[PATCH v2 1/1] iommu: Add config option to set lazy mode as default

[PATCH v2 1/1] iommu: Add config option to set lazy mode as default

[Zhen Lei]

This allows the default behaviour to be controlled by a kernel config
option instead of changing the command line for the kernel to include
"iommu.strict=0" on ARM64 where this is desired.

This is similar to CONFIG_IOMMU_DEFAULT_PASSTHROUGH.

Note: At present, intel_iommu, amd_iommu and s390_iommu use lazy mode as
default, so there is no need to add code for them.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
---
 drivers/iommu/Kconfig | 14 ++++++++++++++
 drivers/iommu/iommu.c |  5 +++++
 2 files changed, 19 insertions(+)

diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 6f07f3b21816c64..5daa110d0e83a07 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -85,6 +85,20 @@ config IOMMU_DEFAULT_PASSTHROUGH

 	  If unsure, say N here.

+config IOMMU_DMA_DEFAULT_LAZY_MODE
+	bool "IOMMU DMA use lazy mode to flush IOTLB and free IOVA"
+	depends on IOMMU_API
+	help
+	  Support lazy mode, where for every IOMMU DMA unmap operation, the
+	  flush operation of IOTLB and the free operation of IOVA are deferred.
+	  They are only guaranteed to be done before the related IOVA will be
+	  reused. Removing the need to pass in kernel parameters through
+	  command line. For example, iommu.strict=0 on ARM64. If this is
+	  enabled, you can still disable with kernel parameters, such as
+	  iommu.strict=1 depending on the architecture.
+
+	  If unsure, say N here.
+
 config OF_IOMMU
        def_bool y
        depends on OF && IOMMU_API
diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 33a982e33716369..5acb98e79b5b32d 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -43,7 +43,12 @@
 #else
 static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
 #endif
+
+#ifdef CONFIG_IOMMU_DMA_DEFAULT_LAZY_MODE
+static bool iommu_dma_strict __read_mostly;
+#else
 static bool iommu_dma_strict __read_mostly = true;
+#endif

 struct iommu_callback_data {
 	const struct iommu_ops *ops;
--
1.8.3

Re: [PATCH v2 1/1] iommu: Add config option to set lazy mode as default

[Robin Murphy]

On 27/03/2019 15:00, Zhen Lei wrote:
> This allows the default behaviour to be controlled by a kernel config
> option instead of changing the command line for the kernel to include
> "iommu.strict=0" on ARM64 where this is desired.
> 
> This is similar to CONFIG_IOMMU_DEFAULT_PASSTHROUGH.
> 
> Note: At present, intel_iommu, amd_iommu and s390_iommu use lazy mode as
> default, so there is no need to add code for them.

That seems a bit self-contradictory - if there's a real need for TLB 
flush behaviour to be statically configurable and not command-line-based 
then why should users of other architectures be exempt?

> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
> ---
>   drivers/iommu/Kconfig | 14 ++++++++++++++
>   drivers/iommu/iommu.c |  5 +++++
>   2 files changed, 19 insertions(+)
> 
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 6f07f3b21816c64..5daa110d0e83a07 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -85,6 +85,20 @@ config IOMMU_DEFAULT_PASSTHROUGH
> 
>   	  If unsure, say N here.
> 
> +config IOMMU_DMA_DEFAULT_LAZY_MODE
> +	bool "IOMMU DMA use lazy mode to flush IOTLB and free IOVA"
> +	depends on IOMMU_API
> +	help
> +	  Support lazy mode, where for every IOMMU DMA unmap operation, the
> +	  flush operation of IOTLB and the free operation of IOVA are deferred.
> +	  They are only guaranteed to be done before the related IOVA will be
> +	  reused. Removing the need to pass in kernel parameters through
> +	  command line. For example, iommu.strict=0 on ARM64. If this is
> +	  enabled, you can still disable with kernel parameters, such as
> +	  iommu.strict=1 depending on the architecture.
> +
> +	  If unsure, say N here.
> +
>   config OF_IOMMU
>          def_bool y
>          depends on OF && IOMMU_API
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index 33a982e33716369..5acb98e79b5b32d 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -43,7 +43,12 @@
>   #else
>   static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
>   #endif
> +
> +#ifdef CONFIG_IOMMU_DMA_DEFAULT_LAZY_MODE
> +static bool iommu_dma_strict __read_mostly;
> +#else
>   static bool iommu_dma_strict __read_mostly = true;
> +#endif

For a straightforward boolean, you can simply do:

	static bool foo = IS_ENABLED(CONFIG_FOO);

but that said, I'm still not particularly convinced that there are a 
significant number of users in a position to build and install a custom 
kernel but not edit /etc/default/grub, and who really value the 
combination of less performance than passthrough with less isolation 
than strict.

