diff mbox

staging: vc04_services: setup DMA and coherent mask

Message ID 20161028171159.23973-1-mzoran@crowfest.net (mailing list archive)
State New, archived
Headers show

Commit Message

Michael Zoran Oct. 28, 2016, 5:11 p.m. UTC
Setting the DMA mask is optional on 32 bit but
is mandatory on 64 bit.  Set the DMA mask and coherent
to force all DMA to be in the 32 bit address space.

This is considered a "good practice" and most drivers
already do this.

Signed-off-by: Michael Zoran <mzoran@crowfest.net>
---
 .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

Comments

Eric Anholt Oct. 31, 2016, 6:36 p.m. UTC | #1
Michael Zoran <mzoran@crowfest.net> writes:

> Setting the DMA mask is optional on 32 bit but
> is mandatory on 64 bit.  Set the DMA mask and coherent
> to force all DMA to be in the 32 bit address space.
>
> This is considered a "good practice" and most drivers
> already do this.
>
> Signed-off-by: Michael Zoran <mzoran@crowfest.net>
> ---
>  .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
> index a5afcc5..6fa2b5a 100644
> --- a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
> +++ b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
> @@ -97,6 +97,16 @@ int vchiq_platform_init(struct platform_device *pdev, VCHIQ_STATE_T *state)
>  	int slot_mem_size, frag_mem_size;
>  	int err, irq, i;
>  
> +	/*
> +	 * Setting the DMA mask is necessary in the 64 bit environment.
> +	 * It isn't necessary in a 32 bit environment but is considered
> +	 * a good practice.
> +	 */
> +	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));

I think a better comment here would be simply:

/* VCHI messages between the CPU and firmware use 32-bit bus addresses. */

explaining why the value is chosen (once you know that the 32 bit
restriction exists, reporting it is obviously needed).  I'm curious,
though: what failed when you didn't set it?

> +
> +	if (err < 0)
> +		return err;
> +
>  	(void)of_property_read_u32(dev->of_node, "cache-line-size",
>  				   &g_cache_line_size);
>  	g_fragments_size = 2 * g_cache_line_size;
> -- 
> 2.10.1
Michael Zoran Oct. 31, 2016, 6:40 p.m. UTC | #2
On Mon, 2016-10-31 at 11:36 -0700, Eric Anholt wrote:
> Michael Zoran <mzoran@crowfest.net> writes:
> 
> > Setting the DMA mask is optional on 32 bit but
> > is mandatory on 64 bit.  Set the DMA mask and coherent
> > to force all DMA to be in the 32 bit address space.
> > 
> > This is considered a "good practice" and most drivers
> > already do this.
> > 
> > Signed-off-by: Michael Zoran <mzoran@crowfest.net>
> > ---
> >  .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c |
> > 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> > 
> > diff --git
> > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.
> > c
> > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.
> > c
> > index a5afcc5..6fa2b5a 100644
> > ---
> > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.
> > c
> > +++
> > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.
> > c
> > @@ -97,6 +97,16 @@ int vchiq_platform_init(struct platform_device
> > *pdev, VCHIQ_STATE_T *state)
> >  	int slot_mem_size, frag_mem_size;
> >  	int err, irq, i;
> >  
> > +	/*
> > +	 * Setting the DMA mask is necessary in the 64 bit
> > environment.
> > +	 * It isn't necessary in a 32 bit environment but is
> > considered
> > +	 * a good practice.
> > +	 */
> > +	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
> 
> I think a better comment here would be simply:
> 
> /* VCHI messages between the CPU and firmware use 32-bit bus
> addresses. */
> 
> explaining why the value is chosen (once you know that the 32 bit
> restriction exists, reporting it is obviously needed).  I'm curious,
> though: what failed when you didn't set it?
> 

The comment is easy to change.

I don't have the log available ATM, but if I remember the DMA API's
bugcheck the first time that are used.  

I think this was a policy decision or something because the information
should be available in the dma-ranges.

If it's important, I can setup a test again without the change and e-
mail the logs.

