.c > > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int > > configuration) > > */ > > intf->dev.dma_mask = dev->dev.dma_mask; > > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > > +#endif > > Thanks for looking at this, that said, I see more instances of drivers changing > dma_pfn_offset outside of the core code. Why not doing this there too? > > Also, are we 100% sure that dev->dev.dma_pfn_offset isn't going to be freed > before we're done using intf->dev? Maybe it's safer to copy the ranges? > > > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > > intf->minor = -1; > > device_initialize(&intf->dev); > > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > > index f16c26dc079d..d2ed4d90e56e 100644 > > --- a/drivers/usb/core/usb.c > > +++ b/drivers/usb/core/usb.c > > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device > > *parent, > > */ > > dev->dev.dma_mask = bus->sysdev->dma_mask; > > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > > +#endif > > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > > dev->state = USB_STATE_ATTACHED; > > dev->lpm_disable_count = 1; > > diff --git a/include/linux/device.h b/include/linux/device.h > > index ac8e37cd716a..67a240ad4fc5 100644 > > --- a/include/linux/device.h > > +++ b/include/linux/device.h > > @@ -493,6 +493,8 @@ struct dev_links_info { > > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are > > multiple > > + * pfn offsets for multiple dma-ranges. > > * @dma_parms: A low level driver may set these to teach IOMMU code > > about > > * segment limitations. > > * @dma_pools: Dma pools (if dma'ble device). > > @@ -578,7 +580,13 @@ struct device { > > allocations such descriptors. */ > > u64 bus_dma_limit; /* upstream dma constraint */ > > unsigned long dma_pfn_offset; > > - > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > > + /* Like dma_pfn_offset, but for > > + * the unlikely case of multiple > > + * offsets. If non-null, dma_pfn_offset > > + * will be set to 0. */ > > +#endif > > I'm still sad this doesn't fully replace dma_pfn_offset & bus_dma_limit. I feel > the extra logic involved in incorporating this as default isn't going to be > noticeable as far as performance is concerned to single dma-range users, and > it'd make for a nicer DMA code. Also you'd force everyone to test their changes > on the multi dma-ranges code path, as opposed to having this disabled 99.9% of > the time (hence broken every so often). Good point. > > Note that I sympathize with the amount of work involved on improving that, so > better wait to hear what more knowledgeable people have to say about this :) Yes, I agree. I want to avoid coding and testing one solution only to have a different reviewer NAK it. Many thanks, Jim > > Regards, > Nicolas >
Hi Jim, On Wed, 2020-05-27 at 11:43 -0400, Jim Quinlan wrote: > Hi Nicolas, > > On Wed, May 27, 2020 at 11:00 AM Nicolas Saenz Julienne > <nsaenzjulienne@suse.de> wrote: > > Hi Jim, > > one thing comes to mind, there is a small test suite in > > drivers/of/unittest.c > > (specifically of_unittest_pci_dma_ranges()) you could extend it to include > > your > > use cases. > Sure, will check out. > > On Tue, 2020-05-26 at 15:12 -0400, Jim Quinlan wrote: > > > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > > > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > > > similar to 'dma_pfn_offset' except that the offset chosen depends on the > > > cpu or dma address involved. > > > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > > > --- > > > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > > > drivers/usb/core/message.c | 3 ++ > > > drivers/usb/core/usb.c | 3 ++ > > > include/linux/device.h | 10 +++++- > > > include/linux/dma-direct.h | 10 ++++-- > > > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > > > kernel/dma/Kconfig | 13 ++++++++ > > > 7 files changed, 144 insertions(+), 6 deletions(-) > > > > > > > [...] > > > > > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct > > > device_node *np, u64 *dma_addr, > > > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > > > range.bus_addr, range.cpu_addr, range.size); > > > > > > + num_ranges++; > > > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > > > + pr_warn("Can't handle multiple dma-ranges > > > with > > > different offsets on node(%pOF)\n", node); > > > + pr_warn("Perhaps set > > > DMA_PFN_OFFSET_MAP=y?