LinuxLists.cc - [PATCH] dma-mapping: move dma configuration to bus infrastructure

[permalink] [raw]

Subject: RE: [PATCH] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Sinan Kaya [mailto:[email protected]]
> Sent: Monday, March 12, 2018 22:14
> To: Nipun Gupta <[email protected]>; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Cc: [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected]
> foundation.org; [email protected]
> Subject: Re: [PATCH] dma-mapping: move dma configuration to bus
> infrastructure
>
> On 3/12/2018 11:24 AM, Nipun Gupta wrote:
> > + if (dma_dev->of_node) {
> > + ret = of_dma_configure(dev, dma_dev->of_node);
> > + } else if (has_acpi_companion(dma_dev)) {
> > + attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev-
> >fwnode));
> > + if (attr != DEV_DMA_NOT_SUPPORTED)
> > + ret = acpi_dma_configure(dev, attr);
> > + }
> > +
> > + pci_put_host_bridge_device(bridge);
> > +
> > + return ret;
> > +}
> > +
> > +void pci_dma_deconfigure(struct device *dev)
> > +{
> > + of_dma_deconfigure(dev);
> > + acpi_dma_deconfigure(dev);
> > +}
>
> Isn't this one or the other one but not both?
>
> Something like:
>
> if (dev->of_node)
> of_dma_deconfigure(dev);
> else
> acpi_dma_deconfigure(dev);
>
> should work.

I understand your point. Seems reasonable as we should not expect
the 'of/acpi DMA deconfigure' API to not fail when they are not configured.

But, here we would also need to get dma_device (just as we get in
'pci_dma_configure') and need a check on it as for PCI there 'of_node'
is present in the dma_dev.

Ill update this in v2, and also make similar changes for platform and AMBA bus.

Thanks,
Nipun

>
> --
> Sinan Kaya
> Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm
> Technologies, Inc.
> Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux
> Foundation Collaborative Project.

2018-03-13 07:29:32

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

On Tue, Mar 13, 2018 at 04:22:53AM +0000, Nipun Gupta wrote:
> > Isn't this one or the other one but not both?
> >
> > Something like:
> >
> > if (dev->of_node)
> > of_dma_deconfigure(dev);
> > else
> > acpi_dma_deconfigure(dev);
> >
> > should work.
>
> I understand your point. Seems reasonable as we should not expect
> the 'of/acpi DMA deconfigure' API to not fail when they are not configured.
>
> But, here we would also need to get dma_device (just as we get in
> 'pci_dma_configure') and need a check on it as for PCI there 'of_node'
> is present in the dma_dev.

Both of_dma_deconfigure and acpi_dma_deconfigure just end up calling
arch_teardown_dma_ops. So my preference would be to just remove
of_dma_deconfigure and acpi_dma_deconfigure and call arch_teardown_dma_ops
as a prep patch before this one.

2018-03-13 07:36:26

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

> +int amba_dma_configure(struct device *dev)
> +{
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + return ret;

This code sniplet is duplicated so many times that I think we should
just have some sort of dma_common_configure() for it that the various
busses can use.

> +void amba_dma_deconfigure(struct device *dev)
> +{
> + of_dma_deconfigure(dev);
> + acpi_dma_deconfigure(dev);
> +}

As mention in my previous reply I think we don't even need a deconfigure
callback at this point - just remove the ACPI and OF wrappers and
clear the dma ops.

Also in this series we should replace the force_dma flag by use of the
proper method, e.g. give a force parameter to of_dma_configure and the
new dma_common_configure helper that the busses that want it can set.

2018-03-13 11:37:16

by Robin Murphy

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

On 12/03/18 15:24, Nipun Gupta wrote:
> The change introduces 'dma_configure' & 'dma_deconfigure'as
> bus callback functions so each bus can choose to implement
> its own dma configuration function.
> This eases the addition of new busses w.r.t. adding the dma
> configuration functionality.

It's probably worth clarifying - either in the commit message, the
kerneldoc, or both - that the bus-specific aspect is that of mapping
between a given device on the bus and the relevant firmware description
of its DMA configuration.

> The change also updates the PCI, Platform and ACPI bus to use
> new introduced callbacks.
>
> Signed-off-by: Nipun Gupta <[email protected]>
> ---
> - This patch is based on the comments on:
> https://patchwork.kernel.org/patch/10259087/
> - I have validated for PCI and platform, but not for AMBA as I
> do not have infrastructure to validate it.
> Can anyone please validate them on AMBA?
>
> drivers/amba/bus.c | 38 ++++++++++++++++++++++++-----
> drivers/base/dd.c | 14 +++++++----
> drivers/base/dma-mapping.c | 41 -------------------------------
> drivers/base/platform.c | 36 ++++++++++++++++++++++-----
> drivers/pci/pci-driver.c | 59 ++++++++++++++++++++++++++++++++++++---------
> include/linux/device.h | 6 +++++
> include/linux/dma-mapping.h | 12 ---------
> 7 files changed, 124 insertions(+), 82 deletions(-)
>
> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
> index 594c228..58241d2 100644
> --- a/drivers/amba/bus.c
> +++ b/drivers/amba/bus.c
> @@ -20,6 +20,8 @@
> #include <linux/sizes.h>
> #include <linux/limits.h>
> #include <linux/clk/clk-conf.h>
> +#include <linux/acpi.h>
> +#include <linux/of_device.h>
>
> #include <asm/irq.h>
>
> @@ -171,6 +173,28 @@ static int amba_pm_runtime_resume(struct device *dev)
> }
> #endif /* CONFIG_PM */
>
> +int amba_dma_configure(struct device *dev)
> +{
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + return ret;
> +}

I would be inclined to have amba_bustype just reference
platform_dma_configure() directly rather than duplicate it like this,
since there's no sensible reason for them to ever differ.

> +
> +void amba_dma_deconfigure(struct device *dev)
> +{
> + of_dma_deconfigure(dev);
> + acpi_dma_deconfigure(dev);
> +}
> +
> static const struct dev_pm_ops amba_pm = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -190,12 +214,14 @@ static int amba_pm_runtime_resume(struct device *dev)
> * so we call the bus "amba".
> */
> struct bus_type amba_bustype = {
> - .name = "amba",
> - .dev_groups = amba_dev_groups,
> - .match = amba_match,
> - .uevent = amba_uevent,
> - .pm = &amba_pm,
> - .force_dma = true,
> + .name = "amba",
> + .dev_groups = amba_dev_groups,
> + .match = amba_match,
> + .uevent = amba_uevent,
> + .pm = &amba_pm,
> + .dma_configure = amba_dma_configure,
> + .dma_deconfigure = amba_dma_deconfigure,
> + .force_dma = true,

This patch should also be removing force_dma because it no longer makes
sense. If DMA configuration is now done by a bus-level callback, then a
bus which wants its children to get DMA configuration needs to implement
that callback; there's nowhere to force a "default" global behaviour any
more.

> };
>
> static int __init amba_init(void)
> diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> index de6fd09..f124f3f 100644
> --- a/drivers/base/dd.c
> +++ b/drivers/base/dd.c
> @@ -421,9 +421,11 @@ static int really_probe(struct device *dev, struct device_driver *drv)
> if (ret)
> goto pinctrl_bind_failed;
>
> - ret = dma_configure(dev);
> - if (ret)
> - goto dma_failed;
> + if (dev->bus->dma_configure) {
> + ret = dev->bus->dma_configure(dev);
> + if (ret)
> + goto dma_failed;
> + }
>
> if (driver_sysfs_add(dev)) {
> printk(KERN_ERR "%s: driver_sysfs_add(%s) failed\n",
> @@ -486,7 +488,8 @@ static int really_probe(struct device *dev, struct device_driver *drv)
> goto done;
>
> probe_failed:
> - dma_deconfigure(dev);
> + if (dev->bus->dma_deconfigure)
> + dev->bus->dma_deconfigure(dev);
> dma_failed:
> if (dev->bus)
> blocking_notifier_call_chain(&dev->bus->p->bus_notifier,
> @@ -895,7 +898,8 @@ static void __device_release_driver(struct device *dev, struct device *parent)
> drv->remove(dev);
>
> device_links_driver_cleanup(dev);
> - dma_deconfigure(dev);
> + if (dev->bus->dma_deconfigure)
> + dev->bus->dma_deconfigure(dev);
>
> devres_release_all(dev);
> dev->driver = NULL;
> diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
> index 3b11835..f16bd49 100644
> --- a/drivers/base/dma-mapping.c
> +++ b/drivers/base/dma-mapping.c
> @@ -6,11 +6,9 @@
> * Copyright (c) 2006 Tejun Heo <[email protected]>
> */
>
> -#include <linux/acpi.h>
> #include <linux/dma-mapping.h>
> #include <linux/export.h>
> #include <linux/gfp.h>
> -#include <linux/of_device.h>
> #include <linux/slab.h>
> #include <linux/vmalloc.h>
>
> @@ -329,42 +327,3 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
> vunmap(cpu_addr);
> }
> #endif
> -
> -/*
> - * Common configuration to enable DMA API use for a device
> - */
> -#include <linux/pci.h>
> -
> -int dma_configure(struct device *dev)
> -{
> - struct device *bridge = NULL, *dma_dev = dev;
> - enum dev_dma_attr attr;
> - int ret = 0;
> -
> - if (dev_is_pci(dev)) {
> - bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> - dma_dev = bridge;
> - if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> - dma_dev->parent->of_node)
> - dma_dev = dma_dev->parent;
> - }
> -
> - if (dma_dev->of_node) {
> - ret = of_dma_configure(dev, dma_dev->of_node);
> - } else if (has_acpi_companion(dma_dev)) {
> - attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> - if (attr != DEV_DMA_NOT_SUPPORTED)
> - ret = acpi_dma_configure(dev, attr);
> - }
> -
> - if (bridge)
> - pci_put_host_bridge_device(bridge);
> -
> - return ret;
> -}
> -
> -void dma_deconfigure(struct device *dev)
> -{
> - of_dma_deconfigure(dev);
> - acpi_dma_deconfigure(dev);
> -}
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index f1bf7b3..adf94eb 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -1130,6 +1130,28 @@ int platform_pm_restore(struct device *dev)
>
> #endif /* CONFIG_HIBERNATE_CALLBACKS */
>
> +int platform_dma_configure(struct device *dev)
> +{
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + return ret;
> +}
> +
> +void platform_dma_deconfigure(struct device *dev)
> +{
> + of_dma_deconfigure(dev);
> + acpi_dma_deconfigure(dev);
> +}
> +
> static const struct dev_pm_ops platform_dev_pm_ops = {
> .runtime_suspend = pm_generic_runtime_suspend,
> .runtime_resume = pm_generic_runtime_resume,
> @@ -1137,12 +1159,14 @@ int platform_pm_restore(struct device *dev)
> };
>
> struct bus_type platform_bus_type = {
> - .name = "platform",
> - .dev_groups = platform_dev_groups,
> - .match = platform_match,
> - .uevent = platform_uevent,
> - .pm = &platform_dev_pm_ops,
> - .force_dma = true,
> + .name = "platform",
> + .dev_groups = platform_dev_groups,
> + .match = platform_match,
> + .uevent = platform_uevent,
> + .pm = &platform_dev_pm_ops,
> + .dma_configure = platform_dma_configure,
> + .dma_deconfigure = platform_dma_deconfigure,
> + .force_dma = true,
> };
> EXPORT_SYMBOL_GPL(platform_bus_type);
>
> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> index 3bed6be..4a77814 100644
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -18,6 +18,8 @@
> #include <linux/pm_runtime.h>
> #include <linux/suspend.h>
> #include <linux/kexec.h>
> +#include <linux/acpi.h>
> +#include <linux/of_device.h>
> #include "pci.h"
>
> struct pci_dynid {
> @@ -1522,19 +1524,52 @@ static int pci_bus_num_vf(struct device *dev)
> return pci_num_vf(to_pci_dev(dev));
> }
>
> +int pci_dma_configure(struct device *dev)
> +{
> + struct device *bridge, *dma_dev;

You don't need dma_dev here; see the code removed in 09515ef5ddad for
how the logic originally worked.

> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> + dma_dev = bridge;
> + if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> + dma_dev->parent->of_node)
> + dma_dev = dma_dev->parent;
> +
> + if (dma_dev->of_node) {
> + ret = of_dma_configure(dev, dma_dev->of_node);
> + } else if (has_acpi_companion(dma_dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + pci_put_host_bridge_device(bridge);
> +
> + return ret;
> +}
> +
> +void pci_dma_deconfigure(struct device *dev)
> +{
> + of_dma_deconfigure(dev);
> + acpi_dma_deconfigure(dev);
> +}
> +
> struct bus_type pci_bus_type = {
> - .name = "pci",
> - .match = pci_bus_match,
> - .uevent = pci_uevent,
> - .probe = pci_device_probe,
> - .remove = pci_device_remove,
> - .shutdown = pci_device_shutdown,
> - .dev_groups = pci_dev_groups,
> - .bus_groups = pci_bus_groups,
> - .drv_groups = pci_drv_groups,
> - .pm = PCI_PM_OPS_PTR,
> - .num_vf = pci_bus_num_vf,
> - .force_dma = true,
> + .name = "pci",
> + .match = pci_bus_match,
> + .uevent = pci_uevent,
> + .probe = pci_device_probe,
> + .remove = pci_device_remove,
> + .shutdown = pci_device_shutdown,
> + .dev_groups = pci_dev_groups,
> + .bus_groups = pci_bus_groups,
> + .drv_groups = pci_drv_groups,
> + .pm = PCI_PM_OPS_PTR,
> + .num_vf = pci_bus_num_vf,
> + .dma_configure = pci_dma_configure,
> + .dma_deconfigure = pci_dma_deconfigure,
> + .force_dma = true,
> };
> EXPORT_SYMBOL(pci_bus_type);
>
> diff --git a/include/linux/device.h b/include/linux/device.h
> index b093405..9b2dcf6 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -88,6 +88,9 @@ extern int __must_check bus_create_file(struct bus_type *,
> * @resume: Called to bring a device on this bus out of sleep mode.
> * @num_vf: Called to find out how many virtual functions a device on this
> * bus supports.
> + * @dma_configure: Called to setup DMA configuration on a device on
> + this bus.
> + * @dma_deconfigure: Called to tear down the DMA configuration.
> * @pm: Power management operations of this bus, callback the specific
> * device driver's pm-ops.
> * @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
> @@ -130,6 +133,9 @@ struct bus_type {
>
> int (*num_vf)(struct device *dev);
>
> + int (*dma_configure)(struct device *dev);
> + void (*dma_deconfigure)(struct device *dev);

Seeing it laid out in the patch, I really don't think we need a
deconfigure callback like this - the fact that we're just copy-pasting
the existing implementation everywhere is a big hint, but more
conceptually I can't see a good reason for it to ever need bus-specific
behaviour in the same way that configure does.

Maybe that means we keep dma_configure() around for the sake of
symmetry, but just reduce it to:

int dma_configure(struct device *dev)
{
if (dev->bus->dma_configure)
return dev->bus->dma_configure(dev);
return 0;
}

Realistically though, dma_deconfigure() only exists for the sake of
calling arch_teardown_dma_ops(), and that only really exists for the
sake of the old ARM IOMMU code, so I'm not inclined to pretend it's
anywhere near as important as the dma_configure() path in design terms.

Robin.

> +
> const struct dev_pm_ops *pm;
>
> const struct iommu_ops *iommu_ops;
> diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> index eb9eab4..039224b 100644
> --- a/include/linux/dma-mapping.h
> +++ b/include/linux/dma-mapping.h
> @@ -761,18 +761,6 @@ void *dma_mark_declared_memory_occupied(struct device *dev,
> }
> #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
>
> -#ifdef CONFIG_HAS_DMA
> -int dma_configure(struct device *dev);
> -void dma_deconfigure(struct device *dev);
> -#else
> -static inline int dma_configure(struct device *dev)
> -{
> - return 0;
> -}
> -
> -static inline void dma_deconfigure(struct device *dev) {}
> -#endif
> -
> /*
> * Managed DMA API
> */
>

2018-03-13 16:02:24

[permalink] [raw]

Subject: RE: [PATCH] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Christoph Hellwig [mailto:[email protected]]
> Sent: Tuesday, March 13, 2018 13:05
> > +int amba_dma_configure(struct device *dev)
> > +{
> > + enum dev_dma_attr attr;
> > + int ret = 0;
> > +
> > + if (dev->of_node) {
> > + ret = of_dma_configure(dev, dev->of_node);
> > + } else if (has_acpi_companion(dev)) {
> > + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> > + if (attr != DEV_DMA_NOT_SUPPORTED)
> > + ret = acpi_dma_configure(dev, attr);
> > + }
> > +
> > + return ret;
>
> This code sniplet is duplicated so many times that I think we should
> just have some sort of dma_common_configure() for it that the various
> busses can use.

Agree. There is no good point in duplicating the code.
So this new API will be part of 'drivers/base/dma-mapping.c' file?

>
> > +void amba_dma_deconfigure(struct device *dev)
> > +{
> > + of_dma_deconfigure(dev);
> > + acpi_dma_deconfigure(dev);
> > +}
>
> As mention in my previous reply I think we don't even need a deconfigure
> callback at this point - just remove the ACPI and OF wrappers and
> clear the dma ops.
>
> Also in this series we should replace the force_dma flag by use of the
> proper method, e.g. give a force parameter to of_dma_configure and the
> new dma_common_configure helper that the busses that want it can set.

I am more inclined to what Robin states in other mail to keep symmetry.
i.e. to keep dma_configure() and dma_deconfigure() and call
dev->bus->dma_configure from dma_configure(). Is this okay?

Thanks,
Nipun

2018-03-13 16:14:02

[permalink] [raw]

Subject: RE: [PATCH] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Robin Murphy [mailto:[email protected]]
> Sent: Tuesday, March 13, 2018 17:06
>
> On 12/03/18 15:24, Nipun Gupta wrote:
> > The change introduces 'dma_configure' & 'dma_deconfigure'as
> > bus callback functions so each bus can choose to implement
> > its own dma configuration function.
> > This eases the addition of new busses w.r.t. adding the dma
> > configuration functionality.
>
> It's probably worth clarifying - either in the commit message, the
> kerneldoc, or both - that the bus-specific aspect is that of mapping
> between a given device on the bus and the relevant firmware description
> of its DMA configuration.

Okay.