If you look at the DWC2 driver you will see that it also sets this
mask.
Michael Zoran Oct. 31, 2016, 7:53 p.m. UTC | #3
On Mon, 2016-10-31 at 11:40 -0700, Michael Zoran wrote:
> On Mon, 2016-10-31 at 11:36 -0700, Eric Anholt wrote:
> > Michael Zoran <mzoran@crowfest.net> writes:
> > 
> > > Setting the DMA mask is optional on 32 bit but
> > > is mandatory on 64 bit.  Set the DMA mask and coherent
> > > to force all DMA to be in the 32 bit address space.
> > > 
> > > This is considered a "good practice" and most drivers
> > > already do this.
> > > 
> > > Signed-off-by: Michael Zoran <mzoran@crowfest.net>
> > > ---
> > >  .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c |
> > > 10 ++++++++++
> > >  1 file changed, 10 insertions(+)
> > > 
> > > diff --git
> > > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_ar
> > > m.
> > > c
> > > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_ar
> > > m.
> > > c
> > > index a5afcc5..6fa2b5a 100644
> > > ---
> > > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_ar
> > > m.
> > > c
> > > +++
> > > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_ar
> > > m.
> > > c
> > > @@ -97,6 +97,16 @@ int vchiq_platform_init(struct platform_device
> > > *pdev, VCHIQ_STATE_T *state)
> > >  	int slot_mem_size, frag_mem_size;
> > >  	int err, irq, i;
> > >  
> > > +	/*
> > > +	 * Setting the DMA mask is necessary in the 64 bit
> > > environment.
> > > +	 * It isn't necessary in a 32 bit environment but is
> > > considered
> > > +	 * a good practice.
> > > +	 */
> > > +	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
> > 
> > I think a better comment here would be simply:
> > 
> > /* VCHI messages between the CPU and firmware use 32-bit bus
> > addresses. */
> > 
> > explaining why the value is chosen (once you know that the 32 bit
> > restriction exists, reporting it is obviously needed).  I'm
> > curious,
> > though: what failed when you didn't set it?
> > 
> 
> The comment is easy to change.
> 
> I don't have the log available ATM, but if I remember the DMA API's
> bugcheck the first time that are used.  
> 
> I think this was a policy decision or something because the
> information
> should be available in the dma-ranges.
> 
> If it's important, I can setup a test again without the change and e-
> mail the logs.
> 
> If you look at the DWC2 driver you will see that it also sets this
> mask.

OK, I'm begging to understand this.  It appears the architecture
specific paths are very different.

In arm the mask and coherent is set to DMA_BIT_MASK(32) in mm/dma-
mapping.c the first time the dma APIs are used.  On arm64, it appears
this variable is uninitialized and will contain random crude.

Like I said, I don't know if this is a policy decision or if it just
slipped through the cracks.

arch/arm/mm/dma-mapping.c(Note the call to get_coherent_dma_mask(dev))

static u64 get_coherent_dma_mask(struct device *dev)
{
	u64 mask = (u64)DMA_BIT_MASK(32);

	if (dev) {
		mask = dev->coherent_dma_mask;

		/*
		 * Sanity check the DMA mask - it must be non-zero, and
		 * must be able to be satisfied by a DMA allocation.
		 */
		if (mask == 0) {
			dev_warn(dev, "coherent DMA mask is unset\n");
			return 0;
		}

		if (!__dma_supported(dev, mask, true))
			return 0;
	}

	return mask;
}

static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t
*handle,
			 gfp_t gfp, pgprot_t prot, bool is_coherent,
			 unsigned long attrs, const void *caller)
{
	u64 mask = get_coherent_dma_mask(dev);
	struct page *page = NULL;
	void *addr;
	bool allowblock, cma;
	struct arm_dma_buffer *buf;
	struct arm_dma_alloc_args args = {
		.dev = dev,
		.size = PAGE_ALIGN(size),
		.gfp = gfp,
		.prot = prot,
		.caller = caller,
		.want_vaddr = ((attrs & DMA_ATTR_NO_KERNEL_MAPPING) ==
0),
		.coherent_flag = is_coherent ? COHERENT : NORMAL,
	};

arch/arm64/include/asm/dma-mapping.h

/* do not use this function in a driver */
static inline bool is_device_dma_coherent(struct device *dev)
{
	if (!dev)
		return false;
	return dev->archdata.dma_coherent;
}

