2023-03-24 11:22:48

by Steven Price

[permalink] [raw]
Subject: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

Similar to exynos, we need a set_platform_dma_ops() callback for proper
operation on ARM 32 bit after recent changes in the IOMMU framework
(detach ops removal). But also the use of a NULL domain is confusing.

Rework the code to have a singleton rk_identity_domain which is assigned
to domain when using an identity mapping rather than "detaching". This
makes the code easier to reason about.

Signed-off-by: Steven Price <[email protected]>
---
Changes since v1[1]:

* Reworked the code to avoid a NULL domain, instead a singleton
rk_identity_domain is used instead. The 'detach' language is no
longer used.

[1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com

drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
1 file changed, 39 insertions(+), 11 deletions(-)

diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
index f30db22ea5d7..437541004994 100644
--- a/drivers/iommu/rockchip-iommu.c
+++ b/drivers/iommu/rockchip-iommu.c
@@ -124,6 +124,7 @@ struct rk_iommudata {

static struct device *dma_dev;
static const struct rk_iommu_ops *rk_ops;
+static struct iommu_domain rk_identity_domain;

static inline void rk_table_flush(struct rk_iommu_domain *dom, dma_addr_t dma,
unsigned int count)
@@ -980,26 +981,27 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
return ret;
}

-static void rk_iommu_detach_device(struct iommu_domain *domain,
- struct device *dev)
+static int rk_iommu_identity_attach(struct iommu_domain *identity_domain,
+ struct device *dev)
{
struct rk_iommu *iommu;
- struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
+ struct rk_iommu_domain *rk_domain;
unsigned long flags;
int ret;

/* Allow 'virtual devices' (eg drm) to detach from domain */
iommu = rk_iommu_from_dev(dev);
if (!iommu)
- return;
+ return -ENODEV;
+
+ rk_domain = to_rk_domain(iommu->domain);

dev_dbg(dev, "Detaching from iommu domain\n");

- /* iommu already detached */
- if (iommu->domain != domain)
- return;
+ if (iommu->domain == identity_domain)
+ return 0;

- iommu->domain = NULL;
+ iommu->domain = identity_domain;

spin_lock_irqsave(&rk_domain->iommus_lock, flags);
list_del_init(&iommu->node);
@@ -1011,7 +1013,25 @@ static void rk_iommu_detach_device(struct iommu_domain *domain,
rk_iommu_disable(iommu);
pm_runtime_put(iommu->dev);
}
+
+ return 0;
+}
+
+static struct iommu_domain_ops rk_identity_ops = {
+ .attach_dev = rk_iommu_identity_attach,
+};
+
+static struct iommu_domain rk_identity_domain = {
+ .type = IOMMU_DOMAIN_IDENTITY,
+ .ops = &rk_identity_ops,
+};
+
+#ifdef CONFIG_ARM
+static void rk_iommu_set_platform_dma(struct device *dev)
+{
+ WARN_ON(rk_iommu_identity_attach(&rk_identity_domain, dev));
}
+#endif

static int rk_iommu_attach_device(struct iommu_domain *domain,
struct device *dev)
@@ -1035,8 +1055,9 @@ static int rk_iommu_attach_device(struct iommu_domain *domain,
if (iommu->domain == domain)
return 0;

- if (iommu->domain)
- rk_iommu_detach_device(iommu->domain, dev);
+ ret = rk_iommu_identity_attach(&rk_identity_domain, dev);
+ if (ret)
+ return ret;

iommu->domain = domain;

@@ -1050,7 +1071,7 @@ static int rk_iommu_attach_device(struct iommu_domain *domain,

ret = rk_iommu_enable(iommu);
if (ret)
- rk_iommu_detach_device(iommu->domain, dev);
+ WARN_ON(rk_iommu_identity_attach(&rk_identity_domain, dev));

pm_runtime_put(iommu->dev);

@@ -1061,6 +1082,9 @@ static struct iommu_domain *rk_iommu_domain_alloc(unsigned type)
{
struct rk_iommu_domain *rk_domain;

+ if (type == IOMMU_DOMAIN_IDENTITY)
+ return &rk_identity_domain;
+
if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
return NULL;