\n"); > > > + /* > > > + * Don't error out as we'd break some > > > existing > > > + * DTs that are using configs w/o > > > + * CONFIG_DMA_PFN_OFFSET_MAP set. > > > + */ > > > + continue; > > > > dev->bus_dma_limit is set in of_dma_configure(), this function's caller, > > based > > on dma_start's value (set after this continue). So you'd be effectively > > setting > > the dev->bus_dma_limit to whatever we get from the first dma-range. > I'm not seeing that at all. On the evaluation of each dma-range, > dma_start and dma_end are re-evaluated to be the lowest and highest > bus values of the dma-ranges seen so far. After all dma-ranges are > examined, dev->bus_dma_limit being set to the highest. In fact, the > current code -- ie before my commits -- already does this for multiple > dma-ranges as long as the cpu-bus offset is the same in the > dma-ranges. Sorry I got carried away, you're right. So I understand there is an underlaying assumption that the non DMAble memory space will always sit on top of the bus memory space, as intertwined as it might be, so as to every phys_to_dma() call on non DMAble memory to fall above bus_dma_limit. Regards, Nicolas
On Wed, May 27, 2020 at 9:43 AM Jim Quinlan <james.quinlan@broadcom.com> wrote: > > Hi Nicolas, > > On Wed, May 27, 2020 at 11:00 AM Nicolas Saenz Julienne > <nsaenzjulienne@suse.de> wrote: > > > > Hi Jim, > > one thing comes to mind, there is a small test suite in drivers/of/unittest.c > > (specifically of_unittest_pci_dma_ranges()) you could extend it to include your > > use cases. > Sure, will check out. > > > > On Tue, 2020-05-26 at 15:12 -0400, Jim Quinlan wrote: > > > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > > > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > > > similar to 'dma_pfn_offset' except that the offset chosen depends on the > > > cpu or dma address involved. > > > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > > > --- > > > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > > > drivers/usb/core/message.c | 3 ++ > > > drivers/usb/core/usb.c | 3 ++ > > > include/linux/device.h | 10 +++++- > > > include/linux/dma-direct.h | 10 ++++-- > > > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > > > kernel/dma/Kconfig | 13 ++++++++ > > > 7 files changed, 144 insertions(+), 6 deletions(-) > > > > > > > [...] > > > > > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct > > > device_node *np, u64 *dma_addr, > > > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > > > range.bus_addr, range.cpu_addr, range.size); > > > > > > + num_ranges++; > > > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > > > + pr_warn("Can't handle multiple dma-ranges with > > > different offsets on node(%pOF)\n", node); > > > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > > > + /* > > > + * Don't error out as we'd break some existing > > > + * DTs that are using configs w/o > > > + * CONFIG_DMA_PFN_OFFSET_MAP set. > > > + */ > > > + continue; > > > > dev->bus_dma_limit is set in of_dma_configure(), this function's caller, based > > on dma_start's value (set after this continue). So you'd be effectively setting > > the dev->bus_dma_limit to whatever we get from the first dma-range. > I'm not seeing that at all. On the evaluation of each dma-range, > dma_start and dma_end are re-evaluated to be the lowest and highest > bus values of the dma-ranges seen so far. After all dma-ranges are > examined, dev->bus_dma_limit being set to the highest. In fact, the > current code -- ie before my commits -- already does this for multiple > dma-ranges as long as the cpu-bus offset is the same in the > dma-ranges. > > > > This can be troublesome depending on how the dma-ranges are setup, for example > > if the first dma-range doesn't include the CMA area, in arm64 generally set as > > high