>
> > The change also updates the PCI, Platform and ACPI bus to use
> > new introduced callbacks.
> >
> > Signed-off-by: Nipun Gupta <[email protected]>
> > ---
> > - This patch is based on the comments on:
> >
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpatchwo
> rk.kernel.org%2Fpatch%2F10259087%2F&data=02%7C01%7Cnipun.gupta%40nxp.com%7
> Cc541100ecb944e7650a408d588d69ab0%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7
> C0%7C636565377665676631&sdata=k2Xjn5B1GECx4UjCg9tChOpOrD3NPM7BkzIXLLSv3rI%
> 3D&reserved=0
> > - I have validated for PCI and platform, but not for AMBA as I
> > do not have infrastructure to validate it.
> > Can anyone please validate them on AMBA?
> >
> > drivers/amba/bus.c | 38 ++++++++++++++++++++++++-----
> > drivers/base/dd.c | 14 +++++++----
> > drivers/base/dma-mapping.c | 41 -------------------------------
> > drivers/base/platform.c | 36 ++++++++++++++++++++++-----
> > drivers/pci/pci-driver.c | 59 ++++++++++++++++++++++++++++++++++++-
> --------
> > include/linux/device.h | 6 +++++
> > include/linux/dma-mapping.h | 12 ---------
> > 7 files changed, 124 insertions(+), 82 deletions(-)
> >
> > diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
> > index 594c228..58241d2 100644
> > --- a/drivers/amba/bus.c
> > +++ b/drivers/amba/bus.c
> > @@ -20,6 +20,8 @@
> > #include <linux/sizes.h>
> > #include <linux/limits.h>
> > #include <linux/clk/clk-conf.h>
> > +#include <linux/acpi.h>
> > +#include <linux/of_device.h>
> >
> > #include <asm/irq.h>
> >
> > @@ -171,6 +173,28 @@ static int amba_pm_runtime_resume(struct device
> *dev)
> > }
> > #endif /* CONFIG_PM */
> >
> > +int amba_dma_configure(struct device *dev)
> > +{
> > + enum dev_dma_attr attr;
> > + int ret = 0;
> > +
> > + if (dev->of_node) {
> > + ret = of_dma_configure(dev, dev->of_node);
> > + } else if (has_acpi_companion(dev)) {
> > + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> > + if (attr != DEV_DMA_NOT_SUPPORTED)
> > + ret = acpi_dma_configure(dev, attr);
> > + }
> > +
> > + return ret;
> > +}
>
> I would be inclined to have amba_bustype just reference
> platform_dma_configure() directly rather than duplicate it like this,
> since there's no sensible reason for them to ever differ.

I think dma_common_configure() having this as the common code seems pretty
Decent. All the busses will probably call this API.

>
> > +
> > +void amba_dma_deconfigure(struct device *dev)
> > +{
> > + of_dma_deconfigure(dev);
> > + acpi_dma_deconfigure(dev);
> > +}
> > +
> > static const struct dev_pm_ops amba_pm = {
> > .suspend = pm_generic_suspend,
> > .resume = pm_generic_resume,
> > @@ -190,12 +214,14 @@ static int amba_pm_runtime_resume(struct device
> *dev)
> > * so we call the bus "amba".
> > */
> > struct bus_type amba_bustype = {
> > - .name = "amba",
> > - .dev_groups = amba_dev_groups,
> > - .match = amba_match,
> > - .uevent = amba_uevent,
> > - .pm = &amba_pm,
> > - .force_dma = true,
> > + .name = "amba",
> > + .dev_groups = amba_dev_groups,
> > + .match = amba_match,
> > + .uevent = amba_uevent,
> > + .pm = &amba_pm,
> > + .dma_configure = amba_dma_configure,
> > + .dma_deconfigure = amba_dma_deconfigure,
> > + .force_dma = true,
>
> This patch should also be removing force_dma because it no longer makes
> sense. If DMA configuration is now done by a bus-level callback, then a
> bus which wants its children to get DMA configuration needs to implement
> that callback; there's nowhere to force a "default" global behaviour any
> more.

Agree. We will also need to pass a force_dma flag in of_dma_configure() as
Christoph suggests. Ill update this.

>
> > };
> >
> > static int __init amba_init(void)
> > diff --git a/drivers/base/dd.c b/drivers/base/dd.c
> > index de6fd09..f124f3f 100644
> > --- a/drivers/base/dd.c
> > +++ b/drivers/base/dd.c
> > @@ -421,9 +421,11 @@ static int really_probe(struct device *dev, struct
> device_driver *drv)
> > if (ret)
> > goto pinctrl_bind_failed;
> >
> > - ret = dma_configure(dev);
> > - if (ret)
> > - goto dma_failed;
> > + if (dev->bus->dma_configure) {
> > + ret = dev->bus->dma_configure(dev);
> > + if (ret)
> > + goto dma_failed;
> > + }
> >
> > if (driver_sysfs_add(dev)) {
> > printk(KERN_ERR "%s: driver_sysfs_add(%s) failed\n",
> > @@ -486,7 +488,8 @@ static int really_probe(struct device *dev, struct
> device_driver *drv)
> > goto done;
> >
> > probe_failed:
> > - dma_deconfigure(dev);
> > + if (dev->bus->dma_deconfigure)
> > + dev->bus->dma_deconfigure(dev);
> > dma_failed:
> > if (dev->bus)
> > blocking_notifier_call_chain(&dev->bus->p->bus_notifier,
> > @@ -895,7 +898,8 @@ static void __device_release_driver(struct device
> *dev, struct device *parent)
> > drv->remove(dev);
> >
> > device_links_driver_cleanup(dev);
> > - dma_deconfigure(dev);
> > + if (dev->bus->dma_deconfigure)
> > + dev->bus->dma_deconfigure(dev);
> >
> > devres_release_all(dev);
> > dev->driver = NULL;
> > diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
> > index 3b11835..f16bd49 100644
> > --- a/drivers/base/dma-mapping.c
> > +++ b/drivers/base/dma-mapping.c
> > @@ -6,11 +6,9 @@
> > * Copyright (c) 2006 Tejun Heo <[email protected]>
> > */
> >
> > -#include <linux/acpi.h>
> > #include <linux/dma-mapping.h>
> > #include <linux/export.h>
> > #include <linux/gfp.h>
> > -#include <linux/of_device.h>
> > #include <linux/slab.h>
> > #include <linux/vmalloc.h>
> >
> > @@ -329,42 +327,3 @@ void dma_common_free_remap(void *cpu_addr, size_t
> size, unsigned long vm_flags)
> > vunmap(cpu_addr);
> > }
> > #endif
> > -
> > -/*
> > - * Common configuration to enable DMA API use for a device
> > - */
> > -#include <linux/pci.h>
> > -
> > -int dma_configure(struct device *dev)
> > -{
> > - struct device *bridge = NULL, *dma_dev = dev;
> > - enum dev_dma_attr attr;
> > - int ret = 0;
> > -
> > - if (dev_is_pci(dev)) {
> > - bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> > - dma_dev = bridge;
> > - if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> > - dma_dev->parent->of_node)
> > - dma_dev = dma_dev->parent;
> > - }
> > -
> > - if (dma_dev->of_node) {
> > - ret = of_dma_configure(dev, dma_dev->of_node);
> > - } else if (has_acpi_companion(dma_dev)) {
> > - attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> > - if (attr != DEV_DMA_NOT_SUPPORTED)
> > - ret = acpi_dma_configure(dev, attr);
> > - }
> > -
> > - if (bridge)
> > - pci_put_host_bridge_device(bridge);
> > -
> > - return ret;
> > -}
> > -
> > -void dma_deconfigure(struct device *dev)
> > -{
> > - of_dma_deconfigure(dev);
> > - acpi_dma_deconfigure(dev);
> > -}
> > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > index f1bf7b3..adf94eb 100644
> > --- a/drivers/base/platform.c
> > +++ b/drivers/base/platform.c
> > @@ -1130,6 +1130,28 @@ int platform_pm_restore(struct device *dev)
> >
> > #endif /* CONFIG_HIBERNATE_CALLBACKS */
> >
> > +int platform_dma_configure(struct device *dev)
> > +{
> > + enum dev_dma_attr attr;
> > + int ret = 0;
> > +
> > + if (dev->of_node) {
> > + ret = of_dma_configure(dev, dev->of_node);
> > + } else if (has_acpi_companion(dev)) {
> > + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> > + if (attr != DEV_DMA_NOT_SUPPORTED)
> > + ret = acpi_dma_configure(dev, attr);
> > + }
> > +
> > + return ret;
> > +}
> > +
> > +void platform_dma_deconfigure(struct device *dev)
> > +{
> > + of_dma_deconfigure(dev);
> > + acpi_dma_deconfigure(dev);
> > +}
> > +
> > static const struct dev_pm_ops platform_dev_pm_ops = {
> > .runtime_suspend = pm_generic_runtime_suspend,
> > .runtime_resume = pm_generic_runtime_resume,
> > @@ -1137,12 +1159,14 @@ int platform_pm_restore(struct device *dev)
> > };
> >
> > struct bus_type platform_bus_type = {
> > - .name = "platform",
> > - .dev_groups = platform_dev_groups,
> > - .match = platform_match,
> > - .uevent = platform_uevent,
> > - .pm = &platform_dev_pm_ops,
> > - .force_dma = true,
> > + .name = "platform",
> > + .dev_groups = platform_dev_groups,
> > + .match = platform_match,
> > + .uevent = platform_uevent,
> > + .pm = &platform_dev_pm_ops,
> > + .dma_configure = platform_dma_configure,
> > + .dma_deconfigure = platform_dma_deconfigure,
> > + .force_dma = true,
> > };
> > EXPORT_SYMBOL_GPL(platform_bus_type);
> >
> > diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> > index 3bed6be..4a77814 100644
> > --- a/drivers/pci/pci-driver.c
> > +++ b/drivers/pci/pci-driver.c
> > @@ -18,6 +18,8 @@
> > #include <linux/pm_runtime.h>
> > #include <linux/suspend.h>
> > #include <linux/kexec.h>
> > +#include <linux/acpi.h>
> > +#include <linux/of_device.h>
> > #include "pci.h"
> >
> > struct pci_dynid {
> > @@ -1522,19 +1524,52 @@ static int pci_bus_num_vf(struct device *dev)
> > return pci_num_vf(to_pci_dev(dev));
> > }
> >
> > +int pci_dma_configure(struct device *dev)
> > +{
> > + struct device *bridge, *dma_dev;
>
> You don't need dma_dev here; see the code removed in 09515ef5ddad for
> how the logic originally worked.

Okay. I will have a look in this commit id.

>
> > + enum dev_dma_attr attr;
> > + int ret = 0;
> > +
> > + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> > + dma_dev = bridge;
> > + if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> > + dma_dev->parent->of_node)
> > + dma_dev = dma_dev->parent;
> > +
> > + if (dma_dev->of_node) {
> > + ret = of_dma_configure(dev, dma_dev->of_node);
> > + } else if (has_acpi_companion(dma_dev)) {
> > + attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> > + if (attr != DEV_DMA_NOT_SUPPORTED)
> > + ret = acpi_dma_configure(dev, attr);
> > + }
> > +
> > + pci_put_host_bridge_device(bridge);
> > +
> > + return ret;
> > +}
> > +
> > +void pci_dma_deconfigure(struct device *dev)
> > +{
> > + of_dma_deconfigure(dev);
> > + acpi_dma_deconfigure(dev);
> > +}
> > +
> > struct bus_type pci_bus_type = {
> > - .name = "pci",
> > - .match = pci_bus_match,
> > - .uevent = pci_uevent,
> > - .probe = pci_device_probe,
> > - .remove = pci_device_remove,
> > - .shutdown = pci_device_shutdown,
> > - .dev_groups = pci_dev_groups,
> > - .bus_groups = pci_bus_groups,
> > - .drv_groups = pci_drv_groups,
> > - .pm = PCI_PM_OPS_PTR,
> > - .num_vf = pci_bus_num_vf,
> > - .force_dma = true,
> > + .name = "pci",
> > + .match = pci_bus_match,
> > + .uevent = pci_uevent,
> > + .probe = pci_device_probe,
> > + .remove = pci_device_remove,
> > + .shutdown = pci_device_shutdown,
> > + .dev_groups = pci_dev_groups,
> > + .bus_groups = pci_bus_groups,
> > + .drv_groups = pci_drv_groups,
> > + .pm = PCI_PM_OPS_PTR,
> > + .num_vf = pci_bus_num_vf,
> > + .dma_configure = pci_dma_configure,
> > + .dma_deconfigure = pci_dma_deconfigure,
> > + .force_dma = true,
> > };
> > EXPORT_SYMBOL(pci_bus_type);
> >
> > diff --git a/include/linux/device.h b/include/linux/device.h
> > index b093405..9b2dcf6 100644
> > --- a/include/linux/device.h
> > +++ b/include/linux/device.h
> > @@ -88,6 +88,9 @@ extern int __must_check bus_create_file(struct
> bus_type *,
> > * @resume: Called to bring a device on this bus out of sleep mode.
> > * @num_vf: Called to find out how many virtual functions a device on
> this
> > * bus supports.
> > + * @dma_configure: Called to setup DMA configuration on a device on
> > + this bus.
> > + * @dma_deconfigure: Called to tear down the DMA configuration.
> > * @pm: Power management operations of this bus, callback
> the specific
> > * device driver's pm-ops.
> > * @iommu_ops: IOMMU specific operations for this bus, used to attach
> IOMMU
> > @@ -130,6 +133,9 @@ struct bus_type {
> >
> > int (*num_vf)(struct device *dev);
> >
> > + int (*dma_configure)(struct device *dev);
> > + void (*dma_deconfigure)(struct device *dev);
>
> Seeing it laid out in the patch, I really don't think we need a
> deconfigure callback like this - the fact that we're just copy-pasting
> the existing implementation everywhere is a big hint, but more
> conceptually I can't see a good reason for it to ever need bus-specific
> behaviour in the same way that configure does.
>
> Maybe that means we keep dma_configure() around for the sake of
> symmetry, but just reduce it to:
>
> int dma_configure(struct device *dev)
> {
> if (dev->bus->dma_configure)
> return dev->bus->dma_configure(dev);
> return 0;
> }
>
> Realistically though, dma_deconfigure() only exists for the sake of
> calling arch_teardown_dma_ops(), and that only really exists for the
> sake of the old ARM IOMMU code, so I'm not inclined to pretend it's
> anywhere near as important as the dma_configure() path in design terms.

Yes, I will remove dma_deconfigure callback.

Thanks,
Nipun

>
> Robin.
>
> > +
> > const struct dev_pm_ops *pm;
> >
> > const struct iommu_ops *iommu_ops;
> > diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> > index eb9eab4..039224b 100644
> > --- a/include/linux/dma-mapping.h
> > +++ b/include/linux/dma-mapping.h
> > @@ -761,18 +761,6 @@ void *dma_mark_declared_memory_occupied(struct
> device *dev,
> > }
> > #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
> >
> > -#ifdef CONFIG_HAS_DMA
> > -int dma_configure(struct device *dev);
> > -void dma_deconfigure(struct device *dev);
> > -#else
> > -static inline int dma_configure(struct device *dev)
> > -{
> > - return 0;
> > -}
> > -
> > -static inline void dma_deconfigure(struct device *dev) {}
> > -#endif
> > -
> > /*
> > * Managed DMA API
> > */
> >

2018-03-13 21:56:21

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

Hi Nipun,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on linus/master]
[also build test WARNING on v4.16-rc5 next-20180313]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Nipun-Gupta/dma-mapping-move-dma-configuration-to-bus-infrastructure/20180313-225250
reproduce:
# apt-get install sparse
make ARCH=x86_64 allmodconfig
make C=1 CF=-D__CHECK_ENDIAN__

sparse warnings: (new ones prefixed by >>)

>> drivers/base/platform.c:1133:5: sparse: symbol 'platform_dma_configure' was not declared. Should it be static?
>> drivers/base/platform.c:1149:6: sparse: symbol 'platform_dma_deconfigure' was not declared. Should it be static?

Please review and possibly fold the followup patch.

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

2018-03-13 21:57:09

[permalink] [raw]

Subject: [RFC PATCH] dma-mapping: platform_dma_configure() can be static

Fixes: 9a019f425175 ("dma-mapping: move dma configuration to bus infrastructure")
Signed-off-by: Fengguang Wu <[email protected]>
---
platform.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index adf94eb..dc9c790 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1130,7 +1130,7 @@ int platform_pm_restore(struct device *dev)

#endif /* CONFIG_HIBERNATE_CALLBACKS */

-int platform_dma_configure(struct device *dev)
+static int platform_dma_configure(struct device *dev)
{
enum dev_dma_attr attr;
int ret = 0;
@@ -1146,7 +1146,7 @@ int platform_dma_configure(struct device *dev)
return ret;
}

-void platform_dma_deconfigure(struct device *dev)
+static void platform_dma_deconfigure(struct device *dev)
{
of_dma_deconfigure(dev);
acpi_dma_deconfigure(dev);

2018-03-14 05:51:05

by Fengguang Wu

[permalink] [raw]

Subject: [dma] 9a019f4251: BUG:unable_to_handle_kernel

FYI, we noticed the following commit (built with gcc-7):

commit: 9a019f425175e1c42b68b5b628492ed07e6b1e22 ("dma-mapping: move dma configuration to bus infrastructure")
url: https://github.com/0day-ci/linux/commits/Nipun-Gupta/dma-mapping-move-dma-configuration-to-bus-infrastructure/20180313-225250

in testcase: trinity
with following parameters:

runtime: 300s

test-description: Trinity is a linux system call fuzz tester.
test-url: http://codemonkey.org.uk/projects/trinity/

on test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -m 512M

caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):

+------------------------------------------+------------+------------+
| | 3266b5bd97 | 9a019f4251 |
+------------------------------------------+------------+------------+
| boot_successes | 8 | 0 |
| boot_failures | 0 | 8 |
| BUG:unable_to_handle_kernel | 0 | 8 |
| Oops:#[##] | 0 | 8 |
| RIP:device_release_driver_internal | 0 | 8 |
| Kernel_panic-not_syncing:Fatal_exception | 0 | 8 |
+------------------------------------------+------------+------------+

[ 69.592314] BUG: unable to handle kernel NULL pointer dereference at 0000000000000088
[ 69.593011] IP: device_release_driver_internal+0x157/0x230
[ 69.593011] PGD 0 P4D 0
[ 69.593011] Oops: 0000 [#1] SMP
[ 69.593011] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.16.0-rc4-00340-g9a019f4 #1
[ 69.593011] RIP: 0010:device_release_driver_internal+0x157/0x230
[ 69.593011] RSP: 0000:ffff89575e43fd70 EFLAGS: 00010292
[ 69.593011] RAX: 0000000000000000 RBX: ffff89575c6f58a8 RCX: 0000000000000000
[ 69.593011] RDX: 0000000000000000 RSI: 0000000000000002 RDI: 0000000000000246
[ 69.593011] RBP: ffff89575e43fd98 R08: 0000000000000001 R09: 0000000000000000
[ 69.593011] R10: ffff89575e43fd48 R11: 0000000000000000 R12: 0000000000000000
[ 69.593011] R13: ffff89575c6f5908 R14: ffffffffbd87fba8 R15: 0000000000000060
[ 69.593011] FS: 0000000000000000(0000) GS:ffff89575f600000(0000) knlGS:0000000000000000
[ 69.593011] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 69.593011] CR2: 0000000000000088 CR3: 0000000017a1d000 CR4: 00000000000406b0
[ 69.593011] Call Trace:
[ 69.593011] device_release_driver+0xd/0x10
[ 69.593011] mac80211_hwsim_new_radio+0x390/0x1000
[ 69.593011] ? hwsim_init_net+0x4e/0x4e
[ 69.593011] init_mac80211_hwsim+0x26a/0x671
[ 69.593011] ? hwsim_init_net+0x4e/0x4e
[ 69.593011] do_one_initcall+0xe1/0x256
[ 69.593011] ? set_debug_rodata+0x3c/0x3c
[ 69.593011] kernel_init_freeable+0x175/0x283
[ 69.593011] ? rest_init+0xc0/0xc0
[ 69.593011] kernel_init+0x9/0x100
[ 69.593011] ret_from_fork+0x3a/0x50
[ 69.593011] Code: c0 0f 84 e1 00 00 00 48 8b 40 48 48 85 c0 0f 84 d4 00 00 00 48 89 df e8 98 1c 20 01 48 89 df e8 d0 d0 ff ff 48 8b 83 e0 00 00 00 <48> 8b 80 88 00 00 00 48 85 c0 74 08 48 89 df e8 75 1c 20 01 48
[ 69.593011] RIP: device_release_driver_internal+0x157/0x230 RSP: ffff89575e43fd70
[ 69.593011] CR2: 0000000000000088
[ 69.593011] ---[ end trace 60349fca5b9c7ff9 ]---

To reproduce:

git clone https://github.com/intel/lkp-tests.git
cd lkp-tests
bin/lkp qemu -k <bzImage> job-script # job-script is attached in this email

Thanks,
lkp

Attachments:

(No filename) (3.52 kB)
config-4.16.0-rc4-00340-g9a019f4 (131.96 kB)
job-script (3.84 kB)
dmesg.xz (12.03 kB)
Download all attachments

2018-03-14 09:03:39

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

>> + .dev_groups = amba_dev_groups,
>> + .match = amba_match,
>> + .uevent = amba_uevent,
>> + .pm = &amba_pm,
>> + .dma_configure = amba_dma_configure,
>> + .dma_deconfigure = amba_dma_deconfigure,
>> + .force_dma = true,
>
> This patch should also be removing force_dma because it no longer makes
> sense. If DMA configuration is now done by a bus-level callback, then a bus
> which wants its children to get DMA configuration needs to implement that
> callback; there's nowhere to force a "default" global behaviour any more.