@@ -1176,6 +1200,7 @@ static int rk_iommu_of_xlate(struct device *dev,
iommu_dev = of_find_device_by_node(args->np);

data->iommu = platform_get_drvdata(iommu_dev);
+ data->iommu->domain = &rk_identity_domain;
dev_iommu_priv_set(dev, data);

platform_device_put(iommu_dev);
@@ -1188,6 +1213,9 @@ static const struct iommu_ops rk_iommu_ops = {
.probe_device = rk_iommu_probe_device,
.release_device = rk_iommu_release_device,
.device_group = rk_iommu_device_group,
+#ifdef CONFIG_ARM
+ .set_platform_dma_ops = rk_iommu_set_platform_dma,
+#endif
.pgsize_bitmap = RK_IOMMU_PGSIZE_BITMAP,
.of_xlate = rk_iommu_of_xlate,
.default_domain_ops = &(const struct iommu_domain_ops) {
--
2.39.2


2023-03-24 13:18:01

by Lu Baolu

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On 2023/3/24 19:11, Steven Price wrote:
> Similar to exynos, we need a set_platform_dma_ops() callback for proper
> operation on ARM 32 bit after recent changes in the IOMMU framework
> (detach ops removal). But also the use of a NULL domain is confusing.
>
> Rework the code to have a singleton rk_identity_domain which is assigned
> to domain when using an identity mapping rather than "detaching". This
> makes the code easier to reason about.
>
> Signed-off-by: Steven Price <[email protected]>
> ---
> Changes since v1[1]:
>
> * Reworked the code to avoid a NULL domain, instead a singleton
> rk_identity_domain is used instead. The 'detach' language is no
> longer used.
>
> [1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>
> drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
> 1 file changed, 39 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
> index f30db22ea5d7..437541004994 100644
> --- a/drivers/iommu/rockchip-iommu.c
> +++ b/drivers/iommu/rockchip-iommu.c
> @@ -124,6 +124,7 @@ struct rk_iommudata {
>
> static struct device *dma_dev;
> static const struct rk_iommu_ops *rk_ops;
> +static struct iommu_domain rk_identity_domain;
>
> static inline void rk_table_flush(struct rk_iommu_domain *dom, dma_addr_t dma,
> unsigned int count)
> @@ -980,26 +981,27 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
> return ret;
> }
>
> -static void rk_iommu_detach_device(struct iommu_domain *domain,
> - struct device *dev)
> +static int rk_iommu_identity_attach(struct iommu_domain *identity_domain,
> + struct device *dev)
> {
> struct rk_iommu *iommu;
> - struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
> + struct rk_iommu_domain *rk_domain;
> unsigned long flags;
> int ret;
>
> /* Allow 'virtual devices' (eg drm) to detach from domain */
> iommu = rk_iommu_from_dev(dev);
> if (!iommu)
> - return;
> + return -ENODEV;
> +
> + rk_domain = to_rk_domain(iommu->domain);
>
> dev_dbg(dev, "Detaching from iommu domain\n");
>
> - /* iommu already detached */
> - if (iommu->domain != domain)
> - return;
> + if (iommu->domain == identity_domain)
> + return 0;
>
> - iommu->domain = NULL;
> + iommu->domain = identity_domain;

Where did identity_domain come from? Is it rk_identity_domain?