arch/arm64/mm/dma-mapping.c(Note no call to get_coherent_dma_mask)

static void *__dma_alloc(struct device *dev, size_t size,
			 dma_addr_t *dma_handle, gfp_t flags,
			 unsigned long attrs)
{
	struct page *page;
	void *ptr, *coherent_ptr;
	bool coherent = is_device_dma_coherent(dev);
	pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, false);

	size = PAGE_ALIGN(size);

	if (!coherent && !gfpflags_allow_blocking(flags)) {
		struct page *page = NULL;
		void *addr = __alloc_from_pool(size, &page, flags);

		if (addr)
			*dma_handle = phys_to_dma(dev,
page_to_phys(page));

		return addr;
	}

	ptr = __dma_alloc_coherent(dev, size, dma_handle, flags,
attrs);
Michael Zoran Oct. 31, 2016, 7:57 p.m. UTC | #4
On Mon, 2016-10-31 at 12:53 -0700, Michael Zoran wrote:
> On Mon, 2016-10-31 at 11:40 -0700, Michael Zoran wrote:
> > On Mon, 2016-10-31 at 11:36 -0700, Eric Anholt wrote:
> > > Michael Zoran <mzoran@crowfest.net> writes:
> > > 
> > > > Setting the DMA mask is optional on 32 bit but
> > > > is mandatory on 64 bit.  Set the DMA mask and coherent
> > > > to force all DMA to be in the 32 bit address space.
> > > > 
> > > > This is considered a "good practice" and most drivers
> > > > already do this.
> > > > 
> > > > Signed-off-by: Michael Zoran <mzoran@crowfest.net>
> > > > ---
> > > >  .../staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
> > > > |
> > > > 10 ++++++++++
> > > >  1 file changed, 10 insertions(+)
> > > > 
> > > > diff --git
> > > > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_
> > > > ar
> > > > m.
> > > > c
> > > > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_
> > > > ar
> > > > m.
> > > > c
> > > > index a5afcc5..6fa2b5a 100644
> > > > ---
> > > > a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_
> > > > ar
> > > > m.
> > > > c
> > > > +++
> > > > b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_
> > > > ar
> > > > m.
> > > > c
> > > > @@ -97,6 +97,16 @@ int vchiq_platform_init(struct
> > > > platform_device
> > > > *pdev, VCHIQ_STATE_T *state)
> > > >  	int slot_mem_size, frag_mem_size;
> > > >  	int err, irq, i;
> > > >  
> > > > +	/*
> > > > +	 * Setting the DMA mask is necessary in the 64 bit
> > > > environment.
> > > > +	 * It isn't necessary in a 32 bit environment but is
> > > > considered
> > > > +	 * a good practice.
> > > > +	 */
> > > > +	err = dma_set_mask_and_coherent(dev,
> > > > DMA_BIT_MASK(32));
> > > 
> > > I think a better comment here would be simply:
> > > 
> > > /* VCHI messages between the CPU and firmware use 32-bit bus
> > > addresses. */
> > > 
> > > explaining why the value is chosen (once you know that the 32 bit
> > > restriction exists, reporting it is obviously needed).  I'm
> > > curious,
> > > though: what failed when you didn't set it?
> > > 
> > 
> > The comment is easy to change.
> > 
> > I don't have the log available ATM, but if I remember the DMA API's
> > bugcheck the first time that are used.  
> > 
> > I think this was a policy decision or something because the
> > information
> > should be available in the dma-ranges.
> > 
> > If it's important, I can setup a test again without the change and
> > e-
> > mail the logs.
> > 
> > If you look at the DWC2 driver you will see that it also sets this
> > mask.
> 
> OK, I'm begging to understand this.  It appears the architecture
> specific paths are very different.
> 
> In arm the mask and coherent is set to DMA_BIT_MASK(32) in mm/dma-
> mapping.c the first time the dma APIs are used.  On arm64, it appears
> this variable is uninitialized and will contain random crude.
> 
> Like I said, I don't know if this is a policy decision or if it just
> slipped through the cracks.
> 
Actually, I'm getting confused here.   If I need to prove this is
needed, is their anybody I can send e-mail to that has a deep
understanding of the two different architecture paths.   Perhaps they
can explain exactly why arm64 is not defaulting to 32 bit DMA.
Russell King (Oracle) Nov. 1, 2016, 9:31 a.m. UTC | #5
On Mon, Oct 31, 2016 at 12:53:13PM -0700, Michael Zoran wrote:
> On Mon, 2016-10-31 at 11:40 -0700, Michael Zoran wrote:
> > The comment is easy to change.
> > 
> > I don't have the log available ATM, but if I remember the DMA API's
> > bugcheck the first time that are used.  
> > 
> > I think this was a policy decision or something because the
> > information
> > should be available in the dma-ranges.
> > 
> > If it's important, I can setup a test again without the change and e-
> > mail the logs.
> > 
> > If you look at the DWC2 driver you will see that it also sets this
> > mask.
> 
> OK, I'm begging to understand this.  It appears the architecture
> specific paths are very different.
> 
> In arm the mask and coherent is set to DMA_BIT_MASK(32) in mm/dma-
> mapping.c the first time the dma APIs are used.