It's also not necessarily obvious to the user how this interacts with 
IOMMU_DEFAULT_PASSTHROUGH, so if we really do go down this route, maybe 
it would be better to refactor the whole lot into a single selection of 
something like IOMMU_DEFAULT_MODE anyway.

Robin.

> 
>   struct iommu_callback_data {
>   	const struct iommu_ops *ops;
> --
> 1.8.3
> 
> 

Re: [PATCH v2 1/1] iommu: Add config option to set lazy mode as default

[Leizhen (ThunderTown)]

On 2019/3/28 3:18, Robin Murphy wrote:
> On 27/03/2019 15:00, Zhen Lei wrote:
>> This allows the default behaviour to be controlled by a kernel config
>> option instead of changing the command line for the kernel to include
>> "iommu.strict=0" on ARM64 where this is desired.
>>
>> This is similar to CONFIG_IOMMU_DEFAULT_PASSTHROUGH.
>>
>> Note: At present, intel_iommu, amd_iommu and s390_iommu use lazy mode as
>> default, so there is no need to add code for them.
> 
> That seems a bit self-contradictory - if there's a real need for TLB flush behaviour to be statically configurable and not command-line-based then why should users of other architectures be exempt?
Yes，you're right. I will make this configuration to be effective for other architectures.

BTW, can you give some opinion about below patches?
https://patchwork.kernel.org/patch/10857601/

> 
>> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
>> ---
>>   drivers/iommu/Kconfig | 14 ++++++++++++++
>>   drivers/iommu/iommu.c |  5 +++++
>>   2 files changed, 19 insertions(+)
>>
>> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
>> index 6f07f3b21816c64..5daa110d0e83a07 100644
>> --- a/drivers/iommu/Kconfig
>> +++ b/drivers/iommu/Kconfig
>> @@ -85,6 +85,20 @@ config IOMMU_DEFAULT_PASSTHROUGH
>>
>>         If unsure, say N here.
>>
>> +config IOMMU_DMA_DEFAULT_LAZY_MODE
>> +    bool "IOMMU DMA use lazy mode to flush IOTLB and free IOVA"
>> +    depends on IOMMU_API
>> +    help
>> +      Support lazy mode, where for every IOMMU DMA unmap operation, the
>> +      flush operation of IOTLB and the free operation of IOVA are deferred.
>> +      They are only guaranteed to be done before the related IOVA will be
>> +      reused. Removing the need to pass in kernel parameters through
>> +      command line. For example, iommu.strict=0 on ARM64. If this is
>> +      enabled, you can still disable with kernel parameters, such as
>> +      iommu.strict=1 depending on the architecture.
>> +
>> +      If unsure, say N here.
>> +
>>   config OF_IOMMU
>>          def_bool y
>>          depends on OF && IOMMU_API
>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
>> index 33a982e33716369..5acb98e79b5b32d 100644
>> --- a/drivers/iommu/iommu.c
>> +++ b/drivers/iommu/iommu.c
>> @@ -43,7 +43,12 @@
>>   #else
>>   static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
>>   #endif
>> +
>> +#ifdef CONFIG_IOMMU_DMA_DEFAULT_LAZY_MODE
>> +static bool iommu_dma_strict __read_mostly;
>> +#else
>>   static bool iommu_dma_strict __read_mostly = true;
>> +#endif
> 
> For a straightforward boolean, you can simply do:
> 
>     static bool foo = IS_ENABLED(CONFIG_FOO);
> 
> but that said, I'm still not particularly convinced that there are a significant number of users in a position to build and install a custom kernel but not edit /etc/default/grub, and who really value the combination of less performance than passthrough with less isolation than strict.
> 
> It's also not necessarily obvious to the user how this interacts with IOMMU_DEFAULT_PASSTHROUGH, so if we really do go down this route, maybe it would be better to refactor the whole lot into a single selection of something like IOMMU_DEFAULT_MODE anyway.
> 
> Robin.
> 
>>
>>   struct iommu_callback_data {
>>       const struct iommu_ops *ops;
>> -- 
>> 1.8.3
>>
>>
> 
> .
> 

-- 
Thanks!
BestRegards