Btw, we don't really know how many busses currently rely on OF or ACPI
configuration. So maybe we need to keep those as a default?

2018-03-14 09:04:52

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

> Agree. There is no good point in duplicating the code.
> So this new API will be part of 'drivers/base/dma-mapping.c' file?

Yes.

> > As mention in my previous reply I think we don't even need a deconfigure
> > callback at this point - just remove the ACPI and OF wrappers and
> > clear the dma ops.
> >
> > Also in this series we should replace the force_dma flag by use of the
> > proper method, e.g. give a force parameter to of_dma_configure and the
> > new dma_common_configure helper that the busses that want it can set.
>
> I am more inclined to what Robin states in other mail to keep symmetry.
> i.e. to keep dma_configure() and dma_deconfigure() and call
> dev->bus->dma_configure from dma_configure(). Is this okay?

Sure.

2018-03-21 06:57:41

[permalink] [raw]

Subject: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

It's bus specific aspect to map a given device on the bus and
relevant firmware description of its DMA configuration.
So, this change introduces '/dma_configure/' as bus callback
giving flexibility to busses for implementing its own dma
configuration function.

The change eases the addition of new busses w.r.t. adding the dma
configuration functionality.

This patch also updates the PCI, Platform, ACPI and host1x bus to
use new introduced callbacks.

Suggested-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nipun Gupta <[email protected]>
---
- The patches are based on the comments on:
https://patchwork.kernel.org/patch/10259087/

Changes in v2:
- Do not have dma_deconfigure callback
- Have '/dma_common_configure/' API to provide a common DMA
configuration which can be used by busses if it suits them.
- Platform and ACPI bus to use '/dma_common_configure/' in
'/dma_configure/' callback.
- Updated commit message
- Updated pci_dma_configure API with changes suggested by Robin

drivers/amba/bus.c | 7 +++++++
drivers/base/dma-mapping.c | 35 +++++++++++++++--------------------
drivers/base/platform.c | 6 ++++++
drivers/gpu/host1x/bus.c | 9 +++++++++
drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
include/linux/device.h | 4 ++++
include/linux/dma-mapping.h | 1 +
7 files changed, 74 insertions(+), 20 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 594c228..2fa1e8b 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -20,6 +20,7 @@
#include <linux/sizes.h>
#include <linux/limits.h>
#include <linux/clk/clk-conf.h>
+#include <linux/dma-mapping.h>

#include <asm/irq.h>

@@ -171,6 +172,11 @@ static int amba_pm_runtime_resume(struct device *dev)
}
#endif /* CONFIG_PM */

+static int amba_dma_configure(struct device *dev)
+{
+ return dma_common_configure(dev);
+}
+
static const struct dev_pm_ops amba_pm = {
.suspend = pm_generic_suspend,
.resume = pm_generic_resume,
@@ -194,6 +200,7 @@ struct bus_type amba_bustype = {
.dev_groups = amba_dev_groups,
.match = amba_match,
.uevent = amba_uevent,
+ .dma_configure = amba_dma_configure,
.pm = &amba_pm,
.force_dma = true,
};
diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
index 3b11835..48f9af0 100644
--- a/drivers/base/dma-mapping.c
+++ b/drivers/base/dma-mapping.c
@@ -331,38 +331,33 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
#endif

/*
- * Common configuration to enable DMA API use for a device
+ * Common configuration to enable DMA API use for a device.
+ * A bus can use this function in its 'dma_configure' callback, if
+ * suitable for the bus.
*/
-#include <linux/pci.h>
-
-int dma_configure(struct device *dev)
+int dma_common_configure(struct device *dev)
{
- struct device *bridge = NULL, *dma_dev = dev;
enum dev_dma_attr attr;
int ret = 0;

- if (dev_is_pci(dev)) {
- bridge = pci_get_host_bridge_device(to_pci_dev(dev));
- dma_dev = bridge;
- if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
- dma_dev->parent->of_node)
- dma_dev = dma_dev->parent;
- }
-
- if (dma_dev->of_node) {
- ret = of_dma_configure(dev, dma_dev->of_node);
- } else if (has_acpi_companion(dma_dev)) {
- attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
+ if (dev->of_node) {
+ ret = of_dma_configure(dev, dev->of_node);
+ } else if (has_acpi_companion(dev)) {
+ attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
if (attr != DEV_DMA_NOT_SUPPORTED)
ret = acpi_dma_configure(dev, attr);
}

- if (bridge)
- pci_put_host_bridge_device(bridge);
-
return ret;
}

+int dma_configure(struct device *dev)
+{
+ if (dev->bus->dma_configure)
+ return dev->bus->dma_configure(dev);
+
+ return 0;
+}
void dma_deconfigure(struct device *dev)
{
of_dma_deconfigure(dev);
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index f1bf7b3..d2d5891 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1130,6 +1130,11 @@ int platform_pm_restore(struct device *dev)

#endif /* CONFIG_HIBERNATE_CALLBACKS */

+static int platform_dma_configure(struct device *dev)
+{
+ return dma_common_configure(dev);
+}
+
static const struct dev_pm_ops platform_dev_pm_ops = {
.runtime_suspend = pm_generic_runtime_suspend,
.runtime_resume = pm_generic_runtime_resume,
@@ -1141,6 +1146,7 @@ struct bus_type platform_bus_type = {
.dev_groups = platform_dev_groups,
.match = platform_match,
.uevent = platform_uevent,
+ .dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
.force_dma = true,
};
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index 88a3558..fa9896d 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -314,6 +314,14 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
return strcmp(dev_name(dev), drv->name) == 0;
}

+static int host1x_dma_configure(struct device *dev)
+{
+ if (dev->of_node)
+ return of_dma_configure(dev, dev->of_node);
+
+ return 0;
+}
+
static const struct dev_pm_ops host1x_device_pm_ops = {
.suspend = pm_generic_suspend,
.resume = pm_generic_resume,
@@ -326,6 +334,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
struct bus_type host1x_bus_type = {
.name = "host1x",
.match = host1x_device_match,
+ .dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
.force_dma = true,
};
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index 3bed6be..b473a4c 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -18,6 +18,8 @@
#include <linux/pm_runtime.h>
#include <linux/suspend.h>
#include <linux/kexec.h>
+#include <linux/of_device.h>
+#include <linux/acpi.h>
#include "pci.h"

struct pci_dynid {
@@ -1522,6 +1524,35 @@ static int pci_bus_num_vf(struct device *dev)
return pci_num_vf(to_pci_dev(dev));
}

+/**
+ * pci_dma_configure - Setup DMA configuration
+ * @dev: ptr to dev structure
+ *
+ * Function to update PCI devices's DMA configuration using the same
+ * info from the OF node or ACPI node of host bridge's parent (if any).
+ */
+static int pci_dma_configure(struct device *dev)
+{
+ struct device *bridge;
+ enum dev_dma_attr attr;
+ int ret = 0;
+
+ bridge = pci_get_host_bridge_device(to_pci_dev(dev));
+
+ if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
+ bridge->parent->of_node) {
+ ret = of_dma_configure(dev, bridge->parent->of_node);
+ } else if (has_acpi_companion(bridge)) {
+ attr = acpi_get_dma_attr(to_acpi_device_node(bridge->fwnode));
+ if (attr != DEV_DMA_NOT_SUPPORTED)
+ ret = acpi_dma_configure(dev, attr);
+ }
+
+ pci_put_host_bridge_device(bridge);
+
+ return ret;
+}
+
struct bus_type pci_bus_type = {
.name = "pci",
.match = pci_bus_match,
@@ -1534,6 +1565,7 @@ struct bus_type pci_bus_type = {
.drv_groups = pci_drv_groups,
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
+ .dma_configure = pci_dma_configure,
.force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);
diff --git a/include/linux/device.h b/include/linux/device.h
index b093405..1832d90 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type *,
* @resume: Called to bring a device on this bus out of sleep mode.
* @num_vf: Called to find out how many virtual functions a device on this
* bus supports.
+ * @dma_configure: Called to setup DMA configuration on a device on
+ this bus.
* @pm: Power management operations of this bus, callback the specific
* device driver's pm-ops.
* @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
@@ -130,6 +132,8 @@ struct bus_type {

int (*num_vf)(struct device *dev);

+ int (*dma_configure)(struct device *dev);
+
const struct dev_pm_ops *pm;

const struct iommu_ops *iommu_ops;
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index eb9eab4..c15986b 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -761,6 +761,7 @@ void *dma_mark_declared_memory_occupied(struct device *dev,
}
#endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */

+int dma_common_configure(struct device *dev);
#ifdef CONFIG_HAS_DMA
int dma_configure(struct device *dev);
void dma_deconfigure(struct device *dev);
--
1.9.1

2018-03-21 06:58:22

[permalink] [raw]

Subject: [PATCH v2 2/2] drivers: remove force dma flag from buses

With each bus implementing its own DMA configuration callback,
there is no need for bus to explicitly have force_dma in its
global structure. This patch modifies of_dma_configure API to
accept an input parameter which specifies if implicit DMA
configuration is required even when it is not described by the
firmware.

Signed-off-by: Nipun Gupta <[email protected]>
---

Changes in v2:
- This is a new change suggested by Robin and Christoph
and is added to the series.

drivers/amba/bus.c | 3 +--
drivers/base/dma-mapping.c | 4 ++--
drivers/base/platform.c | 3 +--
drivers/bcma/main.c | 2 +-
drivers/dma/qcom/hidma_mgmt.c | 2 +-
drivers/gpu/host1x/bus.c | 5 ++---
drivers/of/device.c | 6 ++++--
drivers/of/of_reserved_mem.c | 2 +-
drivers/pci/pci-driver.c | 3 +--
include/linux/device.h | 4 ----
include/linux/dma-mapping.h | 2 +-
include/linux/of_device.h | 4 +++-
12 files changed, 18 insertions(+), 22 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 2fa1e8b..1d58348 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -174,7 +174,7 @@ static int amba_pm_runtime_resume(struct device *dev)

static int amba_dma_configure(struct device *dev)
{
- return dma_common_configure(dev);
+ return dma_common_configure(dev, true);
}

static const struct dev_pm_ops amba_pm = {
@@ -202,7 +202,6 @@ struct bus_type amba_bustype = {
.uevent = amba_uevent,
.dma_configure = amba_dma_configure,
.pm = &amba_pm,
- .force_dma = true,
};

static int __init amba_init(void)
diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
index 48f9af0..03f8584 100644
--- a/drivers/base/dma-mapping.c
+++ b/drivers/base/dma-mapping.c
@@ -335,13 +335,13 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
* A bus can use this function in its 'dma_configure' callback, if
* suitable for the bus.
*/
-int dma_common_configure(struct device *dev)
+int dma_common_configure(struct device *dev, bool force_dma)
{
enum dev_dma_attr attr;
int ret = 0;

if (dev->of_node) {
- ret = of_dma_configure(dev, dev->of_node);
+ ret = of_dma_configure(dev, dev->of_node, force_dma);
} else if (has_acpi_companion(dev)) {
attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
if (attr != DEV_DMA_NOT_SUPPORTED)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index d2d5891..154707c 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1132,7 +1132,7 @@ int platform_pm_restore(struct device *dev)

static int platform_dma_configure(struct device *dev)
{
- return dma_common_configure(dev);
+ return dma_common_configure(dev, true);
}

static const struct dev_pm_ops platform_dev_pm_ops = {
@@ -1148,7 +1148,6 @@ struct bus_type platform_bus_type = {
.uevent = platform_uevent,
.dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
- .force_dma = true,
};
EXPORT_SYMBOL_GPL(platform_bus_type);

diff --git a/drivers/bcma/main.c b/drivers/bcma/main.c
index e6986c7..fc1f4ac 100644
--- a/drivers/bcma/main.c
+++ b/drivers/bcma/main.c
@@ -207,7 +207,7 @@ static void bcma_of_fill_device(struct device *parent,

core->irq = bcma_of_get_irq(parent, core, 0);

- of_dma_configure(&core->dev, node);
+ of_dma_configure(&core->dev, node, false);
}

unsigned int bcma_core_irq(struct bcma_device *core, int num)
diff --git a/drivers/dma/qcom/hidma_mgmt.c b/drivers/dma/qcom/hidma_mgmt.c
index 000c7019..d64edeb 100644
--- a/drivers/dma/qcom/hidma_mgmt.c
+++ b/drivers/dma/qcom/hidma_mgmt.c
@@ -398,7 +398,7 @@ static int __init hidma_mgmt_of_populate_channels(struct device_node *np)
}
of_node_get(child);
new_pdev->dev.of_node = child;
- of_dma_configure(&new_pdev->dev, child);
+ of_dma_configure(&new_pdev->dev, child, true);
/*
* It is assumed that calling of_msi_configure is safe on
* platforms with or without MSI support.
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index fa9896d..211eb6b 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -317,7 +317,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
static int host1x_dma_configure(struct device *dev)
{
if (dev->of_node)
- return of_dma_configure(dev, dev->of_node);
+ return of_dma_configure(dev, dev->of_node, true);

return 0;
}
@@ -336,7 +336,6 @@ struct bus_type host1x_bus_type = {
.match = host1x_device_match,
.dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
- .force_dma = true,
};

static void __host1x_device_del(struct host1x_device *device)
@@ -425,7 +424,7 @@ static int host1x_device_add(struct host1x *host1x,
device->dev.bus = &host1x_bus_type;
device->dev.parent = host1x->dev;

- of_dma_configure(&device->dev, host1x->dev->of_node);
+ of_dma_configure(&device->dev, host1x->dev->of_node, true);

err = host1x_device_parse_dt(device, driver);
if (err < 0) {
diff --git a/drivers/of/device.c b/drivers/of/device.c
index 064c818..33d8551 100644
--- a/drivers/of/device.c
+++ b/drivers/of/device.c
@@ -76,6 +76,8 @@ int of_device_add(struct platform_device *ofdev)
* of_dma_configure - Setup DMA configuration
* @dev: Device to apply DMA configuration
* @np: Pointer to OF node having DMA configuration
+ * @force_dma: Whether device is to be set up by of_dma_configure() even if
+ * DMA capability is not explicitly described by firmware.
*
* Try to get devices's DMA configuration from DT and update it
* accordingly.
@@ -84,7 +86,7 @@ int of_device_add(struct platform_device *ofdev)
* can use a platform bus notifier and handle BUS_NOTIFY_ADD_DEVICE events
* to fix up DMA configuration.
*/
-int of_dma_configure(struct device *dev, struct device_node *np)
+int of_dma_configure(struct device *dev, struct device_node *np, bool force_dma)
{
u64 dma_addr, paddr, size = 0;
int ret;
@@ -100,7 +102,7 @@ int of_dma_configure(struct device *dev, struct device_node *np)
* DMA configuration regardless of whether "dma-ranges" is
* correctly specified or not.
*/
- if (!dev->bus->force_dma)
+ if (!force_dma)
return ret == -ENODEV ? 0 : ret;

dma_addr = offset = 0;
diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
index 9a4f4246..895c83e 100644
--- a/drivers/of/of_reserved_mem.c
+++ b/drivers/of/of_reserved_mem.c
@@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device *dev,
/* ensure that dma_ops is set for virtual devices
* using reserved memory
*/
- of_dma_configure(dev, np);
+ of_dma_configure(dev, np, true);

dev_info(dev, "assigned reserved memory node %s\n", rmem->name);
} else {
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index b473a4c..b15637a 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -1541,7 +1541,7 @@ static int pci_dma_configure(struct device *dev)

if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
bridge->parent->of_node) {
- ret = of_dma_configure(dev, bridge->parent->of_node);
+ ret = of_dma_configure(dev, bridge->parent->of_node, true);
} else if (has_acpi_companion(bridge)) {
attr = acpi_get_dma_attr(to_acpi_device_node(bridge->fwnode));
if (attr != DEV_DMA_NOT_SUPPORTED)
@@ -1566,7 +1566,6 @@ struct bus_type pci_bus_type = {
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
.dma_configure = pci_dma_configure,
- .force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);

diff --git a/include/linux/device.h b/include/linux/device.h
index 1832d90..92c530e 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -98,8 +98,6 @@ extern int __must_check bus_create_file(struct bus_type *,
* @p: The private data of the driver core, only the driver core can
* touch this.
* @lock_key: Lock class key for use by the lock validator
- * @force_dma: Assume devices on this bus should be set up by dma_configure()
- * even if DMA capability is not explicitly described by firmware.
*
* A bus is a channel between the processor and one or more devices. For the
* purposes of the device model, all devices are connected via a bus, even if
@@ -140,8 +138,6 @@ struct bus_type {

struct subsys_private *p;
struct lock_class_key lock_key;
-
- bool force_dma;
};

extern int __must_check bus_register(struct bus_type *bus);
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index c15986b..3dba3e6 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -761,7 +761,7 @@ void *dma_mark_declared_memory_occupied(struct device *dev,
}
#endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */

-int dma_common_configure(struct device *dev);
+int dma_common_configure(struct device *dev, bool force_dma);
#ifdef CONFIG_HAS_DMA
int dma_configure(struct device *dev);
void dma_deconfigure(struct device *dev);
diff --git a/include/linux/of_device.h b/include/linux/of_device.h
index 8da5a1b..b0be3a5 100644
--- a/include/linux/of_device.h
+++ b/include/linux/of_device.h
@@ -55,7 +55,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
return of_node_get(cpu_dev->of_node);
}

-int of_dma_configure(struct device *dev, struct device_node *np);
+int of_dma_configure(struct device *dev,
+ struct device_node *np,
+ bool force_dma);
void of_dma_deconfigure(struct device *dev);
#else /* CONFIG_OF */

--
1.9.1

2018-03-21 07:21:28

by Bharat Bhushan

[permalink] [raw]

Subject: RE: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Nipun Gupta
> Sent: Wednesday, March 21, 2018 12:25 PM
> To: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]; [email protected]; [email protected]; linux-
> [email protected]; [email protected]; linux-
> [email protected]; [email protected]; linux-
> [email protected]; [email protected]; dri-
> [email protected]; [email protected];
> [email protected]; [email protected]; Bharat Bhushan
> <[email protected]>; Leo Li <[email protected]>; Nipun Gupta
> <[email protected]>
> Subject: [PATCH v2 1/2] dma-mapping: move dma configuration to bus
> infrastructure
>
> It's bus specific aspect to map a given device on the bus and relevant firmware
> description of its DMA configuration.
> So, this change introduces '/dma_configure/' as bus callback giving flexibility to
> busses for implementing its own dma configuration function.
>
> The change eases the addition of new busses w.r.t. adding the dma
> configuration functionality.
>
> This patch also updates the PCI, Platform, ACPI and host1x bus to use new
> introduced callbacks.
>
> Suggested-by: Christoph Hellwig <[email protected]>
> Signed-off-by: Nipun Gupta <[email protected]>
> ---
> - The patches are based on the comments on:
> https://patchwork.kernel.org/patch/10259087/
>
> Changes in v2:
> - Do not have dma_deconfigure callback
> - Have '/dma_common_configure/' API to provide a common DMA
> configuration which can be used by busses if it suits them.
> - Platform and ACPI bus to use '/dma_common_configure/' in
> '/dma_configure/' callback.
> - Updated commit message
> - Updated pci_dma_configure API with changes suggested by Robin
>
> drivers/amba/bus.c | 7 +++++++
> drivers/base/dma-mapping.c | 35 +++++++++++++++--------------------
> drivers/base/platform.c | 6 ++++++
> drivers/gpu/host1x/bus.c | 9 +++++++++
> drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
> include/linux/device.h | 4 ++++
> include/linux/dma-mapping.h | 1 +
> 7 files changed, 74 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c index 594c228..2fa1e8b
> 100644
> --- a/drivers/amba/bus.c
> +++ b/drivers/amba/bus.c
> @@ -20,6 +20,7 @@
> #include <linux/sizes.h>
> #include <linux/limits.h>
> #include <linux/clk/clk-conf.h>
> +#include <linux/dma-mapping.h>
>
> #include <asm/irq.h>
>
> @@ -171,6 +172,11 @@ static int amba_pm_runtime_resume(struct device
> *dev) } #endif /* CONFIG_PM */
>
> +static int amba_dma_configure(struct device *dev) {
> + return dma_common_configure(dev);
> +}
> +
> static const struct dev_pm_ops amba_pm = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -194,6 +200,7 @@ struct bus_type amba_bustype = {
> .dev_groups = amba_dev_groups,
> .match = amba_match,
> .uevent = amba_uevent,
> + .dma_configure = amba_dma_configure,
> .pm = &amba_pm,
> .force_dma = true,
> };
> diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c index
> 3b11835..48f9af0 100644
> --- a/drivers/base/dma-mapping.c
> +++ b/drivers/base/dma-mapping.c
> @@ -331,38 +331,33 @@ void dma_common_free_remap(void *cpu_addr,
> size_t size, unsigned long vm_flags) #endif
>
> /*
> - * Common configuration to enable DMA API use for a device
> + * Common configuration to enable DMA API use for a device.
> + * A bus can use this function in its 'dma_configure' callback, if
> + * suitable for the bus.
> */
> -#include <linux/pci.h>
> -
> -int dma_configure(struct device *dev)
> +int dma_common_configure(struct device *dev)
> {
> - struct device *bridge = NULL, *dma_dev = dev;
> enum dev_dma_attr attr;
> int ret = 0;
>
> - if (dev_is_pci(dev)) {
> - bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> - dma_dev = bridge;
> - if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> - dma_dev->parent->of_node)
> - dma_dev = dma_dev->parent;
> - }
> -
> - if (dma_dev->of_node) {
> - ret = of_dma_configure(dev, dma_dev->of_node);
> - } else if (has_acpi_companion(dma_dev)) {
> - attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev-
> >fwnode));
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> if (attr != DEV_DMA_NOT_SUPPORTED)
> ret = acpi_dma_configure(dev, attr);
> }
>
> - if (bridge)
> - pci_put_host_bridge_device(bridge);
> -
> return ret;
> }
>
> +int dma_configure(struct device *dev)
> +{
> + if (dev->bus->dma_configure)
> + return dev->bus->dma_configure(dev);

What if dma_common_configure() is called in case "bus->dma_configure" is not defined?

Thanks
-Bharat

> +
> + return 0;
> +}
> void dma_deconfigure(struct device *dev) {
> of_dma_deconfigure(dev);
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c index
> f1bf7b3..d2d5891 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -1130,6 +1130,11 @@ int platform_pm_restore(struct device *dev)
>
> #endif /* CONFIG_HIBERNATE_CALLBACKS */
>
> +static int platform_dma_configure(struct device *dev) {
> + return dma_common_configure(dev);
> +}
> +
> static const struct dev_pm_ops platform_dev_pm_ops = {
> .runtime_suspend = pm_generic_runtime_suspend,
> .runtime_resume = pm_generic_runtime_resume, @@ -1141,6 +1146,7
> @@ struct bus_type platform_bus_type = {
> .dev_groups = platform_dev_groups,
> .match = platform_match,
> .uevent = platform_uevent,
> + .dma_configure = platform_dma_configure,
> .pm = &platform_dev_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c index
> 88a3558..fa9896d 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -314,6 +314,14 @@ static int host1x_device_match(struct device *dev,
> struct device_driver *drv)
> return strcmp(dev_name(dev), drv->name) == 0; }
>
> +static int host1x_dma_configure(struct device *dev) {
> + if (dev->of_node)
> + return of_dma_configure(dev, dev->of_node);
> +
> + return 0;
> +}
> +
> static const struct dev_pm_ops host1x_device_pm_ops = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -326,6 +334,7 @@ static int host1x_device_match(struct device *dev,
> struct device_driver *drv) struct bus_type host1x_bus_type = {
> .name = "host1x",
> .match = host1x_device_match,
> + .dma_configure = host1x_dma_configure,
> .pm = &host1x_device_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c index
> 3bed6be..b473a4c 100644
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -18,6 +18,8 @@
> #include <linux/pm_runtime.h>
> #include <linux/suspend.h>
> #include <linux/kexec.h>
> +#include <linux/of_device.h>
> +#include <linux/acpi.h>
> #include "pci.h"
>
> struct pci_dynid {
> @@ -1522,6 +1524,35 @@ static int pci_bus_num_vf(struct device *dev)
> return pci_num_vf(to_pci_dev(dev));
> }
>
> +/**
> + * pci_dma_configure - Setup DMA configuration
> + * @dev: ptr to dev structure
> + *
> + * Function to update PCI devices's DMA configuration using the same
> + * info from the OF node or ACPI node of host bridge's parent (if any).
> + */
> +static int pci_dma_configure(struct device *dev) {
> + struct device *bridge;
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> +
> + if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
> + bridge->parent->of_node) {
> + ret = of_dma_configure(dev, bridge->parent->of_node);
> + } else if (has_acpi_companion(bridge)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(bridge-
> >fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + pci_put_host_bridge_device(bridge);
> +
> + return ret;
> +}
> +
> struct bus_type pci_bus_type = {
> .name = "pci",
> .match = pci_bus_match,
> @@ -1534,6 +1565,7 @@ struct bus_type pci_bus_type = {
> .drv_groups = pci_drv_groups,
> .pm = PCI_PM_OPS_PTR,
> .num_vf = pci_bus_num_vf,
> + .dma_configure = pci_dma_configure,
> .force_dma = true,
> };
> EXPORT_SYMBOL(pci_bus_type);
> diff --git a/include/linux/device.h b/include/linux/device.h index
> b093405..1832d90 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type
> *,
> * @resume: Called to bring a device on this bus out of sleep mode.
> * @num_vf: Called to find out how many virtual functions a device on this
> * bus supports.
> + * @dma_configure: Called to setup DMA configuration on a device on
> + this bus.
> * @pm: Power management operations of this bus, callback the
> specific
> * device driver's pm-ops.
> * @iommu_ops: IOMMU specific operations for this bus, used to attach
> IOMMU @@ -130,6 +132,8 @@ struct bus_type {
>
> int (*num_vf)(struct device *dev);
>
> + int (*dma_configure)(struct device *dev);
> +
> const struct dev_pm_ops *pm;
>
> const struct iommu_ops *iommu_ops;
> diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index
> eb9eab4..c15986b 100644
> --- a/include/linux/dma-mapping.h
> +++ b/include/linux/dma-mapping.h
> @@ -761,6 +761,7 @@ void *dma_mark_declared_memory_occupied(struct
> device *dev, } #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
>
> +int dma_common_configure(struct device *dev);
> #ifdef CONFIG_HAS_DMA
> int dma_configure(struct device *dev);
> void dma_deconfigure(struct device *dev);
> --
> 1.9.1

2018-03-21 07:32:20

[permalink] [raw]

Subject: RE: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Bharat Bhushan
> Sent: Wednesday, March 21, 2018 12:49

> >
> > +int dma_configure(struct device *dev)
> > +{
> > + if (dev->bus->dma_configure)
> > + return dev->bus->dma_configure(dev);
>
> What if dma_common_configure() is called in case "bus->dma_configure" is
> not defined?
>
> Thanks
> -Bharat

I think it is cleaner for bus to call '/dma_common_configure/' rather
than this been called implicitly, but Robin/Christoph can comment
better on this.

Thanks,
Nipun

2018-03-21 09:31:05

[permalink] [raw]

Subject: Re: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

On Wed, Mar 21, 2018 at 12:25:22PM +0530, Nipun Gupta wrote:
> It's bus specific aspect to map a given device on the bus and
> relevant firmware description of its DMA configuration.
> So, this change introduces '/dma_configure/' as bus callback
> giving flexibility to busses for implementing its own dma
> configuration function.
>
> The change eases the addition of new busses w.r.t. adding the dma
> configuration functionality.
>
> This patch also updates the PCI, Platform, ACPI and host1x bus to
> use new introduced callbacks.
>
> Suggested-by: Christoph Hellwig <[email protected]>
> Signed-off-by: Nipun Gupta <[email protected]>
> ---
> - The patches are based on the comments on:
> https://patchwork.kernel.org/patch/10259087/
>
> Changes in v2:
> - Do not have dma_deconfigure callback
> - Have '/dma_common_configure/' API to provide a common DMA
> configuration which can be used by busses if it suits them.
> - Platform and ACPI bus to use '/dma_common_configure/' in
> '/dma_configure/' callback.
> - Updated commit message
> - Updated pci_dma_configure API with changes suggested by Robin
>
> drivers/amba/bus.c | 7 +++++++
> drivers/base/dma-mapping.c | 35 +++++++++++++++--------------------
> drivers/base/platform.c | 6 ++++++
> drivers/gpu/host1x/bus.c | 9 +++++++++
> drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
> include/linux/device.h | 4 ++++
> include/linux/dma-mapping.h | 1 +
> 7 files changed, 74 insertions(+), 20 deletions(-)
>
> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
> index 594c228..2fa1e8b 100644
> --- a/drivers/amba/bus.c
> +++ b/drivers/amba/bus.c
> @@ -20,6 +20,7 @@
> #include <linux/sizes.h>
> #include <linux/limits.h>
> #include <linux/clk/clk-conf.h>
> +#include <linux/dma-mapping.h>
>
> #include <asm/irq.h>
>
> @@ -171,6 +172,11 @@ static int amba_pm_runtime_resume(struct device *dev)
> }
> #endif /* CONFIG_PM */
>
> +static int amba_dma_configure(struct device *dev)
> +{
> + return dma_common_configure(dev);
> +}
> +
> static const struct dev_pm_ops amba_pm = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -194,6 +200,7 @@ struct bus_type amba_bustype = {
> .dev_groups = amba_dev_groups,
> .match = amba_match,
> .uevent = amba_uevent,
> + .dma_configure = amba_dma_configure,
> .pm = &amba_pm,
> .force_dma = true,
> };
> diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
> index 3b11835..48f9af0 100644
> --- a/drivers/base/dma-mapping.c
> +++ b/drivers/base/dma-mapping.c
> @@ -331,38 +331,33 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
> #endif
>
> /*
> - * Common configuration to enable DMA API use for a device
> + * Common configuration to enable DMA API use for a device.
> + * A bus can use this function in its 'dma_configure' callback, if
> + * suitable for the bus.
> */
> -#include <linux/pci.h>
> -
> -int dma_configure(struct device *dev)
> +int dma_common_configure(struct device *dev)
> {
> - struct device *bridge = NULL, *dma_dev = dev;
> enum dev_dma_attr attr;
> int ret = 0;
>
> - if (dev_is_pci(dev)) {
> - bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> - dma_dev = bridge;
> - if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> - dma_dev->parent->of_node)
> - dma_dev = dma_dev->parent;
> - }
> -
> - if (dma_dev->of_node) {
> - ret = of_dma_configure(dev, dma_dev->of_node);
> - } else if (has_acpi_companion(dma_dev)) {
> - attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> if (attr != DEV_DMA_NOT_SUPPORTED)
> ret = acpi_dma_configure(dev, attr);
> }
>
> - if (bridge)
> - pci_put_host_bridge_device(bridge);
> -
> return ret;
> }
>
> +int dma_configure(struct device *dev)
> +{
> + if (dev->bus->dma_configure)
> + return dev->bus->dma_configure(dev);
> +
> + return 0;
> +}
> void dma_deconfigure(struct device *dev)

Empty line after this new function? Sorry, couldn't help it :)

> {
> of_dma_deconfigure(dev);
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index f1bf7b3..d2d5891 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -1130,6 +1130,11 @@ int platform_pm_restore(struct device *dev)
>
> #endif /* CONFIG_HIBERNATE_CALLBACKS */
>
> +static int platform_dma_configure(struct device *dev)
> +{
> + return dma_common_configure(dev);
> +}
> +
> static const struct dev_pm_ops platform_dev_pm_ops = {
> .runtime_suspend = pm_generic_runtime_suspend,
> .runtime_resume = pm_generic_runtime_resume,
> @@ -1141,6 +1146,7 @@ struct bus_type platform_bus_type = {
> .dev_groups = platform_dev_groups,
> .match = platform_match,
> .uevent = platform_uevent,
> + .dma_configure = platform_dma_configure,
> .pm = &platform_dev_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index 88a3558..fa9896d 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -314,6 +314,14 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> return strcmp(dev_name(dev), drv->name) == 0;
> }
>
> +static int host1x_dma_configure(struct device *dev)
> +{
> + if (dev->of_node)
> + return of_dma_configure(dev, dev->of_node);
> +
> + return 0;
> +}
> +
> static const struct dev_pm_ops host1x_device_pm_ops = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -326,6 +334,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> struct bus_type host1x_bus_type = {
> .name = "host1x",
> .match = host1x_device_match,
> + .dma_configure = host1x_dma_configure,
> .pm = &host1x_device_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> index 3bed6be..b473a4c 100644
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -18,6 +18,8 @@
> #include <linux/pm_runtime.h>
> #include <linux/suspend.h>
> #include <linux/kexec.h>
> +#include <linux/of_device.h>
> +#include <linux/acpi.h>
> #include "pci.h"
>
> struct pci_dynid {
> @@ -1522,6 +1524,35 @@ static int pci_bus_num_vf(struct device *dev)
> return pci_num_vf(to_pci_dev(dev));
> }
>
> +/**
> + * pci_dma_configure - Setup DMA configuration
> + * @dev: ptr to dev structure
> + *
> + * Function to update PCI devices's DMA configuration using the same
> + * info from the OF node or ACPI node of host bridge's parent (if any).
> + */
> +static int pci_dma_configure(struct device *dev)
> +{
> + struct device *bridge;
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> +
> + if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
> + bridge->parent->of_node) {
> + ret = of_dma_configure(dev, bridge->parent->of_node);
> + } else if (has_acpi_companion(bridge)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(bridge->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + pci_put_host_bridge_device(bridge);
> +
> + return ret;
> +}
> +
> struct bus_type pci_bus_type = {
> .name = "pci",
> .match = pci_bus_match,
> @@ -1534,6 +1565,7 @@ struct bus_type pci_bus_type = {
> .drv_groups = pci_drv_groups,
> .pm = PCI_PM_OPS_PTR,
> .num_vf = pci_bus_num_vf,
> + .dma_configure = pci_dma_configure,
> .force_dma = true,
> };
> EXPORT_SYMBOL(pci_bus_type);
> diff --git a/include/linux/device.h b/include/linux/device.h
> index b093405..1832d90 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type *,
> * @resume: Called to bring a device on this bus out of sleep mode.
> * @num_vf: Called to find out how many virtual functions a device on this
> * bus supports.
> + * @dma_configure: Called to setup DMA configuration on a device on
> + this bus.
> * @pm: Power management operations of this bus, callback the specific
> * device driver's pm-ops.
> * @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
> @@ -130,6 +132,8 @@ struct bus_type {
>
> int (*num_vf)(struct device *dev);
>
> + int (*dma_configure)(struct device *dev);
> +
> const struct dev_pm_ops *pm;
>
> const struct iommu_ops *iommu_ops;
> diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> index eb9eab4..c15986b 100644
> --- a/include/linux/dma-mapping.h
> +++ b/include/linux/dma-mapping.h
> @@ -761,6 +761,7 @@ void *dma_mark_declared_memory_occupied(struct device *dev,
> }
> #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
>
> +int dma_common_configure(struct device *dev);
> #ifdef CONFIG_HAS_DMA

Blank line after the new function declaration?