as possible in ZONE_DMA32, that would render it useless for > > dma/{direct/swiotlb}. Again depending on the bus_dma_limit value, if smaller > > than ZONE_DMA you'd be unable to allocate any DMA memory. > > > > IMO, a solution to this calls for a revamp of dma-direct's dma_capable(): match > > the target DMA memory area with each dma-range we have to see if it fits. > > > > > + } > > > + dma_multi_pfn_offset = true; > > > } > > > dma_offset = range.cpu_addr - range.bus_addr; > > > > > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct > > > device_node *np, u64 *dma_addr, > > > dma_end = range.bus_addr + range.size; > > > } > > > > > > + if (dma_multi_pfn_offset) { > > > + dma_offset = 0; > > > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > > > + if (ret) > > > + return ret; > > > + } > > > + > > > if (dma_start >= dma_end) { > > > ret = -EINVAL; > > > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > > > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > > > index 6197938dcc2d..aaa3e58f5eb4 100644 > > > --- a/drivers/usb/core/message.c > > > +++ b/drivers/usb/core/message.c > > > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int > > > configuration) > > > */ > > > intf->dev.dma_mask = dev->dev.dma_mask; > > > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > > > +#endif > > > > Thanks for looking at this, that said, I see more instances of drivers changing > > dma_pfn_offset outside of the core code. Why not doing this there too? > > > > Also, are we 100% sure that dev->dev.dma_pfn_offset isn't going to be freed > > before we're done using intf->dev? Maybe it's safer to copy the ranges? > > > > > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > > > intf->minor = -1; > > > device_initialize(&intf->dev); > > > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > > > index f16c26dc079d..d2ed4d90e56e 100644 > > > --- a/drivers/usb/core/usb.c > > > +++ b/drivers/usb/core/usb.c > > > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device > > > *parent, > > > */ > > > dev->dev.dma_mask = bus->sysdev->dma_mask; > > > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > > > +#endif > > > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > > > dev->state = USB_STATE_ATTACHED; > > > dev->lpm_disable_count = 1; > > > diff --git a/include/linux/device.h b/include/linux/device.h > > > index ac8e37cd716a..67a240ad4fc5 100644 > > > --- a/include/linux/device.h > > > +++ b/include/linux/device.h > > > @@ -493,6 +493,8 @@ struct dev_links_info { > > > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are > > > multiple > > > + * pfn offsets for multiple dma-ranges. > > > * @dma_parms: A low level driver may set these to teach IOMMU code > > > about > > > * segment limitations. > > > * @dma_pools: Dma pools (if dma'ble device). > > > @@ -578,7 +580,13 @@ struct device { > > > allocations such descriptors. */ > > > u64 bus_dma_limit; /* upstream dma constraint */ > > > unsigned long dma_pfn_offset; > > > - > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > > > + /* Like dma_pfn_offset, but for > > > + * the unlikely case of multiple > > > + * offsets. If non-null, dma_pfn_offset > > > + * will be set to 0. */ > > > +#endif > > > > I'm still sad this doesn't fully replace dma_pfn_offset & bus_dma_limit. I feel > > the extra logic involved in incorporating this as default isn't going to be > > noticeable as far as performance is concerned to single dma-range users, and > > it'd make for a nicer DMA code. Also you'd force everyone to test their changes > > on the multi dma-ranges code path, as opposed to having this disabled 99.9% of > > the time (hence broken every so often). > Good point. +1 > > Note that I sympathize with the amount of work involved on improving that, so > > better wait to hear what more knowledgeable people have to say about this :) > Yes, I agree. I want to avoid coding and testing one solution only to > have a different reviewer NAK it. It's a pretty safe bet that everyone will prefer one code path over 2. Rob