I'm not sure where you get that from, this is absolutely not the case.

>  On arm64, it appears
> this variable is uninitialized and will contain random crude.
> 
> Like I said, I don't know if this is a policy decision or if it just
> slipped through the cracks.
> 
> arch/arm/mm/dma-mapping.c(Note the call to get_coherent_dma_mask(dev))
> 
> static u64 get_coherent_dma_mask(struct device *dev)
> {
> 	u64 mask = (u64)DMA_BIT_MASK(32);
> 
> 	if (dev) {
> 		mask = dev->coherent_dma_mask;
> 
> 		/*
> 		 * Sanity check the DMA mask - it must be non-zero, and
> 		 * must be able to be satisfied by a DMA allocation.
> 		 */
> 		if (mask == 0) {
> 			dev_warn(dev, "coherent DMA mask is unset\n");
> 			return 0;

If the mask is unset (iow, zero) this function returns zero.  There
is no "the first time" here anywhere - dev->coherent_dma_mask remains
at zero and doesn't change as a result of calling this path.

> 		}
> 
> 		if (!__dma_supported(dev, mask, true))
> 			return 0;
> 	}
> 
> 	return mask;
> }
> 
> static void *__dma_alloc(struct device *dev, size_t size, dma_addr_t
> *handle,
> 			 gfp_t gfp, pgprot_t prot, bool is_coherent,
> 			 unsigned long attrs, const void *caller)
> {
> 	u64 mask = get_coherent_dma_mask(dev);
> 	struct page *page = NULL;
> 	void *addr;
> 	bool allowblock, cma;
> 	struct arm_dma_buffer *buf;
> 	struct arm_dma_alloc_args args = {
> 		.dev = dev,
> 		.size = PAGE_ALIGN(size),
> 		.gfp = gfp,
> 		.prot = prot,
> 		.caller = caller,
> 		.want_vaddr = ((attrs & DMA_ATTR_NO_KERNEL_MAPPING) ==
> 0),
> 		.coherent_flag = is_coherent ? COHERENT : NORMAL,
> 	};

which then causes:

        if (!mask)
                return NULL;

this function to return NULL, hence failing the request.  Rightfully
so, because we don't know what an acceptable allocation for the device
would be, as the device is effectively saying "I support 0 bits of DMA
address."

Now, firstly, the driver is required to call one of:

	dma_set_mask_and_coherent()
	dma_set_coherent_mask()

if it is to use coherent DMA - see Documentation/DMA-API-HOWTO.txt.
However, the bus code should already have setup a default mask in both
dev->dma_mask and the coherent DMA mask if DMA is possible, which
normally will be the 32-bit mask.

As explained in the document, if the device and driver supports 64-bit
addressing, and wants to make use of it, it must successfully negotiate
with the kernel before using it, which includes making calls to change
the DMA masks.
Robin Murphy Nov. 1, 2016, 12:56 p.m. UTC | #6
Hi Michael,

On 31/10/16 19:53, Michael Zoran wrote:
> On Mon, 2016-10-31 at 11:40 -0700, Michael Zoran wrote:
>> On Mon, 2016-10-31 at 11:36 -0700, Eric Anholt wrote:
>>> Michael Zoran <mzoran@crowfest.net> writes:
>>>
>>>> Setting the DMA mask is optional on 32 bit but
>>>> is mandatory on 64 bit.  Set the DMA mask and coherent
>>>> to force all DMA to be in the 32 bit address space.
>>>>
>>>> This is considered a "good practice" and most drivers
>>>> already do this.
>>>>
>>>> Signed-off-by: Michael Zoran <mzoran@crowfest.net>
>>>> ---

[...]