Best regards,
baolu

2023-03-24 13:32:29

by Steven Price

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On 24/03/2023 13:16, Baolu Lu wrote:
> On 2023/3/24 19:11, Steven Price wrote:
>> Similar to exynos, we need a set_platform_dma_ops() callback for proper
>> operation on ARM 32 bit after recent changes in the IOMMU framework
>> (detach ops removal). But also the use of a NULL domain is confusing.
>>
>> Rework the code to have a singleton rk_identity_domain which is assigned
>> to domain when using an identity mapping rather than "detaching". This
>> makes the code easier to reason about.
>>
>> Signed-off-by: Steven Price <[email protected]>
>> ---
>> Changes since v1[1]:
>>
>>   * Reworked the code to avoid a NULL domain, instead a singleton
>>     rk_identity_domain is used instead. The 'detach' language is no
>>     longer used.
>>
>> [1]
>> https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>>
>>   drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
>>   1 file changed, 39 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/iommu/rockchip-iommu.c
>> b/drivers/iommu/rockchip-iommu.c
>> index f30db22ea5d7..437541004994 100644
>> --- a/drivers/iommu/rockchip-iommu.c
>> +++ b/drivers/iommu/rockchip-iommu.c
>> @@ -124,6 +124,7 @@ struct rk_iommudata {
>>     static struct device *dma_dev;
>>   static const struct rk_iommu_ops *rk_ops;
>> +static struct iommu_domain rk_identity_domain;
>>     static inline void rk_table_flush(struct rk_iommu_domain *dom,
>> dma_addr_t dma,
>>                     unsigned int count)
>> @@ -980,26 +981,27 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
>>       return ret;
>>   }
>>   -static void rk_iommu_detach_device(struct iommu_domain *domain,
>> -                   struct device *dev)
>> +static int rk_iommu_identity_attach(struct iommu_domain
>> *identity_domain,
>> +                    struct device *dev)
>>   {
>>       struct rk_iommu *iommu;
>> -    struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
>> +    struct rk_iommu_domain *rk_domain;
>>       unsigned long flags;
>>       int ret;
>>         /* Allow 'virtual devices' (eg drm) to detach from domain */
>>       iommu = rk_iommu_from_dev(dev);
>>       if (!iommu)
>> -        return;
>> +        return -ENODEV;
>> +
>> +    rk_domain = to_rk_domain(iommu->domain);
>>         dev_dbg(dev, "Detaching from iommu domain\n");
>>   -    /* iommu already detached */
>> -    if (iommu->domain != domain)
>> -        return;
>> +    if (iommu->domain == identity_domain)
>> +        return 0;
>>   -    iommu->domain = NULL;
>> +    iommu->domain = identity_domain;
>
> Where did identity_domain come from? Is it rk_identity_domain?

It's a parameter of the function. In the case of the call in
rk_iommu_attach_device() then, yes, it's rk_identity_domain. But this
function is also the "attach_dev" callback of "rk_identity_ops".

I'll admit this is cargo-culted from Jason's example:

https://lore.kernel.org/linux-iommu/[email protected]/

Steve

> Best regards,
> baolu

2023-03-24 13:33:38

by Lu Baolu

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On 2023/3/24 21:24, Steven Price wrote:
> On 24/03/2023 13:16, Baolu Lu wrote:
>> On 2023/3/24 19:11, Steven Price wrote:
>>> Similar to exynos, we need a set_platform_dma_ops() callback for proper
>>> operation on ARM 32 bit after recent changes in the IOMMU framework
>>> (detach ops removal). But also the use of a NULL domain is confusing.
>>>
>>> Rework the code to have a singleton rk_identity_domain which is assigned
>>> to domain when using an identity mapping rather than "detaching". This
>>> makes the code easier to reason about.
>>>
>>> Signed-off-by: Steven Price<[email protected]>
>>> ---
>>> Changes since v1[1]:
>>>
>>>   * Reworked the code to avoid a NULL domain, instead a singleton
>>>     rk_identity_domain is used instead. The 'detach' language is no
>>>     longer used.
>>>
>>> [1]
>>> https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>>>
>>>   drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
>>>   1 file changed, 39 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/drivers/iommu/rockchip-iommu.c
>>> b/drivers/iommu/rockchip-iommu.c
>>> index f30db22ea5d7..437541004994 100644
>>> --- a/drivers/iommu/rockchip-iommu.c
>>> +++ b/drivers/iommu/rockchip-iommu.c
>>> @@ -124,6 +124,7 @@ struct rk_iommudata {
>>>     static struct device *dma_dev;
>>>   static const struct rk_iommu_ops *rk_ops;
>>> +static struct iommu_domain rk_identity_domain;
>>>     static inline void rk_table_flush(struct rk_iommu_domain *dom,
>>> dma_addr_t dma,
>>>                     unsigned int count)
>>> @@ -980,26 +981,27 @@ static int rk_iommu_enable(struct rk_iommu *iommu)
>>>       return ret;
>>>   }
>>>   -static void rk_iommu_detach_device(struct iommu_domain *domain,
>>> -                   struct device *dev)
>>> +static int rk_iommu_identity_attach(struct iommu_domain
>>> *identity_domain,
>>> +                    struct device *dev)
>>>   {
>>>       struct rk_iommu *iommu;
>>> -    struct rk_iommu_domain *rk_domain = to_rk_domain(domain);
>>> +    struct rk_iommu_domain *rk_domain;
>>>       unsigned long flags;
>>>       int ret;
>>>         /* Allow 'virtual devices' (eg drm) to detach from domain */
>>>       iommu = rk_iommu_from_dev(dev);
>>>       if (!iommu)
>>> -        return;
>>> +        return -ENODEV;
>>> +
>>> +    rk_domain = to_rk_domain(iommu->domain);
>>>         dev_dbg(dev, "Detaching from iommu domain\n");
>>>   -    /* iommu already detached */
>>> -    if (iommu->domain != domain)
>>> -        return;
>>> +    if (iommu->domain == identity_domain)
>>> +        return 0;
>>>   -    iommu->domain = NULL;
>>> +    iommu->domain = identity_domain;
>> Where did identity_domain come from? Is it rk_identity_domain?
> It's a parameter of the function. In the case of the call in
> rk_iommu_attach_device() then, yes, it's rk_identity_domain. But this
> function is also the "attach_dev" callback of "rk_identity_ops".
>
> I'll admit this is cargo-culted from Jason's example:
>
> https://lore.kernel.org/linux-iommu/[email protected]/