On Fri, May 29, 2020 at 1:35 PM Rob Herring <robh+dt@kernel.org> wrote: > > On Wed, May 27, 2020 at 9:43 AM Jim Quinlan <james.quinlan@broadcom.com> wrote: > > > > Hi Nicolas, > > > > On Wed, May 27, 2020 at 11:00 AM Nicolas Saenz Julienne > > <nsaenzjulienne@suse.de> wrote: > > > > > > Hi Jim, > > > one thing comes to mind, there is a small test suite in drivers/of/unittest.c > > > (specifically of_unittest_pci_dma_ranges()) you could extend it to include your > > > use cases. > > Sure, will check out. > > > > > > On Tue, 2020-05-26 at 15:12 -0400, Jim Quinlan wrote: > > > > The new field in struct device 'dma_pfn_offset_map' is used to facilitate > > > > the use of multiple pfn offsets between cpu addrs and dma addrs. It is > > > > similar to 'dma_pfn_offset' except that the offset chosen depends on the > > > > cpu or dma address involved. > > > > > > > > Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> > > > > --- > > > > drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- > > > > drivers/usb/core/message.c | 3 ++ > > > > drivers/usb/core/usb.c | 3 ++ > > > > include/linux/device.h | 10 +++++- > > > > include/linux/dma-direct.h | 10 ++++-- > > > > include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ > > > > kernel/dma/Kconfig | 13 ++++++++ > > > > 7 files changed, 144 insertions(+), 6 deletions(-) > > > > > > > > > > [...] > > > > > > > @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct > > > > device_node *np, u64 *dma_addr, > > > > pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", > > > > range.bus_addr, range.cpu_addr, range.size); > > > > > > > > + num_ranges++; > > > > 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { > > > > + pr_warn("Can't handle multiple dma-ranges with > > > > different offsets on node(%pOF)\n", node); > > > > + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); > > > > + /* > > > > + * Don't error out as we'd break some existing > > > > + * DTs that are using configs w/o > > > > + * CONFIG_DMA_PFN_OFFSET_MAP set. > > > > + */ > > > > + continue; > > > > > > dev->bus_dma_limit is set in of_dma_configure(), this function's caller, based > > > on dma_start's value (set after this continue). So you'd be effectively setting > > > the dev->bus_dma_limit to whatever we get from the first dma-range. > > I'm not seeing that at all. On the evaluation of each dma-range, > > dma_start and dma_end are re-evaluated to be the lowest and highest > > bus values of the dma-ranges seen so far. After all dma-ranges are > > examined, dev->bus_dma_limit being set to the highest. In fact, the > > current code -- ie before my commits -- already does this for multiple > > dma-ranges as long as the cpu-bus offset is the same in the > > dma-ranges. > > > > > > This can be troublesome depending on how the dma-ranges are setup, for example > > > if the first dma-range doesn't include the CMA area, in arm64 generally set as > > > high as possible in ZONE_DMA32, that would render it useless for > > > dma/{direct/swiotlb}. Again depending on the bus_dma_limit value, if smaller > > > than ZONE_DMA you'd be unable to allocate any DMA memory. > > > > > > IMO, a solution to this calls for a revamp of dma-direct's dma_capable(): match > > > the target DMA memory area with each dma-range we have to see if it fits. > > > > > > > + } > > > > + dma_multi_pfn_offset = true; > > > > } > > > > dma_offset = range.cpu_addr - range.bus_addr; > > > > > > > > @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct > > > > device_node *np, u64 *dma_addr, > > > > dma_end = range.bus_addr + range.size; > > > > } > > > > > > > > + if (dma_multi_pfn_offset) { > > > > + dma_offset = 0; > > > > + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); > > > > + if (ret) > > > > + return ret; > > > > + } > > > > + > > > > if (dma_start >= dma_end) { > > > > ret = -EINVAL; > > > > pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", > > > > diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c > > > > index 6197938dcc2d..aaa3e58f5eb4 100644 > > > > --- a/drivers/usb/core/message.c > > > > +++ b/drivers/usb/core/message.c > > > > @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int > > > > configuration) > > > > */ > > > > intf->dev.dma_mask = dev->dev.dma_mask; > > > > intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; > > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > > + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; > > > > +#endif > > > > > > Thanks for looking at this, that said, I see more instances of drivers changing > > > dma_pfn_offset outside of the core code. Why not doing this there too? > > > > > > Also, are we 100% sure that dev->dev.dma_pfn_offset isn't going to be freed > > > before we're done using intf->dev? Maybe it's safer to copy the ranges? > > > > > > > INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); > > > > intf->minor = -1; > > > > device_initialize(&intf->dev); > > > > diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c > > > > index f16c26dc079d..d2ed4d90e56e 100644 > > > > --- a/drivers/usb/core/usb.c > > > > +++ b/drivers/usb/core/usb.c > > > > @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device > > > > *parent, > > > > */ > > > > dev->dev.dma_mask = bus->sysdev->dma_mask; > > > > dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; > > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > > + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; > > > > +#endif > > > > set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); > > > > dev->state = USB_STATE_ATTACHED; > > > > dev->lpm_disable_count = 1; > > > > diff --git a/include/linux/device.h b/include/linux/device.h > > > > index ac8e37cd716a..67a240ad4fc5 100644 > > > > --- a/include/linux/device.h > > > > +++ b/include/linux/device.h > > > > @@ -493,6 +493,8 @@ struct dev_links_info { > > > > * @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_pfn_offset_map: Like dma_pfn_offset but used when there are > > > > multiple > > > > + * pfn offsets for multiple dma-ranges. > > > > * @dma_parms: A low level driver may set these to teach IOMMU code > > > > about > > > > * segment limitations. > > > > * @dma_pools: Dma pools (if dma'ble device). > > > > @@ -578,7 +580,13 @@ struct device { > > > > allocations such descriptors. */ > > > > u64 bus_dma_limit; /* upstream dma constraint */ > > > > unsigned long dma_pfn_offset; > > > > - > > > > +#ifdef CONFIG_DMA_PFN_OFFSET_MAP > > > > + const struct dma_pfn_offset_region *dma_pfn_offset_map; > > > > + /* Like dma_pfn_offset, but for > > > > + * the unlikely case of multiple > > > > + * offsets. If non-null, dma_pfn_offset > > > > + * will be set to 0. */ > > > > +#endif > > > > > > I'm still sad this doesn't fully replace dma_pfn_offset & bus_dma_limit. I feel > > > the extra logic involved in incorporating this as default isn't going to be > > > noticeable as far as performance is concerned to single dma-range users, and > > > it'd make for a nicer DMA code. Also you'd force everyone to test their changes > > > on the multi dma-ranges code path, as opposed to having this disabled 99.9% of > > > the time (hence broken every so often). > > Good point. > > +1 > > > > Note that I sympathize with the amount of work involved on improving that, so > > > better wait to hear what more knowledgeable people have to say about this :) > > Yes, I agree. I want to avoid coding and testing one solution only to > > have a different reviewer NAK it. > > It's a pretty safe bet that everyone will prefer one code path over 2. > > Rob Thanks for the input. Will do, and send out v3 ASAP. Thanks, Jim
diff --git a/drivers/of/address.c b/drivers/of/address.c index 96d8cfb14a60..a01afffcde7d 100644 --- a/drivers/of/address.c +++ b/drivers/of/address.c @@ -918,6 +918,47 @@ void __iomem *of_io_request_and_map(struct device_node *np, int index, } EXPORT_SYMBOL(of_io_request_and_map); +#ifdef CONFIG_DMA_PFN_OFFSET_MAP +static int attach_dma_pfn_offset_map(struct device *dev, + struct device_node *node, int num_ranges) +{ + struct of_range_parser parser; + struct of_range range; + size_t r_size = (num_ranges + 1) + * sizeof(struct dma_pfn_offset_region); + struct dma_pfn_offset_region *r; + + r = devm_kzalloc(dev, r_size, GFP_KERNEL); + if (!r) + return -ENOMEM; + dev->dma_pfn_offset_map = r; + of_dma_range_parser_init(&parser, node); + + /* + * Record all info for DMA ranges array. We could + * just use the of_range struct, but if we did that it + * would require more calculations for phys_to_dma and + * dma_to_phys conversions. + */ + for_each_of_range(&parser, &range) { + r->cpu_beg = range.cpu_addr; + r->cpu_end = r->cpu_beg + range.size; + r->dma_beg = range.bus_addr; + r->dma_end = r->dma_beg + range.size; + r->pfn_offset = PFN_DOWN(range.cpu_addr) + - PFN_DOWN(range.bus_addr); + r++; + } + return 0; +} +#else +static int attach_dma_pfn_offset_map(struct device *dev, + struct device_node *node, int num_ranges) +{ + return 0; +} +#endif + /** * of_dma_get_range - Get DMA range info * @dev: device pointer; only needed for a corner case. @@ -947,6 +988,8 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, struct of_range_parser parser; struct of_range range; u64 dma_start = U64_MAX, dma_end = 0, dma_offset = 0; + bool dma_multi_pfn_offset = false; + int num_ranges = 0; while (node) { ranges = of_get_property(node, "dma-ranges", &len); @@ -977,10 +1020,19 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, pr_debug("dma_addr(%llx) cpu_addr(%llx) size(%llx)\n", range.bus_addr, range.cpu_addr, range.size); + num_ranges++; 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 (!IS_ENABLED(CONFIG_DMA_PFN_OFFSET_MAP)) { + pr_warn("Can't handle multiple dma-ranges with different offsets on node(%pOF)\n", node); + pr_warn("Perhaps set DMA_PFN_OFFSET_MAP=y?\n"); + /* + * Don't error out as we'd break some existing + * DTs that are using configs w/o + * CONFIG_DMA_PFN_OFFSET_MAP set. + */ + continue; + } + dma_multi_pfn_offset = true; } dma_offset = range.cpu_addr - range.bus_addr; @@ -991,6 +1043,13 @@ int of_dma_get_range(struct device *dev, struct device_node *np, u64 *dma_addr, dma_end = range.bus_addr + range.size; } + if (dma_multi_pfn_offset) { + dma_offset = 0; + ret = attach_dma_pfn_offset_map(dev, node, num_ranges); + if (ret) + return ret; + } + if (dma_start >= dma_end) { ret = -EINVAL; pr_debug("Invalid DMA ranges configuration on node(%pOF)\n", diff --git a/drivers/usb/core/message.c b/drivers/usb/core/message.c index 6197938dcc2d..aaa3e58f5eb4 100644 --- a/drivers/usb/core/message.c +++ b/drivers/usb/core/message.c @@ -1960,6 +1960,9 @@ int usb_set_configuration(struct usb_device *dev, int configuration) */ intf->dev.dma_mask = dev->dev.dma_mask; intf->dev.dma_pfn_offset = dev->dev.dma_pfn_offset; +#ifdef CONFIG_DMA_PFN_OFFSET_MAP + intf->dev.dma_pfn_offset_map = dev->dev.dma_pfn_offset_map; +#endif INIT_WORK(&intf->reset_ws, __usb_queue_reset_device); intf->minor = -1; device_initialize(&intf->dev); diff --git a/drivers/usb/core/usb.c b/drivers/usb/core/usb.c index f16c26dc079d..d2ed4d90e56e 100644 --- a/drivers/usb/core/usb.c +++ b/drivers/usb/core/usb.c @@ -612,6 +612,9 @@ struct usb_device *usb_alloc_dev(struct usb_device *parent, */ dev->dev.dma_mask = bus->sysdev->dma_mask; dev->dev.dma_pfn_offset = bus->sysdev->dma_pfn_offset; +#ifdef CONFIG_DMA_PFN_OFFSET_MAP + dev->dev.dma_pfn_offset_map = bus->sysdev->dma_pfn_offset_map; +#endif set_dev_node(&dev->dev, dev_to_node(bus->sysdev)); dev->state = USB_STATE_ATTACHED; dev->lpm_disable_count = 1; diff --git a/include/linux/device.h b/include/linux/device.h index ac8e37cd716a..67a240ad4fc5 100644 --- a/include/linux/device.h +++ b/include/linux/device.h @@ -493,6 +493,8 @@ struct dev_links_info { * @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_pfn_offset_map: Like dma_pfn_offset but used when there are multiple + * pfn offsets for multiple dma-ranges. * @dma_parms: A low level driver may set these to teach IOMMU code about * segment limitations. * @dma_pools: Dma pools (if dma'ble device). @@ -578,7 +580,13 @@ struct device { allocations such descriptors. */ u64 bus_dma_limit; /* upstream dma constraint */ unsigned long dma_pfn_offset; - +#ifdef CONFIG_DMA_PFN_OFFSET_MAP + const struct dma_pfn_offset_region *dma_pfn_offset_map; + /* Like dma_pfn_offset, but for + * the