Other than those very minor things, nice job, this looks good:

Reviewed-by: Greg Kroah-Hartman <[email protected]>

2018-03-21 09:36:42

[permalink] [raw]

Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses

On Wed, Mar 21, 2018 at 12:25:23PM +0530, Nipun Gupta wrote:
> With each bus implementing its own DMA configuration callback,
> there is no need for bus to explicitly have force_dma in its
> global structure. This patch modifies of_dma_configure API to
> accept an input parameter which specifies if implicit DMA
> configuration is required even when it is not described by the
> firmware.

Having to "remember" what that bool variable means on the end of the
function call is a royal pain over time, right?

Why not just create a new function:
dma_common_configure_force(dma)
that always does this? Leave "dma_common_configure()" alone, and then
wrap the old code with these two helper functions that call the 'core'
code with the bool set properly?

That way you do not have to "know" what that parameter is, the function
name just documents it automatically, so when you see it in the
bus-specific code, no need to go and have to hunt for anything. And if
you are reading the dma core code, it's obvious what is happening as the
functions are all right there.

thanks,

greg k-h

2018-03-21 16:15:31

[permalink] [raw]

Subject: RE: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Greg KH [mailto:[email protected]]
> Sent: Wednesday, March 21, 2018 15:00

<snip>

> > +int dma_configure(struct device *dev)
> > +{
> > + if (dev->bus->dma_configure)
> > + return dev->bus->dma_configure(dev);
> > +
> > + return 0;
> > +}
> > void dma_deconfigure(struct device *dev)
>
> Empty line after this new function? Sorry, couldn't help it :)
>
> > {
> > of_dma_deconfigure(dev);
> > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > index f1bf7b3..d2d5891 100644
> > --- a/drivers/base/platform.c
> > +++ b/drivers/base/platform.c
> > @@ -1130,6 +1130,11 @@ int platform_pm_restore(struct device *dev)
> >
> > #endif /* CONFIG_HIBERNATE_CALLBACKS */
> >

<snip>

> > +
> > const struct dev_pm_ops *pm;
> >
> > const struct iommu_ops *iommu_ops;
> > diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
> > index eb9eab4..c15986b 100644
> > --- a/include/linux/dma-mapping.h
> > +++ b/include/linux/dma-mapping.h
> > @@ -761,6 +761,7 @@ void *dma_mark_declared_memory_occupied(struct
> device *dev,
> > }
> > #endif /* CONFIG_HAVE_GENERIC_DMA_COHERENT */
> >
> > +int dma_common_configure(struct device *dev);
> > #ifdef CONFIG_HAS_DMA
>
> Blank line after the new function declaration?
>
> Other than those very minor things, nice job, this looks good:
>
> Reviewed-by: Greg Kroah-Hartman <[email protected]>

Thank you for the review :). I will fix your comments in next version.

Regards,
Nipun

2018-03-21 16:30:11

[permalink] [raw]

Subject: RE: [PATCH v2 2/2] drivers: remove force dma flag from buses

> -----Original Message-----
> From: Greg KH [mailto:[email protected]]
> Sent: Wednesday, March 21, 2018 15:05
> To: Nipun Gupta <[email protected]>
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected]
> foundation.org; [email protected]; linux-arm-
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]; [email protected]; Bharat Bhushan
> <[email protected]>; Leo Li <[email protected]>
> Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses
>
> On Wed, Mar 21, 2018 at 12:25:23PM +0530, Nipun Gupta wrote:
> > With each bus implementing its own DMA configuration callback,
> > there is no need for bus to explicitly have force_dma in its
> > global structure. This patch modifies of_dma_configure API to
> > accept an input parameter which specifies if implicit DMA
> > configuration is required even when it is not described by the
> > firmware.
>
> Having to "remember" what that bool variable means on the end of the
> function call is a royal pain over time, right?
>
> Why not just create a new function:
> dma_common_configure_force(dma)
> that always does this? Leave "dma_common_configure()" alone, and then
> wrap the old code with these two helper functions that call the 'core'
> code with the bool set properly?
>
> That way you do not have to "know" what that parameter is, the function
> name just documents it automatically, so when you see it in the
> bus-specific code, no need to go and have to hunt for anything. And if
> you are reading the dma core code, it's obvious what is happening as the
> functions are all right there.

How about we do not pass any flag in 'dma_common_configure()', and inside this
API we pass "true" to 'of_dma_configure()'? I am saying this because currently
both the busses (platform and AMBA) which uses 'dma_common_configure()' passes
"true" value. If we create additional 'dma_common_configure_force()', then
'dma_common_configure()' will not be used anytime and will become redundant.

If someday new busses come and they needs to use similar functionality which
'dma_common_configure()' provides, but with passing "false" to 'of_dma_configure()',
then what you suggests of having two separate such API's will be more reasonable
and can be implemented?

Thanks,
Nipun

>
> thanks,
>
> greg k-h

2018-03-21 17:51:29

[permalink] [raw]

Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses

On Wed, Mar 21, 2018 at 04:28:46PM +0000, Nipun Gupta wrote:
>
>
> > -----Original Message-----
> > From: Greg KH [mailto:[email protected]]
> > Sent: Wednesday, March 21, 2018 15:05
> > To: Nipun Gupta <[email protected]>
> > Cc: [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected];
> > [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected]; [email protected]
> > foundation.org; [email protected]; linux-arm-
> > [email protected]; [email protected]; [email protected];
> > [email protected]; [email protected];
> > [email protected]; [email protected]; Bharat Bhushan
> > <[email protected]>; Leo Li <[email protected]>
> > Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses
> >
> > On Wed, Mar 21, 2018 at 12:25:23PM +0530, Nipun Gupta wrote:
> > > With each bus implementing its own DMA configuration callback,
> > > there is no need for bus to explicitly have force_dma in its
> > > global structure. This patch modifies of_dma_configure API to
> > > accept an input parameter which specifies if implicit DMA
> > > configuration is required even when it is not described by the
> > > firmware.
> >
> > Having to "remember" what that bool variable means on the end of the
> > function call is a royal pain over time, right?
> >
> > Why not just create a new function:
> > dma_common_configure_force(dma)
> > that always does this? Leave "dma_common_configure()" alone, and then
> > wrap the old code with these two helper functions that call the 'core'
> > code with the bool set properly?
> >
> > That way you do not have to "know" what that parameter is, the function
> > name just documents it automatically, so when you see it in the
> > bus-specific code, no need to go and have to hunt for anything. And if
> > you are reading the dma core code, it's obvious what is happening as the
> > functions are all right there.
>
> How about we do not pass any flag in 'dma_common_configure()', and inside this
> API we pass "true" to 'of_dma_configure()'? I am saying this because currently
> both the busses (platform and AMBA) which uses 'dma_common_configure()' passes
> "true" value. If we create additional 'dma_common_configure_force()', then
> 'dma_common_configure()' will not be used anytime and will become redundant.
>
> If someday new busses come and they needs to use similar functionality which
> 'dma_common_configure()' provides, but with passing "false" to 'of_dma_configure()',
> then what you suggests of having two separate such API's will be more reasonable
> and can be implemented?

If that makes things "simple", yes, sounds good.

greg k-h

2018-03-22 08:16:49

[permalink] [raw]

Subject: Re: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> +static int amba_dma_configure(struct device *dev)
> +{
> + return dma_common_configure(dev);
> +}

So it turns out we only end with two callers of dma_common_configure
after this series. Based ont hat I'm tempted with the suggestion
from Robin to just have amba call platform_dma_configure, and move
the code from dma_common_configure to platform_dma_configure.

> +int dma_configure(struct device *dev)
> +{
> + if (dev->bus->dma_configure)
> + return dev->bus->dma_configure(dev);
> +
> + return 0;
> +}
> void dma_deconfigure(struct device *dev)

As grep pointed out this wants a blank line after the function, and while
in nitpicking mode I'd suggest to remove the blank line above the return
statement while at it.

> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index 88a3558..fa9896d 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -314,6 +314,14 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> return strcmp(dev_name(dev), drv->name) == 0;
> }
>
> +static int host1x_dma_configure(struct device *dev)
> +{
> + if (dev->of_node)
> + return of_dma_configure(dev, dev->of_node);
> +
> + return 0;

Same here.

> + */
> +static int pci_dma_configure(struct device *dev)
> +{
> + struct device *bridge;
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> +
> + if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
> + bridge->parent->of_node) {
> + ret = of_dma_configure(dev, bridge->parent->of_node);
> + } else if (has_acpi_companion(bridge)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(bridge->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }

The attr declaration can be moved into the inner scope here.

> + pci_put_host_bridge_device(bridge);
> +
> + return ret;

Drop the blank line after the return, please.

2018-03-22 08:18:33

[permalink] [raw]

Subject: Re: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> > +int dma_configure(struct device *dev)
> > +{
> > + if (dev->bus->dma_configure)
> > + return dev->bus->dma_configure(dev);
>
> What if dma_common_configure() is called in case "bus->dma_configure" is not defined?

Then we'd still have a dependency of common code on OF and ACPI.

On the other hand we'd get free OF and ACPI dma ranges parsing for
everyone, which might be handy. And which would really help mitigating
the risk that we missed some bus that gets dma configuration from OF,
so maybe it actually is a good idea. I'd just rename it to
dma_default_configure or similar in that case.

2018-03-22 08:22:02

[permalink] [raw]

Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses

> --- a/drivers/dma/qcom/hidma_mgmt.c
> +++ b/drivers/dma/qcom/hidma_mgmt.c
> @@ -398,7 +398,7 @@ static int __init hidma_mgmt_of_populate_channels(struct device_node *np)
> }
> of_node_get(child);
> new_pdev->dev.of_node = child;
> - of_dma_configure(&new_pdev->dev, child);
> + of_dma_configure(&new_pdev->dev, child, true);
> /*
> * It is assumed that calling of_msi_configure is safe on
> * platforms with or without MSI support.

Where did we mark this bus as force_dma before?

> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> index 9a4f4246..895c83e 100644
> --- a/drivers/of/of_reserved_mem.c
> +++ b/drivers/of/of_reserved_mem.c
> @@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device *dev,
> /* ensure that dma_ops is set for virtual devices
> * using reserved memory
> */
> - of_dma_configure(dev, np);
> + of_dma_configure(dev, np, true);

Did all the callers of this one really force dma? I have a hard time
untangling the call stacks unfortunately.

2018-03-22 15:08:03

[permalink] [raw]

Subject: RE: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

> -----Original Message-----
> From: Christoph Hellwig [mailto:[email protected]]
> Sent: Thursday, March 22, 2018 13:46
> To: Nipun Gupta <[email protected]>
>
> > +static int amba_dma_configure(struct device *dev)
> > +{
> > + return dma_common_configure(dev);
> > +}
>
> So it turns out we only end with two callers of dma_common_configure
> after this series. Based ont hat I'm tempted with the suggestion
> from Robin to just have amba call platform_dma_configure, and move
> the code from dma_common_configure to platform_dma_configure.

okay, that would be fine, trivial query - will it be okay to include
'linux/platform_device.h' in the AMBA bus? I am reluctant for this change
because of including platform file.

Thanks,
Nipun

2018-03-22 15:16:18

[permalink] [raw]

Subject: RE: [PATCH v2 2/2] drivers: remove force dma flag from buses

> -----Original Message-----
> From: Christoph Hellwig [mailto:[email protected]]
> Sent: Thursday, March 22, 2018 13:49
> To: Nipun Gupta <[email protected]>
>
> > --- a/drivers/dma/qcom/hidma_mgmt.c
> > +++ b/drivers/dma/qcom/hidma_mgmt.c
> > @@ -398,7 +398,7 @@ static int __init
> hidma_mgmt_of_populate_channels(struct device_node *np)
> > }
> > of_node_get(child);
> > new_pdev->dev.of_node = child;
> > - of_dma_configure(&new_pdev->dev, child);
> > + of_dma_configure(&new_pdev->dev, child, true);
> > /*
> > * It is assumed that calling of_msi_configure is safe on
> > * platforms with or without MSI support.
>
> Where did we mark this bus as force_dma before?

I thought these devices to be on the platform bus as the device is of type
'struct platform_device', though I am not sure then why 'of_dma_configure()'
is called here. Is this not on platform bus?

>
> > diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> > index 9a4f4246..895c83e 100644
> > --- a/drivers/of/of_reserved_mem.c
> > +++ b/drivers/of/of_reserved_mem.c
> > @@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device
> *dev,
> > /* ensure that dma_ops is set for virtual devices
> > * using reserved memory
> > */
> > - of_dma_configure(dev, np);
> > + of_dma_configure(dev, np, true);
>
> Did all the callers of this one really force dma? I have a hard time
> untangling the call stacks unfortunately.

I see this API being called indirectly from NXP DPAA device driver which
is for platform bus devices. So I marked 'true' out here. There are more places
from where it is being called.

Thanks,
Nipun

2018-03-23 16:12:48

[permalink] [raw]

Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses

Hi Nipun,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on linus/master]
[also build test ERROR on v4.16-rc6 next-20180323]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Nipun-Gupta/dma-mapping-move-dma-configuration-to-bus-infrastructure/20180323-232307
config: i386-randconfig-x013-201811 (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
# save the attached .config to linux build tree
make ARCH=i386

All errors (new ones prefixed by >>):

drivers/base/dma-mapping.c: In function 'dma_common_configure':
>> drivers/base/dma-mapping.c:344:9: error: too many arguments to function 'of_dma_configure'
ret = of_dma_configure(dev, dev->of_node, force_dma);
^~~~~~~~~~~~~~~~
In file included from drivers/base/dma-mapping.c:13:0:
include/linux/of_device.h:110:19: note: declared here
static inline int of_dma_configure(struct device *dev, struct device_node *np)
^~~~~~~~~~~~~~~~
--
drivers/pci/pci-driver.c: In function 'pci_dma_configure':
>> drivers/pci/pci-driver.c:1544:9: error: too many arguments to function 'of_dma_configure'
ret = of_dma_configure(dev, bridge->parent->of_node, true);
^~~~~~~~~~~~~~~~
In file included from drivers/pci/pci-driver.c:21:0:
include/linux/of_device.h:110:19: note: declared here
static inline int of_dma_configure(struct device *dev, struct device_node *np)
^~~~~~~~~~~~~~~~

vim +/of_dma_configure +344 drivers/base/dma-mapping.c

332
333 /*
334 * Common configuration to enable DMA API use for a device.
335 * A bus can use this function in its 'dma_configure' callback, if
336 * suitable for the bus.
337 */
338 int dma_common_configure(struct device *dev, bool force_dma)
339 {
340 enum dev_dma_attr attr;
341 int ret = 0;
342
343 if (dev->of_node) {
> 344 ret = of_dma_configure(dev, dev->of_node, force_dma);
345 } else if (has_acpi_companion(dev)) {
346 attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
347 if (attr != DEV_DMA_NOT_SUPPORTED)
348 ret = acpi_dma_configure(dev, attr);
349 }
350
351 return ret;
352 }
353

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

Attachments:

(No filename) (2.50 kB)
.config.gz (25.25 kB)
Download all attachments

2018-03-23 17:42:47

[permalink] [raw]

Subject: Re: [PATCH v2 2/2] drivers: remove force dma flag from buses

Attachments:

(No filename) (1.75 kB)
.config.gz (30.94 kB)
Download all attachments

2018-03-24 09:28:03

[permalink] [raw]

Subject: Re: [PATCH v2 1/2] dma-mapping: move dma configuration to bus infrastructure

Hi Nipun,

Thank you for the patch! Yet something to improve:

[auto build test ERROR on linus/master]
[also build test ERROR on v4.16-rc6]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url: https://github.com/0day-ci/linux/commits/Nipun-Gupta/dma-mapping-move-dma-configuration-to-bus-infrastructure/20180323-232307
config: score-spct6600_defconfig (attached as .config)
compiler: score-elf-gcc (GCC) 4.9.4
reproduce:
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
make.cross ARCH=score

All errors (new ones prefixed by >>):

init/main.o: In function `try_to_run_init_process':
main.c:(.text+0x40): relocation truncated to fit: R_SCORE_24 against `run_init_process'
main.c:(.text+0x78): relocation truncated to fit: R_SCORE_24 against `.L36'
init/main.o: In function `.L132':
main.c:(.text+0x190): relocation truncated to fit: R_SCORE_24 against `.L129'
main.c:(.text+0x200): relocation truncated to fit: R_SCORE_24 against `.L132'
init/main.o: In function `loglevel':
main.c:(.init.text+0xa4): relocation truncated to fit: R_SCORE_24 against `get_option'
init/main.o: In function `.L15':
main.c:(.init.text+0x110): relocation truncated to fit: R_SCORE_24 against `strcmp'
main.c:(.init.text+0x124): relocation truncated to fit: R_SCORE_24 against `parameq'
main.c:(.init.text+0x14c): relocation truncated to fit: R_SCORE_24 against `printk'
init/main.o: In function `.L31':
main.c:(.init.text+0x160): relocation truncated to fit: R_SCORE_24 against `strcmp'
init/main.o: In function `.L21':
main.c:(.init.text+0x170): relocation truncated to fit: R_SCORE_24 against `.L15'
init/main.o: In function `initcall_blacklist':
main.c:(.init.text+0x198): additional relocation overflows omitted from the output
drivers/base/platform.o: In function `platform_dma_configure':
>> platform.c:(.text.platform_dma_configure+0x0): undefined reference to `dma_common_configure'

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

Attachments:

(No filename) (2.29 kB)
.config.gz (7.81 kB)
Download all attachments

2018-03-30 07:17:34

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

Can you resend the current state of affairs so we can get it in for
4.17?

2018-03-30 07:18:30

[permalink] [raw]

Subject: RE: [PATCH] dma-mapping: move dma configuration to bus infrastructure

I am just going to send it within an hour :)

Thanks,
Nipun

> -----Original Message-----
> From: Christoph Hellwig [mailto:[email protected]]
> Sent: Friday, March 30, 2018 12:46
> To: Nipun Gupta <[email protected]>
> Cc: [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]; [email protected];
> [email protected]; [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: Re: [PATCH] dma-mapping: move dma configuration to bus
> infrastructure
>
> Can you resend the current state of affairs so we can get it in for
> 4.17?

2018-03-30 07:57:07

[permalink] [raw]

Subject: [PATCH v3 2/2] drivers: remove force dma flag from buses

With each bus implementing its own DMA configuration callback,
there is no need for bus to explicitly have force_dma in its
global structure. This patch modifies of_dma_configure API to
accept an input parameter which specifies if implicit DMA
configuration is required even when it is not described by the
firmware.