Oh! I overlooked that. Thank you for the explanation.

Best regards,
baolu

2023-03-27 14:46:49

by John Keeping

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On Fri, Mar 24, 2023 at 11:11:27AM +0000, Steven Price wrote:
> Similar to exynos, we need a set_platform_dma_ops() callback for proper
> operation on ARM 32 bit after recent changes in the IOMMU framework
> (detach ops removal). But also the use of a NULL domain is confusing.
>
> Rework the code to have a singleton rk_identity_domain which is assigned
> to domain when using an identity mapping rather than "detaching". This
> makes the code easier to reason about.
>
> Signed-off-by: Steven Price <[email protected]>
> ---
> Changes since v1[1]:
>
> * Reworked the code to avoid a NULL domain, instead a singleton
> rk_identity_domain is used instead. The 'detach' language is no
> longer used.
>
> [1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>
> drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
> 1 file changed, 39 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
> index f30db22ea5d7..437541004994 100644
> --- a/drivers/iommu/rockchip-iommu.c
> +++ b/drivers/iommu/rockchip-iommu.c
[snip]
> +static struct iommu_domain rk_identity_domain = {
> + .type = IOMMU_DOMAIN_IDENTITY,
> + .ops = &rk_identity_ops,
> +};
> +
> +#ifdef CONFIG_ARM

Is this #ifdef needed? I can't see anything ARM-specific about this
function or .set_platform_dma_ops.

> +static void rk_iommu_set_platform_dma(struct device *dev)
> +{
> + WARN_ON(rk_iommu_identity_attach(&rk_identity_domain, dev));
> }
> +#endif

Not shown in the patch are the pm_runtime hooks. Do they need to
change like this?

static int __maybe_unused rk_iommu_suspend(struct device *dev)
{
struct rk_iommu *iommu = dev_get_drvdata(dev);

- if (!iommu->domain)
+ if (iommu->domain == &rk_identity_domain)
return 0;

rk_iommu_disable(iommu);
return 0;
}

static int __maybe_unused rk_iommu_resume(struct device *dev)
{
struct rk_iommu *iommu = dev_get_drvdata(dev);

- if (!iommu->domain)
+ if (iommu->domain == &rk_identity_domain)
return 0;

return rk_iommu_enable(iommu);
}

2023-03-27 15:29:16

by Jason Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On Mon, Mar 27, 2023 at 03:35:04PM +0100, John Keeping wrote:
> On Fri, Mar 24, 2023 at 11:11:27AM +0000, Steven Price wrote:
> > Similar to exynos, we need a set_platform_dma_ops() callback for proper
> > operation on ARM 32 bit after recent changes in the IOMMU framework
> > (detach ops removal). But also the use of a NULL domain is confusing.
> >
> > Rework the code to have a singleton rk_identity_domain which is assigned
> > to domain when using an identity mapping rather than "detaching". This
> > makes the code easier to reason about.
> >
> > Signed-off-by: Steven Price <[email protected]>
> > ---
> > Changes since v1[1]:
> >
> > * Reworked the code to avoid a NULL domain, instead a singleton
> > rk_identity_domain is used instead. The 'detach' language is no
> > longer used.
> >
> > [1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
> >
> > drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
> > 1 file changed, 39 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
> > index f30db22ea5d7..437541004994 100644
> > --- a/drivers/iommu/rockchip-iommu.c
> > +++ b/drivers/iommu/rockchip-iommu.c
> [snip]
> > +static struct iommu_domain rk_identity_domain = {
> > + .type = IOMMU_DOMAIN_IDENTITY,
> > + .ops = &rk_identity_ops,
> > +};
> > +
> > +#ifdef CONFIG_ARM
>
> Is this #ifdef needed? I can't see anything ARM-specific about this
> function or .set_platform_dma_ops.