>>>> +	/*
>>>> +	 * Setting the DMA mask is necessary in the 64 bit
>>>> environment.
>>>> +	 * It isn't necessary in a 32 bit environment but is
>>>> considered
>>>> +	 * a good practice.
>>>> +	 */
>>>> +	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
>>>
>>> I think a better comment here would be simply:
>>>
>>> /* VCHI messages between the CPU and firmware use 32-bit bus
>>> addresses. */
>>>
>>> explaining why the value is chosen (once you know that the 32 bit
>>> restriction exists, reporting it is obviously needed).  I'm
>>> curious,
>>> though: what failed when you didn't set it?
>>>
>>
>> The comment is easy to change.
>>
>> I don't have the log available ATM, but if I remember the DMA API's
>> bugcheck the first time that are used.  
>>
>> I think this was a policy decision or something because the
>> information
>> should be available in the dma-ranges.
>>
>> If it's important, I can setup a test again without the change and e-
>> mail the logs.
>>
>> If you look at the DWC2 driver you will see that it also sets this
>> mask.
> 
> OK, I'm begging to understand this.  It appears the architecture
> specific paths are very different.
> 
> In arm the mask and coherent is set to DMA_BIT_MASK(32) in mm/dma-
> mapping.c the first time the dma APIs are used.  On arm64, it appears
> this variable is uninitialized and will contain random crude.
> 
> Like I said, I don't know if this is a policy decision or if it just
> slipped through the cracks.

[...]

> arch/arm64/mm/dma-mapping.c(Note no call to get_coherent_dma_mask)
> 
> static void *__dma_alloc(struct device *dev, size_t size,
> 			 dma_addr_t *dma_handle, gfp_t flags,
> 			 unsigned long attrs)
> {
> 	struct page *page;
> 	void *ptr, *coherent_ptr;
> 	bool coherent = is_device_dma_coherent(dev);
> 	pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL, false);
> 
> 	size = PAGE_ALIGN(size);
> 
> 	if (!coherent && !gfpflags_allow_blocking(flags)) {
> 		struct page *page = NULL;
> 		void *addr = __alloc_from_pool(size, &page, flags);
> 
> 		if (addr)
> 			*dma_handle = phys_to_dma(dev,
> page_to_phys(page));
> 
> 		return addr;
> 	}
> 
> 	ptr = __dma_alloc_coherent(dev, size, dma_handle, flags,
> attrs);

In the general case, the device's coherent DMA mask is checked inside
that helper function, and then again further on in the SWIOTLB code if
necessary. For the atomic pool case beforehand, the pool is already
allocated below 4GB, and on arm64 we don't really expect devices to have
DMA masks *smaller* than 32 bits.

Devices created from DT get 32-bit DMA masks by default, although the
dma-ranges property may override that - see of_dma_configure() - so if
you somehow have a device with an uninitialised mask then I can only
assume there's some platform driver shenanigans going on. Adding the
dma_set_mask() call to the driver is no bad thing, but the rationale in
the commit message is bogus.

Robin.
diff mbox

Patch

diff --git a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
index a5afcc5..6fa2b5a 100644
--- a/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
+++ b/drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c
@@ -97,6 +97,16 @@  int vchiq_platform_init(struct platform_device *pdev, VCHIQ_STATE_T *state)
 	int slot_mem_size, frag_mem_size;
 	int err, irq, i;
 
+	/*
+	 * Setting the DMA mask is necessary in the 64 bit environment.
+	 * It isn't necessary in a 32 bit environment but is considered
+	 * a good practice.
+	 */
+	err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
+
+	if (err < 0)
+		return err;
+
 	(void)of_property_read_u32(dev->of_node, "cache-line-size",
 				   &g_cache_line_size);
 	g_fragments_size = 2 * g_cache_line_size;