Signed-off-by: Nipun Gupta <[email protected]>
---
Changes in v2:
- This is a new change suggested by Robin and Christoph
and is added to the series.

Changes in v3:
- Rebase and changes corresponding to the changes in patch 1/2

drivers/amba/bus.c | 1 -
drivers/base/platform.c | 3 +--
drivers/bcma/main.c | 2 +-
drivers/dma/qcom/hidma_mgmt.c | 2 +-
drivers/gpu/host1x/bus.c | 5 ++---
drivers/of/device.c | 6 ++++--
drivers/of/of_reserved_mem.c | 2 +-
drivers/pci/pci-driver.c | 3 +--
include/linux/device.h | 4 ----
include/linux/of_device.h | 8 ++++++--
10 files changed, 17 insertions(+), 19 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 867dc2b..abe73c4 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -199,7 +199,6 @@ struct bus_type amba_bustype = {
.uevent = amba_uevent,
.dma_configure = platform_dma_configure,
.pm = &amba_pm,
- .force_dma = true,
};

static int __init amba_init(void)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 72fdbf6..cfbc569 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1136,7 +1136,7 @@ int platform_dma_configure(struct device *dev)
int ret = 0;

if (dev->of_node) {
- ret = of_dma_configure(dev, dev->of_node);
+ ret = of_dma_configure(dev, dev->of_node, true);
} else if (has_acpi_companion(dev)) {
attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
if (attr != DEV_DMA_NOT_SUPPORTED)
@@ -1159,7 +1159,6 @@ struct bus_type platform_bus_type = {
.uevent = platform_uevent,
.dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
- .force_dma = true,
};
EXPORT_SYMBOL_GPL(platform_bus_type);

diff --git a/drivers/bcma/main.c b/drivers/bcma/main.c
index e6986c7..fc1f4ac 100644
--- a/drivers/bcma/main.c
+++ b/drivers/bcma/main.c
@@ -207,7 +207,7 @@ static void bcma_of_fill_device(struct device *parent,

core->irq = bcma_of_get_irq(parent, core, 0);

- of_dma_configure(&core->dev, node);
+ of_dma_configure(&core->dev, node, false);
}

unsigned int bcma_core_irq(struct bcma_device *core, int num)
diff --git a/drivers/dma/qcom/hidma_mgmt.c b/drivers/dma/qcom/hidma_mgmt.c
index 000c7019..d64edeb 100644
--- a/drivers/dma/qcom/hidma_mgmt.c
+++ b/drivers/dma/qcom/hidma_mgmt.c
@@ -398,7 +398,7 @@ static int __init hidma_mgmt_of_populate_channels(struct device_node *np)
}
of_node_get(child);
new_pdev->dev.of_node = child;
- of_dma_configure(&new_pdev->dev, child);
+ of_dma_configure(&new_pdev->dev, child, true);
/*
* It is assumed that calling of_msi_configure is safe on
* platforms with or without MSI support.
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index a9ec99d..b39c1e9 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -317,7 +317,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
static int host1x_dma_configure(struct device *dev)
{
if (dev->of_node)
- return of_dma_configure(dev, dev->of_node);
+ return of_dma_configure(dev, dev->of_node, true);
return 0;
}

@@ -335,7 +335,6 @@ struct bus_type host1x_bus_type = {
.match = host1x_device_match,
.dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
- .force_dma = true,
};

static void __host1x_device_del(struct host1x_device *device)
@@ -424,7 +423,7 @@ static int host1x_device_add(struct host1x *host1x,
device->dev.bus = &host1x_bus_type;
device->dev.parent = host1x->dev;

- of_dma_configure(&device->dev, host1x->dev->of_node);
+ of_dma_configure(&device->dev, host1x->dev->of_node, true);

err = host1x_device_parse_dt(device, driver);
if (err < 0) {
diff --git a/drivers/of/device.c b/drivers/of/device.c
index 064c818..33d8551 100644
--- a/drivers/of/device.c
+++ b/drivers/of/device.c
@@ -76,6 +76,8 @@ int of_device_add(struct platform_device *ofdev)
* of_dma_configure - Setup DMA configuration
* @dev: Device to apply DMA configuration
* @np: Pointer to OF node having DMA configuration
+ * @force_dma: Whether device is to be set up by of_dma_configure() even if
+ * DMA capability is not explicitly described by firmware.
*
* Try to get devices's DMA configuration from DT and update it
* accordingly.
@@ -84,7 +86,7 @@ int of_device_add(struct platform_device *ofdev)
* can use a platform bus notifier and handle BUS_NOTIFY_ADD_DEVICE events
* to fix up DMA configuration.
*/
-int of_dma_configure(struct device *dev, struct device_node *np)
+int of_dma_configure(struct device *dev, struct device_node *np, bool force_dma)
{
u64 dma_addr, paddr, size = 0;
int ret;
@@ -100,7 +102,7 @@ int of_dma_configure(struct device *dev, struct device_node *np)
* DMA configuration regardless of whether "dma-ranges" is
* correctly specified or not.
*/
- if (!dev->bus->force_dma)
+ if (!force_dma)
return ret == -ENODEV ? 0 : ret;

dma_addr = offset = 0;
diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
index 9a4f4246..895c83e 100644
--- a/drivers/of/of_reserved_mem.c
+++ b/drivers/of/of_reserved_mem.c
@@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device *dev,
/* ensure that dma_ops is set for virtual devices
* using reserved memory
*/
- of_dma_configure(dev, np);
+ of_dma_configure(dev, np, true);

dev_info(dev, "assigned reserved memory node %s\n", rmem->name);
} else {
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index 78e507f..27c832f 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -1540,7 +1540,7 @@ static int pci_dma_configure(struct device *dev)

if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
bridge->parent->of_node) {
- ret = of_dma_configure(dev, bridge->parent->of_node);
+ ret = of_dma_configure(dev, bridge->parent->of_node, true);
} else if (has_acpi_companion(bridge)) {
struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
enum dev_dma_attr attr = acpi_get_dma_attr(adev);
@@ -1566,7 +1566,6 @@ struct bus_type pci_bus_type = {
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
.dma_configure = pci_dma_configure,
- .force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);

diff --git a/include/linux/device.h b/include/linux/device.h
index 1832d90..92c530e 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -98,8 +98,6 @@ extern int __must_check bus_create_file(struct bus_type *,
* @p: The private data of the driver core, only the driver core can
* touch this.
* @lock_key: Lock class key for use by the lock validator
- * @force_dma: Assume devices on this bus should be set up by dma_configure()
- * even if DMA capability is not explicitly described by firmware.
*
* A bus is a channel between the processor and one or more devices. For the
* purposes of the device model, all devices are connected via a bus, even if
@@ -140,8 +138,6 @@ struct bus_type {

struct subsys_private *p;
struct lock_class_key lock_key;
-
- bool force_dma;
};

extern int __must_check bus_register(struct bus_type *bus);
diff --git a/include/linux/of_device.h b/include/linux/of_device.h
index 8da5a1b..165fd30 100644
--- a/include/linux/of_device.h
+++ b/include/linux/of_device.h
@@ -55,7 +55,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
return of_node_get(cpu_dev->of_node);
}

-int of_dma_configure(struct device *dev, struct device_node *np);
+int of_dma_configure(struct device *dev,
+ struct device_node *np,
+ bool force_dma);
void of_dma_deconfigure(struct device *dev);
#else /* CONFIG_OF */

@@ -105,7 +107,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
return NULL;
}

-static inline int of_dma_configure(struct device *dev, struct device_node *np)
+static inline int of_dma_configure(struct device *dev,
+ struct device_node *np,
+ bool force_dma)
{
return 0;
}
--
1.9.1

2018-03-30 07:57:49

[permalink] [raw]

Subject: [PATCH v3 1/2] dma-mapping: move dma configuration to bus infrastructure

It is bus specific aspect to map a given device on the bus and
relevant firmware description of its DMA configuration.
So, this change introduces 'dma_configure' as bus callback
giving flexibility to busses for implementing its own dma
configuration function.

The change eases the addition of new busses w.r.t. adding the dma
configuration functionality.

This patch also updates the PCI, Platform, ACPI and host1x bus to
use new introduced callbacks.

Suggested-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nipun Gupta <[email protected]>
Reviewed-by: Greg Kroah-Hartman <[email protected]>
---
- The patches are based on the comments on:
https://patchwork.kernel.org/patch/10259087/

Changes in v2:
- Do not have dma_deconfigure callback
- Have '/dma_common_configure/' API to provide a common DMA
configuration which can be used by busses if it suits them.
- Platform and ACPI bus to use '/dma_common_configure/' in
'/dma_configure/' callback.
- Updated commit message
- Updated pci_dma_configure API with changes suggested by Robin

Changes in v3
- Move dma_common_configure() within platform_dma_configure() and
reuse platofrm_dma_configure() for AMBA bus too
- Declare 'attr' in pci_dma_configure() inside the else statement
where it is used.

drivers/amba/bus.c | 4 ++++
drivers/base/dma-mapping.c | 31 ++++---------------------------
drivers/base/platform.c | 17 +++++++++++++++++
drivers/gpu/host1x/bus.c | 8 ++++++++
drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
include/linux/device.h | 4 ++++
include/linux/platform_device.h | 2 ++
7 files changed, 71 insertions(+), 27 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 594c228..867dc2b 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -20,6 +20,7 @@
#include <linux/sizes.h>
#include <linux/limits.h>
#include <linux/clk/clk-conf.h>
+#include <linux/platform_device.h>

#include <asm/irq.h>

@@ -188,12 +189,15 @@ static int amba_pm_runtime_resume(struct device *dev)
/*
* Primecells are part of the Advanced Microcontroller Bus Architecture,
* so we call the bus "amba".
+ * DMA configuration for platform and AMBA bus is same. So here we reuse
+ * platform's DMA config routine.
*/
struct bus_type amba_bustype = {
.name = "amba",
.dev_groups = amba_dev_groups,
.match = amba_match,
.uevent = amba_uevent,
+ .dma_configure = platform_dma_configure,
.pm = &amba_pm,
.force_dma = true,
};
diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
index 3b11835..fdc1502 100644
--- a/drivers/base/dma-mapping.c
+++ b/drivers/base/dma-mapping.c
@@ -331,36 +331,13 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
#endif

/*
- * Common configuration to enable DMA API use for a device
+ * enables DMA API use for a device
*/
-#include <linux/pci.h>
-
int dma_configure(struct device *dev)
{
- struct device *bridge = NULL, *dma_dev = dev;
- enum dev_dma_attr attr;
- int ret = 0;
-
- if (dev_is_pci(dev)) {
- bridge = pci_get_host_bridge_device(to_pci_dev(dev));
- dma_dev = bridge;
- if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
- dma_dev->parent->of_node)
- dma_dev = dma_dev->parent;
- }
-
- if (dma_dev->of_node) {
- ret = of_dma_configure(dev, dma_dev->of_node);
- } else if (has_acpi_companion(dma_dev)) {
- attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
- if (attr != DEV_DMA_NOT_SUPPORTED)
- ret = acpi_dma_configure(dev, attr);
- }
-
- if (bridge)
- pci_put_host_bridge_device(bridge);
-
- return ret;
+ if (dev->bus->dma_configure)
+ return dev->bus->dma_configure(dev);
+ return 0;
}

void dma_deconfigure(struct device *dev)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index f1bf7b3..72fdbf6 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1130,6 +1130,22 @@ int platform_pm_restore(struct device *dev)

#endif /* CONFIG_HIBERNATE_CALLBACKS */

+int platform_dma_configure(struct device *dev)
+{
+ enum dev_dma_attr attr;
+ int ret = 0;
+
+ if (dev->of_node) {
+ ret = of_dma_configure(dev, dev->of_node);
+ } else if (has_acpi_companion(dev)) {
+ attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
+ if (attr != DEV_DMA_NOT_SUPPORTED)
+ ret = acpi_dma_configure(dev, attr);
+ }
+
+ return ret;
+}
+
static const struct dev_pm_ops platform_dev_pm_ops = {
.runtime_suspend = pm_generic_runtime_suspend,
.runtime_resume = pm_generic_runtime_resume,
@@ -1141,6 +1157,7 @@ struct bus_type platform_bus_type = {
.dev_groups = platform_dev_groups,
.match = platform_match,
.uevent = platform_uevent,
+ .dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
.force_dma = true,
};
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index 88a3558..a9ec99d 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -314,6 +314,13 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
return strcmp(dev_name(dev), drv->name) == 0;
}

+static int host1x_dma_configure(struct device *dev)
+{
+ if (dev->of_node)
+ return of_dma_configure(dev, dev->of_node);
+ return 0;
+}
+
static const struct dev_pm_ops host1x_device_pm_ops = {
.suspend = pm_generic_suspend,
.resume = pm_generic_resume,
@@ -326,6 +333,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
struct bus_type host1x_bus_type = {
.name = "host1x",
.match = host1x_device_match,
+ .dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
.force_dma = true,
};
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index 3bed6be..78e507f 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -18,6 +18,8 @@
#include <linux/pm_runtime.h>
#include <linux/suspend.h>
#include <linux/kexec.h>
+#include <linux/of_device.h>
+#include <linux/acpi.h>
#include "pci.h"

struct pci_dynid {
@@ -1522,6 +1524,35 @@ static int pci_bus_num_vf(struct device *dev)
return pci_num_vf(to_pci_dev(dev));
}

+/**
+ * pci_dma_configure - Setup DMA configuration
+ * @dev: ptr to dev structure
+ *
+ * Function to update PCI devices's DMA configuration using the same
+ * info from the OF node or ACPI node of host bridge's parent (if any).
+ */
+static int pci_dma_configure(struct device *dev)
+{
+ struct device *bridge;
+ int ret = 0;
+
+ bridge = pci_get_host_bridge_device(to_pci_dev(dev));
+
+ if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
+ bridge->parent->of_node) {
+ ret = of_dma_configure(dev, bridge->parent->of_node);
+ } else if (has_acpi_companion(bridge)) {
+ struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
+ enum dev_dma_attr attr = acpi_get_dma_attr(adev);
+
+ if (attr != DEV_DMA_NOT_SUPPORTED)
+ ret = acpi_dma_configure(dev, attr);
+ }
+
+ pci_put_host_bridge_device(bridge);
+ return ret;
+}
+
struct bus_type pci_bus_type = {
.name = "pci",
.match = pci_bus_match,
@@ -1534,6 +1565,7 @@ struct bus_type pci_bus_type = {
.drv_groups = pci_drv_groups,
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
+ .dma_configure = pci_dma_configure,
.force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);
diff --git a/include/linux/device.h b/include/linux/device.h
index b093405..1832d90 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type *,
* @resume: Called to bring a device on this bus out of sleep mode.
* @num_vf: Called to find out how many virtual functions a device on this
* bus supports.
+ * @dma_configure: Called to setup DMA configuration on a device on
+ this bus.
* @pm: Power management operations of this bus, callback the specific
* device driver's pm-ops.
* @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
@@ -130,6 +132,8 @@ struct bus_type {

int (*num_vf)(struct device *dev);

+ int (*dma_configure)(struct device *dev);
+
const struct dev_pm_ops *pm;

const struct iommu_ops *iommu_ops;
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index 49f634d..3097c94 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -356,6 +356,8 @@ static inline char *early_platform_driver_setup_func(void) \
#define platform_pm_restore NULL
#endif

+extern int platform_dma_configure(struct device *dev);
+
#ifdef CONFIG_PM_SLEEP
#define USE_PLATFORM_PM_SLEEP_OPS \
.suspend = platform_pm_suspend, \
--
1.9.1

2018-03-30 09:07:02

[permalink] [raw]

Subject: Re: [PATCH] dma-mapping: move dma configuration to bus infrastructure

On Fri, Mar 30, 2018 at 09:15:30AM +0200, Christoph Hellwig wrote:
> Can you resend the current state of affairs so we can get it in for
> 4.17?

Changes to the driver core and 3 different busses 2 days before 4.16 is
out, preventing any testing at all in linux-next? This needs to wait
for the next merge window, sorry. I'm away all this weekend, and can't
test on my end, sorry.

greg k-h

2018-04-09 20:35:44

by Rob Herring (Arm)

[permalink] [raw]

Subject: Re: [PATCH v3 2/2] drivers: remove force dma flag from buses

2018-04-10 19:24:23

by Bjorn Helgaas

[permalink] [raw]

Subject: Re: [PATCH v3 1/2] dma-mapping: move dma configuration to bus infrastructure

On Fri, Mar 30, 2018 at 01:24:44PM +0530, Nipun Gupta wrote:
> It is bus specific aspect to map a given device on the bus and
> relevant firmware description of its DMA configuration.
> So, this change introduces 'dma_configure' as bus callback
> giving flexibility to busses for implementing its own dma
> configuration function.
>
> The change eases the addition of new busses w.r.t. adding the dma
> configuration functionality.
>
> This patch also updates the PCI, Platform, ACPI and host1x bus to
> use new introduced callbacks.

s/dma/DMA/ consistently above.

> Suggested-by: Christoph Hellwig <[email protected]>
> Signed-off-by: Nipun Gupta <[email protected]>
> Reviewed-by: Greg Kroah-Hartman <[email protected]>

Acked-by: Bjorn Helgaas <[email protected]> # PCI parts

I assume you'll merge this via some non-PCI tree. Let me know if you
need anything else from me.