unlikely case of multiple + * offsets. If non-null, dma_pfn_offset + * will be set to 0. */ +#endif struct device_dma_parameters *dma_parms; struct list_head dma_pools; /* dma pools (if dma'ble) */ diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h index 24b8684aa21d..03110a57eabc 100644 --- a/include/linux/dma-direct.h +++ b/include/linux/dma-direct.h @@ -14,15 +14,21 @@ extern unsigned int zone_dma_bits; static inline dma_addr_t __phys_to_dma(struct device *dev, phys_addr_t paddr) { dma_addr_t dev_addr = (dma_addr_t)paddr; + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ + unsigned long dma_pfn_offset = dev->dma_pfn_offset + + dma_pfn_offset_from_phys_addr(dev, paddr); - return dev_addr - ((dma_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); + return dev_addr - ((dma_addr_t)dma_pfn_offset << PAGE_SHIFT); } static inline phys_addr_t __dma_to_phys(struct device *dev, dma_addr_t dev_addr) { phys_addr_t paddr = (phys_addr_t)dev_addr; + /* The compiler should remove the 2nd term if !DMA_PFN_OFFSET_MAP */ + unsigned long dma_pfn_offset = dev->dma_pfn_offset + + dma_pfn_offset_from_dma_addr(dev, paddr); - return paddr + ((phys_addr_t)dev->dma_pfn_offset << PAGE_SHIFT); + return paddr + ((phys_addr_t)dma_pfn_offset << PAGE_SHIFT); } #endif /* !CONFIG_ARCH_HAS_PHYS_TO_DMA */ diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index 330ad58fbf4d..91940bba2229 100644 --- a/include/linux/dma-mapping.h +++ b/include/linux/dma-mapping.h @@ -256,6 +256,52 @@ static inline void dma_direct_sync_sg_for_cpu(struct device *dev, size_t dma_direct_max_mapping_size(struct device *dev); #ifdef CONFIG_HAS_DMA +#ifdef CONFIG_DMA_PFN_OFFSET_MAP +struct dma_pfn_offset_region { + phys_addr_t cpu_beg; + phys_addr_t cpu_end; + dma_addr_t dma_beg; + dma_addr_t dma_end; + unsigned long pfn_offset; +}; + +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, + dma_addr_t dma_addr) +{ + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; + + if (m) + for (; m->cpu_end; m++) + if (dma_addr >= m->dma_beg && dma_addr < m->dma_end) + return m->pfn_offset; + return 0; +} + +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, + phys_addr_t paddr) +{ + const struct dma_pfn_offset_region *m = dev->dma_pfn_offset_map; + + if (m) + for (; m->cpu_end; m++) + if (paddr >= m->cpu_beg && paddr < m->cpu_end) + return m->pfn_offset; + return 0; +} +#else /* CONFIG_DMA_PFN_OFFSET_MAP */ +static inline unsigned long dma_pfn_offset_from_dma_addr(struct device *dev, + dma_addr_t dma_addr) +{ + return 0; +} + +static inline unsigned long dma_pfn_offset_from_phys_addr(struct device *dev, + phys_addr_t paddr) +{ + return 0; +} +#endif /* CONFIG_DMA_PFN_OFFSET_MAP */ + #include <asm/dma-mapping.h> static inline const struct dma_map_ops *get_dma_ops(struct device *dev) diff --git a/kernel/dma/Kconfig b/kernel/dma/Kconfig index 4c103a24e380..ceb7e5e8f501 100644 --- a/kernel/dma/Kconfig +++ b/kernel/dma/Kconfig @@ -195,3 +195,16 @@ config DMA_API_DEBUG_SG is technically out-of-spec. If unsure, say N. + +config DMA_PFN_OFFSET_MAP + bool "Uses a DMA range map to calculate PFN offset" + depends on PCIE_BRCMSTB + default n + help + Some devices have a dma-range that gets converted to + a dev->dma_pfn_offset value. This option is for the + atypical case of there being multiple dma-ranges requiring + multiple pfn offsets, which are selected from when + converting to phys to dma and vice versa. + + If unsure, say N.
The new field in struct device 'dma_pfn_offset_map' is used to facilitate the use of multiple pfn offsets between cpu addrs and dma addrs. It is similar to 'dma_pfn_offset' except that the offset chosen depends on the cpu or dma address involved. Signed-off-by: Jim Quinlan <james.quinlan@broadcom.com> --- drivers/of/address.c | 65 +++++++++++++++++++++++++++++++++++-- drivers/usb/core/message.c | 3 ++ drivers/usb/core/usb.c | 3 ++ include/linux/device.h | 10 +++++- include/linux/dma-direct.h | 10 ++++-- include/linux/dma-mapping.h | 46 ++++++++++++++++++++++++++ kernel/dma/Kconfig | 13 ++++++++ 7 files changed, 144 insertions(+), 6 deletions(-)