set_platform_dma_ops is never called on ARM64.

> Not shown in the patch are the pm_runtime hooks. Do they need to
> change like this?

Most likely yes

Jason

2023-03-29 23:45:37

by Jason Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On Fri, Mar 24, 2023 at 11:11:27AM +0000, Steven Price wrote:
> Similar to exynos, we need a set_platform_dma_ops() callback for proper
> operation on ARM 32 bit after recent changes in the IOMMU framework
> (detach ops removal). But also the use of a NULL domain is confusing.
>
> Rework the code to have a singleton rk_identity_domain which is assigned
> to domain when using an identity mapping rather than "detaching". This
> makes the code easier to reason about.
>
> Signed-off-by: Steven Price <[email protected]>
> ---
> Changes since v1[1]:
>
> * Reworked the code to avoid a NULL domain, instead a singleton
> rk_identity_domain is used instead. The 'detach' language is no
> longer used.
>
> [1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>
> drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
> 1 file changed, 39 insertions(+), 11 deletions(-)

Aside from the pm stuff this looks OK to me

Reviewed-by: Jason Gunthorpe <[email protected]>

Thanks,
Jason

2023-03-30 14:46:40

by Steven Price

[permalink] [raw]
Subject: Re: [PATCH v2] iommu/rockchip: Add missing set_platform_dma_ops callback

On 27/03/2023 16:26, Jason Gunthorpe wrote:
> On Mon, Mar 27, 2023 at 03:35:04PM +0100, John Keeping wrote:
>> On Fri, Mar 24, 2023 at 11:11:27AM +0000, Steven Price wrote:
>>> Similar to exynos, we need a set_platform_dma_ops() callback for proper
>>> operation on ARM 32 bit after recent changes in the IOMMU framework
>>> (detach ops removal). But also the use of a NULL domain is confusing.
>>>
>>> Rework the code to have a singleton rk_identity_domain which is assigned
>>> to domain when using an identity mapping rather than "detaching". This
>>> makes the code easier to reason about.
>>>
>>> Signed-off-by: Steven Price <[email protected]>
>>> ---
>>> Changes since v1[1]:
>>>
>>> * Reworked the code to avoid a NULL domain, instead a singleton
>>> rk_identity_domain is used instead. The 'detach' language is no
>>> longer used.
>>>
>>> [1] https://lore.kernel.org/r/20230315164152.333251-1-steven.price%40arm.com
>>>
>>> drivers/iommu/rockchip-iommu.c | 50 ++++++++++++++++++++++++++--------
>>> 1 file changed, 39 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
>>> index f30db22ea5d7..437541004994 100644
>>> --- a/drivers/iommu/rockchip-iommu.c
>>> +++ b/drivers/iommu/rockchip-iommu.c
>> [snip]
>>> +static struct iommu_domain rk_identity_domain = {
>>> + .type = IOMMU_DOMAIN_IDENTITY,
>>> + .ops = &rk_identity_ops,
>>> +};
>>> +
>>> +#ifdef CONFIG_ARM
>>
>> Is this #ifdef needed? I can't see anything ARM-specific about this
>> function or .set_platform_dma_ops.
>
> set_platform_dma_ops is never called on ARM64.

I have to admit this was somewhat cargo-culted from the exynos-iommu
change. This is only needed for Arm, but I'm not sure if there's any
real harm including it for other arches.

>> Not shown in the patch are the pm_runtime hooks. Do they need to
>> change like this?
>
> Most likely yes

Good spot - I'll send a v3 with this fixed.

Thanks,

Steve