> ---
> - The patches are based on the comments on:
> https://patchwork.kernel.org/patch/10259087/
>
> Changes in v2:
> - Do not have dma_deconfigure callback
> - Have '/dma_common_configure/' API to provide a common DMA
> configuration which can be used by busses if it suits them.
> - Platform and ACPI bus to use '/dma_common_configure/' in
> '/dma_configure/' callback.
> - Updated commit message
> - Updated pci_dma_configure API with changes suggested by Robin
>
> Changes in v3
> - Move dma_common_configure() within platform_dma_configure() and
> reuse platofrm_dma_configure() for AMBA bus too
> - Declare 'attr' in pci_dma_configure() inside the else statement
> where it is used.
>
> drivers/amba/bus.c | 4 ++++
> drivers/base/dma-mapping.c | 31 ++++---------------------------
> drivers/base/platform.c | 17 +++++++++++++++++
> drivers/gpu/host1x/bus.c | 8 ++++++++
> drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
> include/linux/device.h | 4 ++++
> include/linux/platform_device.h | 2 ++
> 7 files changed, 71 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
> index 594c228..867dc2b 100644
> --- a/drivers/amba/bus.c
> +++ b/drivers/amba/bus.c
> @@ -20,6 +20,7 @@
> #include <linux/sizes.h>
> #include <linux/limits.h>
> #include <linux/clk/clk-conf.h>
> +#include <linux/platform_device.h>
>
> #include <asm/irq.h>
>
> @@ -188,12 +189,15 @@ static int amba_pm_runtime_resume(struct device *dev)
> /*
> * Primecells are part of the Advanced Microcontroller Bus Architecture,
> * so we call the bus "amba".
> + * DMA configuration for platform and AMBA bus is same. So here we reuse
> + * platform's DMA config routine.
> */
> struct bus_type amba_bustype = {
> .name = "amba",
> .dev_groups = amba_dev_groups,
> .match = amba_match,
> .uevent = amba_uevent,
> + .dma_configure = platform_dma_configure,
> .pm = &amba_pm,
> .force_dma = true,
> };
> diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
> index 3b11835..fdc1502 100644
> --- a/drivers/base/dma-mapping.c
> +++ b/drivers/base/dma-mapping.c
> @@ -331,36 +331,13 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
> #endif
>
> /*
> - * Common configuration to enable DMA API use for a device
> + * enables DMA API use for a device
> */
> -#include <linux/pci.h>
> -
> int dma_configure(struct device *dev)
> {
> - struct device *bridge = NULL, *dma_dev = dev;
> - enum dev_dma_attr attr;
> - int ret = 0;
> -
> - if (dev_is_pci(dev)) {
> - bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> - dma_dev = bridge;
> - if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
> - dma_dev->parent->of_node)
> - dma_dev = dma_dev->parent;
> - }
> -
> - if (dma_dev->of_node) {
> - ret = of_dma_configure(dev, dma_dev->of_node);
> - } else if (has_acpi_companion(dma_dev)) {
> - attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
> - if (attr != DEV_DMA_NOT_SUPPORTED)
> - ret = acpi_dma_configure(dev, attr);
> - }
> -
> - if (bridge)
> - pci_put_host_bridge_device(bridge);
> -
> - return ret;
> + if (dev->bus->dma_configure)
> + return dev->bus->dma_configure(dev);
> + return 0;
> }
>
> void dma_deconfigure(struct device *dev)
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index f1bf7b3..72fdbf6 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -1130,6 +1130,22 @@ int platform_pm_restore(struct device *dev)
>
> #endif /* CONFIG_HIBERNATE_CALLBACKS */
>
> +int platform_dma_configure(struct device *dev)
> +{
> + enum dev_dma_attr attr;
> + int ret = 0;
> +
> + if (dev->of_node) {
> + ret = of_dma_configure(dev, dev->of_node);
> + } else if (has_acpi_companion(dev)) {
> + attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + return ret;
> +}
> +
> static const struct dev_pm_ops platform_dev_pm_ops = {
> .runtime_suspend = pm_generic_runtime_suspend,
> .runtime_resume = pm_generic_runtime_resume,
> @@ -1141,6 +1157,7 @@ struct bus_type platform_bus_type = {
> .dev_groups = platform_dev_groups,
> .match = platform_match,
> .uevent = platform_uevent,
> + .dma_configure = platform_dma_configure,
> .pm = &platform_dev_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index 88a3558..a9ec99d 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -314,6 +314,13 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> return strcmp(dev_name(dev), drv->name) == 0;
> }
>
> +static int host1x_dma_configure(struct device *dev)
> +{
> + if (dev->of_node)
> + return of_dma_configure(dev, dev->of_node);
> + return 0;
> +}
> +
> static const struct dev_pm_ops host1x_device_pm_ops = {
> .suspend = pm_generic_suspend,
> .resume = pm_generic_resume,
> @@ -326,6 +333,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> struct bus_type host1x_bus_type = {
> .name = "host1x",
> .match = host1x_device_match,
> + .dma_configure = host1x_dma_configure,
> .pm = &host1x_device_pm_ops,
> .force_dma = true,
> };
> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> index 3bed6be..78e507f 100644
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -18,6 +18,8 @@
> #include <linux/pm_runtime.h>
> #include <linux/suspend.h>
> #include <linux/kexec.h>
> +#include <linux/of_device.h>
> +#include <linux/acpi.h>
> #include "pci.h"
>
> struct pci_dynid {
> @@ -1522,6 +1524,35 @@ static int pci_bus_num_vf(struct device *dev)
> return pci_num_vf(to_pci_dev(dev));
> }
>
> +/**
> + * pci_dma_configure - Setup DMA configuration
> + * @dev: ptr to dev structure
> + *
> + * Function to update PCI devices's DMA configuration using the same
> + * info from the OF node or ACPI node of host bridge's parent (if any).
> + */
> +static int pci_dma_configure(struct device *dev)
> +{
> + struct device *bridge;
> + int ret = 0;
> +
> + bridge = pci_get_host_bridge_device(to_pci_dev(dev));
> +
> + if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
> + bridge->parent->of_node) {
> + ret = of_dma_configure(dev, bridge->parent->of_node);
> + } else if (has_acpi_companion(bridge)) {
> + struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
> + enum dev_dma_attr attr = acpi_get_dma_attr(adev);
> +
> + if (attr != DEV_DMA_NOT_SUPPORTED)
> + ret = acpi_dma_configure(dev, attr);
> + }
> +
> + pci_put_host_bridge_device(bridge);
> + return ret;
> +}
> +
> struct bus_type pci_bus_type = {
> .name = "pci",
> .match = pci_bus_match,
> @@ -1534,6 +1565,7 @@ struct bus_type pci_bus_type = {
> .drv_groups = pci_drv_groups,
> .pm = PCI_PM_OPS_PTR,
> .num_vf = pci_bus_num_vf,
> + .dma_configure = pci_dma_configure,
> .force_dma = true,
> };
> EXPORT_SYMBOL(pci_bus_type);
> diff --git a/include/linux/device.h b/include/linux/device.h
> index b093405..1832d90 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type *,
> * @resume: Called to bring a device on this bus out of sleep mode.
> * @num_vf: Called to find out how many virtual functions a device on this
> * bus supports.
> + * @dma_configure: Called to setup DMA configuration on a device on
> + this bus.
> * @pm: Power management operations of this bus, callback the specific
> * device driver's pm-ops.
> * @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
> @@ -130,6 +132,8 @@ struct bus_type {
>
> int (*num_vf)(struct device *dev);
>
> + int (*dma_configure)(struct device *dev);
> +
> const struct dev_pm_ops *pm;
>
> const struct iommu_ops *iommu_ops;
> diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
> index 49f634d..3097c94 100644
> --- a/include/linux/platform_device.h
> +++ b/include/linux/platform_device.h
> @@ -356,6 +356,8 @@ static inline char *early_platform_driver_setup_func(void) \
> #define platform_pm_restore NULL
> #endif
>
> +extern int platform_dma_configure(struct device *dev);
> +
> #ifdef CONFIG_PM_SLEEP
> #define USE_PLATFORM_PM_SLEEP_OPS \
> .suspend = platform_pm_suspend, \
> --
> 1.9.1
>

2018-04-10 19:26:05

by Bjorn Helgaas

[permalink] [raw]

Subject: Re: [PATCH v3 2/2] drivers: remove force dma flag from buses

On Fri, Mar 30, 2018 at 01:24:45PM +0530, Nipun Gupta wrote:
> With each bus implementing its own DMA configuration callback,
> there is no need for bus to explicitly have force_dma in its
> global structure. This patch modifies of_dma_configure API to
> accept an input parameter which specifies if implicit DMA
> configuration is required even when it is not described by the
> firmware.
>
> Signed-off-by: Nipun Gupta <[email protected]>

Acked-by: Bjorn Helgaas <[email protected]> # PCI parts

> ---
> Changes in v2:
> - This is a new change suggested by Robin and Christoph
> and is added to the series.
>
> Changes in v3:
> - Rebase and changes corresponding to the changes in patch 1/2
>
> drivers/amba/bus.c | 1 -
> drivers/base/platform.c | 3 +--
> drivers/bcma/main.c | 2 +-
> drivers/dma/qcom/hidma_mgmt.c | 2 +-
> drivers/gpu/host1x/bus.c | 5 ++---
> drivers/of/device.c | 6 ++++--
> drivers/of/of_reserved_mem.c | 2 +-
> drivers/pci/pci-driver.c | 3 +--
> include/linux/device.h | 4 ----
> include/linux/of_device.h | 8 ++++++--
> 10 files changed, 17 insertions(+), 19 deletions(-)
>
> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
> index 867dc2b..abe73c4 100644
> --- a/drivers/amba/bus.c
> +++ b/drivers/amba/bus.c
> @@ -199,7 +199,6 @@ struct bus_type amba_bustype = {
> .uevent = amba_uevent,
> .dma_configure = platform_dma_configure,
> .pm = &amba_pm,
> - .force_dma = true,
> };
>
> static int __init amba_init(void)
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index 72fdbf6..cfbc569 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -1136,7 +1136,7 @@ int platform_dma_configure(struct device *dev)
> int ret = 0;
>
> if (dev->of_node) {
> - ret = of_dma_configure(dev, dev->of_node);
> + ret = of_dma_configure(dev, dev->of_node, true);
> } else if (has_acpi_companion(dev)) {
> attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
> if (attr != DEV_DMA_NOT_SUPPORTED)
> @@ -1159,7 +1159,6 @@ struct bus_type platform_bus_type = {
> .uevent = platform_uevent,
> .dma_configure = platform_dma_configure,
> .pm = &platform_dev_pm_ops,
> - .force_dma = true,
> };
> EXPORT_SYMBOL_GPL(platform_bus_type);
>
> diff --git a/drivers/bcma/main.c b/drivers/bcma/main.c
> index e6986c7..fc1f4ac 100644
> --- a/drivers/bcma/main.c
> +++ b/drivers/bcma/main.c
> @@ -207,7 +207,7 @@ static void bcma_of_fill_device(struct device *parent,
>
> core->irq = bcma_of_get_irq(parent, core, 0);
>
> - of_dma_configure(&core->dev, node);
> + of_dma_configure(&core->dev, node, false);
> }
>
> unsigned int bcma_core_irq(struct bcma_device *core, int num)
> diff --git a/drivers/dma/qcom/hidma_mgmt.c b/drivers/dma/qcom/hidma_mgmt.c
> index 000c7019..d64edeb 100644
> --- a/drivers/dma/qcom/hidma_mgmt.c
> +++ b/drivers/dma/qcom/hidma_mgmt.c
> @@ -398,7 +398,7 @@ static int __init hidma_mgmt_of_populate_channels(struct device_node *np)
> }
> of_node_get(child);
> new_pdev->dev.of_node = child;
> - of_dma_configure(&new_pdev->dev, child);
> + of_dma_configure(&new_pdev->dev, child, true);
> /*
> * It is assumed that calling of_msi_configure is safe on
> * platforms with or without MSI support.
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index a9ec99d..b39c1e9 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -317,7 +317,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> static int host1x_dma_configure(struct device *dev)
> {
> if (dev->of_node)
> - return of_dma_configure(dev, dev->of_node);
> + return of_dma_configure(dev, dev->of_node, true);
> return 0;
> }
>
> @@ -335,7 +335,6 @@ struct bus_type host1x_bus_type = {
> .match = host1x_device_match,
> .dma_configure = host1x_dma_configure,
> .pm = &host1x_device_pm_ops,
> - .force_dma = true,
> };
>
> static void __host1x_device_del(struct host1x_device *device)
> @@ -424,7 +423,7 @@ static int host1x_device_add(struct host1x *host1x,
> device->dev.bus = &host1x_bus_type;
> device->dev.parent = host1x->dev;
>
> - of_dma_configure(&device->dev, host1x->dev->of_node);
> + of_dma_configure(&device->dev, host1x->dev->of_node, true);
>
> err = host1x_device_parse_dt(device, driver);
> if (err < 0) {
> diff --git a/drivers/of/device.c b/drivers/of/device.c
> index 064c818..33d8551 100644
> --- a/drivers/of/device.c
> +++ b/drivers/of/device.c
> @@ -76,6 +76,8 @@ int of_device_add(struct platform_device *ofdev)
> * of_dma_configure - Setup DMA configuration
> * @dev: Device to apply DMA configuration
> * @np: Pointer to OF node having DMA configuration
> + * @force_dma: Whether device is to be set up by of_dma_configure() even if
> + * DMA capability is not explicitly described by firmware.
> *
> * Try to get devices's DMA configuration from DT and update it
> * accordingly.
> @@ -84,7 +86,7 @@ int of_device_add(struct platform_device *ofdev)
> * can use a platform bus notifier and handle BUS_NOTIFY_ADD_DEVICE events
> * to fix up DMA configuration.
> */
> -int of_dma_configure(struct device *dev, struct device_node *np)
> +int of_dma_configure(struct device *dev, struct device_node *np, bool force_dma)
> {
> u64 dma_addr, paddr, size = 0;
> int ret;
> @@ -100,7 +102,7 @@ int of_dma_configure(struct device *dev, struct device_node *np)
> * DMA configuration regardless of whether "dma-ranges" is
> * correctly specified or not.
> */
> - if (!dev->bus->force_dma)
> + if (!force_dma)
> return ret == -ENODEV ? 0 : ret;
>
> dma_addr = offset = 0;
> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> index 9a4f4246..895c83e 100644
> --- a/drivers/of/of_reserved_mem.c
> +++ b/drivers/of/of_reserved_mem.c
> @@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device *dev,
> /* ensure that dma_ops is set for virtual devices
> * using reserved memory
> */
> - of_dma_configure(dev, np);
> + of_dma_configure(dev, np, true);
>
> dev_info(dev, "assigned reserved memory node %s\n", rmem->name);
> } else {
> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> index 78e507f..27c832f 100644
> --- a/drivers/pci/pci-driver.c
> +++ b/drivers/pci/pci-driver.c
> @@ -1540,7 +1540,7 @@ static int pci_dma_configure(struct device *dev)
>
> if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
> bridge->parent->of_node) {
> - ret = of_dma_configure(dev, bridge->parent->of_node);
> + ret = of_dma_configure(dev, bridge->parent->of_node, true);
> } else if (has_acpi_companion(bridge)) {
> struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
> enum dev_dma_attr attr = acpi_get_dma_attr(adev);
> @@ -1566,7 +1566,6 @@ struct bus_type pci_bus_type = {
> .pm = PCI_PM_OPS_PTR,
> .num_vf = pci_bus_num_vf,
> .dma_configure = pci_dma_configure,
> - .force_dma = true,
> };
> EXPORT_SYMBOL(pci_bus_type);
>
> diff --git a/include/linux/device.h b/include/linux/device.h
> index 1832d90..92c530e 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -98,8 +98,6 @@ extern int __must_check bus_create_file(struct bus_type *,
> * @p: The private data of the driver core, only the driver core can
> * touch this.
> * @lock_key: Lock class key for use by the lock validator
> - * @force_dma: Assume devices on this bus should be set up by dma_configure()
> - * even if DMA capability is not explicitly described by firmware.
> *
> * A bus is a channel between the processor and one or more devices. For the
> * purposes of the device model, all devices are connected via a bus, even if
> @@ -140,8 +138,6 @@ struct bus_type {
>
> struct subsys_private *p;
> struct lock_class_key lock_key;
> -
> - bool force_dma;
> };
>
> extern int __must_check bus_register(struct bus_type *bus);
> diff --git a/include/linux/of_device.h b/include/linux/of_device.h
> index 8da5a1b..165fd30 100644
> --- a/include/linux/of_device.h
> +++ b/include/linux/of_device.h
> @@ -55,7 +55,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
> return of_node_get(cpu_dev->of_node);
> }
>
> -int of_dma_configure(struct device *dev, struct device_node *np);
> +int of_dma_configure(struct device *dev,
> + struct device_node *np,
> + bool force_dma);
> void of_dma_deconfigure(struct device *dev);
> #else /* CONFIG_OF */
>
> @@ -105,7 +107,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
> return NULL;
> }
>
> -static inline int of_dma_configure(struct device *dev, struct device_node *np)
> +static inline int of_dma_configure(struct device *dev,
> + struct device_node *np,
> + bool force_dma)
> {
> return 0;
> }
> --
> 1.9.1
>

2018-04-23 12:57:02

[permalink] [raw]

Subject: Re: [PATCH v3 1/2] dma-mapping: move dma configuration to bus infrastructure

On Tue, Apr 10, 2018 at 02:20:41PM -0500, Bjorn Helgaas wrote:
> I assume you'll merge this via some non-PCI tree. Let me know if you
> need anything else from me.

I'll take it through the dma-mapping tree, so you should be fine.

2018-04-23 13:00:07

[permalink] [raw]

Subject: Re: [PATCH v3 1/2] dma-mapping: move dma configuration to bus infrastructure

Can you resend your changes against Linux 4.17-rc2? There are a lot
of conflicts as-is.

2018-04-27 17:12:30

[permalink] [raw]

Subject: RE: [PATCH v3 1/2] dma-mapping: move dma configuration to bus infrastructure

2018-04-28 02:54:19

[permalink] [raw]

Subject: [PATCH v4 2/2] drivers: remove force dma flag from buses

With each bus implementing its own DMA configuration callback,
there is no need for bus to explicitly have force_dma in its
global structure. This patch modifies of_dma_configure API to
accept an input parameter which specifies if implicit DMA
configuration is required even when it is not described by the
firmware.

Signed-off-by: Nipun Gupta <[email protected]>
Acked-by: Bjorn Helgaas <[email protected]> # PCI parts
---
Changes in v2:
- This is a new change suggested by Robin and Christoph
and is added to the series.

Changes in v3:
- Rebase and changes corresponding to the changes in patch 1/2

Changes in v4:
- Rebased on top of 4.17-rc2

drivers/amba/bus.c | 1 -
drivers/base/platform.c | 3 +--
drivers/bcma/main.c | 2 +-
drivers/dma/qcom/hidma_mgmt.c | 2 +-
drivers/gpu/host1x/bus.c | 5 ++---
drivers/of/device.c | 6 ++++--
drivers/of/of_reserved_mem.c | 2 +-
drivers/pci/pci-driver.c | 3 +--
include/linux/device.h | 4 ----
include/linux/of_device.h | 8 ++++++--
10 files changed, 17 insertions(+), 19 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 867dc2b..abe73c4 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -199,7 +199,6 @@ struct bus_type amba_bustype = {
.uevent = amba_uevent,
.dma_configure = platform_dma_configure,
.pm = &amba_pm,
- .force_dma = true,
};

static int __init amba_init(void)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 638d42e..c0ff1e7 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1136,7 +1136,7 @@ int platform_dma_configure(struct device *dev)
int ret = 0;

if (dev->of_node) {
- ret = of_dma_configure(dev, dev->of_node);
+ ret = of_dma_configure(dev, dev->of_node, true);
} else if (has_acpi_companion(dev)) {
attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
if (attr != DEV_DMA_NOT_SUPPORTED)
@@ -1159,7 +1159,6 @@ struct bus_type platform_bus_type = {
.uevent = platform_uevent,
.dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
- .force_dma = true,
};
EXPORT_SYMBOL_GPL(platform_bus_type);

diff --git a/drivers/bcma/main.c b/drivers/bcma/main.c
index e6986c7..fc1f4ac 100644
--- a/drivers/bcma/main.c
+++ b/drivers/bcma/main.c
@@ -207,7 +207,7 @@ static void bcma_of_fill_device(struct device *parent,

core->irq = bcma_of_get_irq(parent, core, 0);

- of_dma_configure(&core->dev, node);
+ of_dma_configure(&core->dev, node, false);
}

unsigned int bcma_core_irq(struct bcma_device *core, int num)
diff --git a/drivers/dma/qcom/hidma_mgmt.c b/drivers/dma/qcom/hidma_mgmt.c
index 000c7019..d64edeb 100644
--- a/drivers/dma/qcom/hidma_mgmt.c
+++ b/drivers/dma/qcom/hidma_mgmt.c
@@ -398,7 +398,7 @@ static int __init hidma_mgmt_of_populate_channels(struct device_node *np)
}
of_node_get(child);
new_pdev->dev.of_node = child;
- of_dma_configure(&new_pdev->dev, child);
+ of_dma_configure(&new_pdev->dev, child, true);
/*
* It is assumed that calling of_msi_configure is safe on
* platforms with or without MSI support.
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index a9ec99d..b39c1e9 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -317,7 +317,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
static int host1x_dma_configure(struct device *dev)
{
if (dev->of_node)
- return of_dma_configure(dev, dev->of_node);
+ return of_dma_configure(dev, dev->of_node, true);
return 0;
}

@@ -335,7 +335,6 @@ struct bus_type host1x_bus_type = {
.match = host1x_device_match,
.dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
- .force_dma = true,
};

static void __host1x_device_del(struct host1x_device *device)
@@ -424,7 +423,7 @@ static int host1x_device_add(struct host1x *host1x,
device->dev.bus = &host1x_bus_type;
device->dev.parent = host1x->dev;

- of_dma_configure(&device->dev, host1x->dev->of_node);
+ of_dma_configure(&device->dev, host1x->dev->of_node, true);

err = host1x_device_parse_dt(device, driver);
if (err < 0) {
diff --git a/drivers/of/device.c b/drivers/of/device.c
index 064c818..33d8551 100644
--- a/drivers/of/device.c
+++ b/drivers/of/device.c
@@ -76,6 +76,8 @@ int of_device_add(struct platform_device *ofdev)
* of_dma_configure - Setup DMA configuration
* @dev: Device to apply DMA configuration
* @np: Pointer to OF node having DMA configuration
+ * @force_dma: Whether device is to be set up by of_dma_configure() even if
+ * DMA capability is not explicitly described by firmware.
*
* Try to get devices's DMA configuration from DT and update it
* accordingly.
@@ -84,7 +86,7 @@ int of_device_add(struct platform_device *ofdev)
* can use a platform bus notifier and handle BUS_NOTIFY_ADD_DEVICE events
* to fix up DMA configuration.
*/
-int of_dma_configure(struct device *dev, struct device_node *np)
+int of_dma_configure(struct device *dev, struct device_node *np, bool force_dma)
{
u64 dma_addr, paddr, size = 0;
int ret;
@@ -100,7 +102,7 @@ int of_dma_configure(struct device *dev, struct device_node *np)
* DMA configuration regardless of whether "dma-ranges" is
* correctly specified or not.
*/
- if (!dev->bus->force_dma)
+ if (!force_dma)
return ret == -ENODEV ? 0 : ret;

dma_addr = offset = 0;
diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
index 9a4f4246..895c83e 100644
--- a/drivers/of/of_reserved_mem.c
+++ b/drivers/of/of_reserved_mem.c
@@ -353,7 +353,7 @@ int of_reserved_mem_device_init_by_idx(struct device *dev,
/* ensure that dma_ops is set for virtual devices
* using reserved memory
*/
- of_dma_configure(dev, np);
+ of_dma_configure(dev, np, true);

dev_info(dev, "assigned reserved memory node %s\n", rmem->name);
} else {
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index ba8c6b4..f8269a7 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -1595,7 +1595,7 @@ static int pci_dma_configure(struct device *dev)

if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
bridge->parent->of_node) {
- ret = of_dma_configure(dev, bridge->parent->of_node);
+ ret = of_dma_configure(dev, bridge->parent->of_node, true);
} else if (has_acpi_companion(bridge)) {
struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
enum dev_dma_attr attr = acpi_get_dma_attr(adev);
@@ -1621,7 +1621,6 @@ struct bus_type pci_bus_type = {
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
.dma_configure = pci_dma_configure,
- .force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);

diff --git a/include/linux/device.h b/include/linux/device.h
index 607c7eb..31dcb29 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -98,8 +98,6 @@ extern int __must_check bus_create_file(struct bus_type *,
* @p: The private data of the driver core, only the driver core can
* touch this.
* @lock_key: Lock class key for use by the lock validator
- * @force_dma: Assume devices on this bus should be set up by dma_configure()
- * even if DMA capability is not explicitly described by firmware.
*
* A bus is a channel between the processor and one or more devices. For the
* purposes of the device model, all devices are connected via a bus, even if
@@ -140,8 +138,6 @@ struct bus_type {

struct subsys_private *p;
struct lock_class_key lock_key;
-
- bool force_dma;
};

extern int __must_check bus_register(struct bus_type *bus);
diff --git a/include/linux/of_device.h b/include/linux/of_device.h
index 8da5a1b..165fd30 100644
--- a/include/linux/of_device.h
+++ b/include/linux/of_device.h
@@ -55,7 +55,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
return of_node_get(cpu_dev->of_node);
}

-int of_dma_configure(struct device *dev, struct device_node *np);
+int of_dma_configure(struct device *dev,
+ struct device_node *np,
+ bool force_dma);
void of_dma_deconfigure(struct device *dev);
#else /* CONFIG_OF */

@@ -105,7 +107,9 @@ static inline struct device_node *of_cpu_device_node_get(int cpu)
return NULL;
}

-static inline int of_dma_configure(struct device *dev, struct device_node *np)
+static inline int of_dma_configure(struct device *dev,
+ struct device_node *np,
+ bool force_dma)
{
return 0;
}
--
1.9.1

2018-04-28 02:54:38

[permalink] [raw]

Subject: [PATCH v4 1/2] dma-mapping: move dma configuration to bus infrastructure

It is bus specific aspect to map a given device on the bus and
relevant firmware description of its DMA configuration.
So, this change introduces 'dma_configure' as bus callback
giving flexibility to busses for implementing its own dma
configuration function.

The change eases the addition of new busses w.r.t. adding the dma
configuration functionality.

This patch also updates the PCI, Platform, ACPI and host1x bus to
use new introduced callbacks.

Suggested-by: Christoph Hellwig <[email protected]>
Signed-off-by: Nipun Gupta <[email protected]>
Reviewed-by: Greg Kroah-Hartman <[email protected]>
Acked-by: Bjorn Helgaas <[email protected]> # PCI parts
---
- The patches are based on the comments on:
https://patchwork.kernel.org/patch/10259087/

Changes in v2:
- Do not have dma_deconfigure callback
- Have '/dma_common_configure/' API to provide a common DMA
configuration which can be used by busses if it suits them.
- Platform and ACPI bus to use '/dma_common_configure/' in
'/dma_configure/' callback.
- Updated commit message
- Updated pci_dma_configure API with changes suggested by Robin

Changes in v3:
- Move dma_common_configure() within platform_dma_configure() and
reuse platofrm_dma_configure() for AMBA bus too
- Declare 'attr' in pci_dma_configure() inside the else statement
where it is used.

Changes in v4:
- Rebased on top of 4.17-rc2

drivers/amba/bus.c | 4 ++++
drivers/base/dma-mapping.c | 31 ++++---------------------------
drivers/base/platform.c | 17 +++++++++++++++++
drivers/gpu/host1x/bus.c | 8 ++++++++
drivers/pci/pci-driver.c | 32 ++++++++++++++++++++++++++++++++
include/linux/device.h | 4 ++++
include/linux/platform_device.h | 2 ++
7 files changed, 71 insertions(+), 27 deletions(-)

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index 594c228..867dc2b 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -20,6 +20,7 @@
#include <linux/sizes.h>
#include <linux/limits.h>
#include <linux/clk/clk-conf.h>
+#include <linux/platform_device.h>

#include <asm/irq.h>

@@ -188,12 +189,15 @@ static int amba_pm_runtime_resume(struct device *dev)
/*
* Primecells are part of the Advanced Microcontroller Bus Architecture,
* so we call the bus "amba".
+ * DMA configuration for platform and AMBA bus is same. So here we reuse
+ * platform's DMA config routine.
*/
struct bus_type amba_bustype = {
.name = "amba",
.dev_groups = amba_dev_groups,
.match = amba_match,
.uevent = amba_uevent,
+ .dma_configure = platform_dma_configure,
.pm = &amba_pm,
.force_dma = true,
};
diff --git a/drivers/base/dma-mapping.c b/drivers/base/dma-mapping.c
index d82566d..f831a58 100644
--- a/drivers/base/dma-mapping.c
+++ b/drivers/base/dma-mapping.c
@@ -329,36 +329,13 @@ void dma_common_free_remap(void *cpu_addr, size_t size, unsigned long vm_flags)
#endif

/*
- * Common configuration to enable DMA API use for a device
+ * enables DMA API use for a device
*/
-#include <linux/pci.h>
-
int dma_configure(struct device *dev)
{
- struct device *bridge = NULL, *dma_dev = dev;
- enum dev_dma_attr attr;
- int ret = 0;
-
- if (dev_is_pci(dev)) {
- bridge = pci_get_host_bridge_device(to_pci_dev(dev));
- dma_dev = bridge;
- if (IS_ENABLED(CONFIG_OF) && dma_dev->parent &&
- dma_dev->parent->of_node)
- dma_dev = dma_dev->parent;
- }
-
- if (dma_dev->of_node) {
- ret = of_dma_configure(dev, dma_dev->of_node);
- } else if (has_acpi_companion(dma_dev)) {
- attr = acpi_get_dma_attr(to_acpi_device_node(dma_dev->fwnode));
- if (attr != DEV_DMA_NOT_SUPPORTED)
- ret = acpi_dma_configure(dev, attr);
- }
-
- if (bridge)
- pci_put_host_bridge_device(bridge);
-
- return ret;
+ if (dev->bus->dma_configure)
+ return dev->bus->dma_configure(dev);
+ return 0;
}

void dma_deconfigure(struct device *dev)
diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 8075ddc..638d42e 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -1130,6 +1130,22 @@ int platform_pm_restore(struct device *dev)

#endif /* CONFIG_HIBERNATE_CALLBACKS */

+int platform_dma_configure(struct device *dev)
+{
+ enum dev_dma_attr attr;
+ int ret = 0;
+
+ if (dev->of_node) {
+ ret = of_dma_configure(dev, dev->of_node);
+ } else if (has_acpi_companion(dev)) {
+ attr = acpi_get_dma_attr(to_acpi_device_node(dev->fwnode));
+ if (attr != DEV_DMA_NOT_SUPPORTED)
+ ret = acpi_dma_configure(dev, attr);
+ }
+
+ return ret;
+}
+
static const struct dev_pm_ops platform_dev_pm_ops = {
.runtime_suspend = pm_generic_runtime_suspend,
.runtime_resume = pm_generic_runtime_resume,
@@ -1141,6 +1157,7 @@ struct bus_type platform_bus_type = {
.dev_groups = platform_dev_groups,
.match = platform_match,
.uevent = platform_uevent,
+ .dma_configure = platform_dma_configure,
.pm = &platform_dev_pm_ops,
.force_dma = true,
};
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
index 88a3558..a9ec99d 100644
--- a/drivers/gpu/host1x/bus.c
+++ b/drivers/gpu/host1x/bus.c
@@ -314,6 +314,13 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
return strcmp(dev_name(dev), drv->name) == 0;
}

+static int host1x_dma_configure(struct device *dev)
+{
+ if (dev->of_node)
+ return of_dma_configure(dev, dev->of_node);
+ return 0;
+}
+
static const struct dev_pm_ops host1x_device_pm_ops = {
.suspend = pm_generic_suspend,
.resume = pm_generic_resume,
@@ -326,6 +333,7 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
struct bus_type host1x_bus_type = {
.name = "host1x",
.match = host1x_device_match,
+ .dma_configure = host1x_dma_configure,
.pm = &host1x_device_pm_ops,
.force_dma = true,
};
diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
index b9a1311..ba8c6b4 100644
--- a/drivers/pci/pci-driver.c
+++ b/drivers/pci/pci-driver.c
@@ -16,6 +16,8 @@
#include <linux/pm_runtime.h>
#include <linux/suspend.h>
#include <linux/kexec.h>
+#include <linux/of_device.h>
+#include <linux/acpi.h>
#include "pci.h"
#include "pcie/portdrv.h"

@@ -1577,6 +1579,35 @@ static int pci_bus_num_vf(struct device *dev)
return pci_num_vf(to_pci_dev(dev));
}

+/**
+ * pci_dma_configure - Setup DMA configuration
+ * @dev: ptr to dev structure
+ *
+ * Function to update PCI devices's DMA configuration using the same
+ * info from the OF node or ACPI node of host bridge's parent (if any).
+ */
+static int pci_dma_configure(struct device *dev)
+{
+ struct device *bridge;
+ int ret = 0;
+
+ bridge = pci_get_host_bridge_device(to_pci_dev(dev));
+
+ if (IS_ENABLED(CONFIG_OF) && bridge->parent &&
+ bridge->parent->of_node) {
+ ret = of_dma_configure(dev, bridge->parent->of_node);
+ } else if (has_acpi_companion(bridge)) {
+ struct acpi_device *adev = to_acpi_device_node(bridge->fwnode);
+ enum dev_dma_attr attr = acpi_get_dma_attr(adev);
+
+ if (attr != DEV_DMA_NOT_SUPPORTED)
+ ret = acpi_dma_configure(dev, attr);
+ }
+
+ pci_put_host_bridge_device(bridge);
+ return ret;
+}
+
struct bus_type pci_bus_type = {
.name = "pci",
.match = pci_bus_match,
@@ -1589,6 +1620,7 @@ struct bus_type pci_bus_type = {
.drv_groups = pci_drv_groups,
.pm = PCI_PM_OPS_PTR,
.num_vf = pci_bus_num_vf,
+ .dma_configure = pci_dma_configure,
.force_dma = true,
};
EXPORT_SYMBOL(pci_bus_type);
diff --git a/include/linux/device.h b/include/linux/device.h
index 0059b99..607c7eb 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -88,6 +88,8 @@ extern int __must_check bus_create_file(struct bus_type *,
* @resume: Called to bring a device on this bus out of sleep mode.
* @num_vf: Called to find out how many virtual functions a device on this
* bus supports.
+ * @dma_configure: Called to setup DMA configuration on a device on
+ this bus.
* @pm: Power management operations of this bus, callback the specific
* device driver's pm-ops.
* @iommu_ops: IOMMU specific operations for this bus, used to attach IOMMU
@@ -130,6 +132,8 @@ struct bus_type {

int (*num_vf)(struct device *dev);

+ int (*dma_configure)(struct device *dev);
+
const struct dev_pm_ops *pm;

const struct iommu_ops *iommu_ops;
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index 49f634d..3097c94 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -356,6 +356,8 @@ static inline char *early_platform_driver_setup_func(void) \
#define platform_pm_restore NULL
#endif

+extern int platform_dma_configure(struct device *dev);
+
#ifdef CONFIG_PM_SLEEP
#define USE_PLATFORM_PM_SLEEP_OPS \
.suspend = platform_pm_suspend, \
--
1.9.1

2018-04-30 10:42:01

by Thierry Reding

[permalink] [raw]

Subject: Re: [PATCH v4 1/2] dma-mapping: move dma configuration to bus infrastructure

On Sat, Apr 28, 2018 at 08:21:58AM +0530, Nipun Gupta wrote:
[...]
> diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c
> index 88a3558..a9ec99d 100644
> --- a/drivers/gpu/host1x/bus.c
> +++ b/drivers/gpu/host1x/bus.c
> @@ -314,6 +314,13 @@ static int host1x_device_match(struct device *dev, struct device_driver *drv)
> return strcmp(dev_name(dev), drv->name) == 0;
> }
>
> +static int host1x_dma_configure(struct device *dev)
> +{
> + if (dev->of_node)
> + return of_dma_configure(dev, dev->of_node);

The conditional here is somewhat pointless since the of_node should
always be set. If it weren't it should be considered a bug and this
function welcome to crash to make that very obvious.

Either way:

Acked-by: Thierry Reding <[email protected]>

Attachments:

(No filename) (795.00 B)
signature.asc (849.00 B)
Download all attachments

2018-05-01 12:35:56

by Rob Herring (Arm)

[permalink] [raw]

Subject: Re: [PATCH v4 2/2] drivers: remove force dma flag from buses

On Sat, Apr 28, 2018 at 08:21:59AM +0530, Nipun Gupta wrote:
> With each bus implementing its own DMA configuration callback,
> there is no need for bus to explicitly have force_dma in its
> global structure. This patch modifies of_dma_configure API to
> accept an input parameter which specifies if implicit DMA
> configuration is required even when it is not described by the
> firmware.
>
> Signed-off-by: Nipun Gupta <[email protected]>
> Acked-by: Bjorn Helgaas <[email protected]> # PCI parts
> ---
> Changes in v2:
> - This is a new change suggested by Robin and Christoph
> and is added to the series.
>
> Changes in v3:
> - Rebase and changes corresponding to the changes in patch 1/2
>
> Changes in v4:
> - Rebased on top of 4.17-rc2
>
> drivers/amba/bus.c | 1 -
> drivers/base/platform.c | 3 +--
> drivers/bcma/main.c | 2 +-
> drivers/dma/qcom/hidma_mgmt.c | 2 +-
> drivers/gpu/host1x/bus.c | 5 ++---
> drivers/of/device.c | 6 ++++--
> drivers/of/of_reserved_mem.c | 2 +-
> drivers/pci/pci-driver.c | 3 +--
> include/linux/device.h | 4 ----
> include/linux/of_device.h | 8 ++++++--
> 10 files changed, 17 insertions(+), 19 deletions(-)

Reviewed-by: Rob Herring <[email protected]>

2018-05-03 12:19:29

[permalink] [raw]

Subject: Re: [PATCH v4 1/2] dma-mapping: move dma configuration to bus infrastructure

This still fails to apply against either 4.17-rc3 or current Linus
master for me. Please resend against it, and mention the tree it
is against.

Also please resend it in a separate email thread (all your patches
were replies to the previous ones) and include a cover letter.

2018-05-03 12:19:49