[PATCH] modify the IO_TLB_SEGSIZE to configtable as flexible requirement about IOMMU/SW-IOMMU.

[PATCH] modify the IO_TLB_SEGSIZE to configtable as flexible requirement about IOMMU/SW-IOMMU.

[xiaomin1]

The maximum of SW-IOMMU is limited to 2^11*128 = 256K.
While in different platform and different requirements this seems improper.
So modify the IO_TLB_SEGSIZE to configtable is make sense.
Example:
If 1M bytes are requied. There has an error like.
[ 33.293120] DMA: Out of SW-IOMMU space for 1048576 bytes at device gadget

Signed-off-by: Chuansheng Liu <chuansheng.liu@intel.com>
Signed-off-by: Zhang Dongxing <dongxing.zhang@intel.com>
Signed-off-by: xiaomin1 <xiaoming.wang@intel.com>
---
 include/linux/swiotlb.h |    2 +-
 lib/Kconfig             |    4 ++++
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
index e7a018e..5ab10b9 100644
--- a/include/linux/swiotlb.h
+++ b/include/linux/swiotlb.h
@@ -14,7 +14,7 @@ extern int swiotlb_force;
  * must be a power of 2.  What is the appropriate value ?
  * The complexity of {map,unmap}_single is linearly dependent on this value.
  */
-#define IO_TLB_SEGSIZE	128
+#define IO_TLB_SEGSIZE	CONFIG_DEFAULT_IO_TLB_SEGSIZE
 
 /*
  * log of the size of each IO TLB slab.  The number of slabs is command line
diff --git a/lib/Kconfig b/lib/Kconfig
index 54cf309..76f1108 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -518,4 +518,8 @@ source "lib/fonts/Kconfig"
 config ARCH_HAS_SG_CHAIN
 	def_bool n
 
+config DEFAULT_IO_TLB_SEGSIZE
+	int
+	default 128
+
 endmenu
-- 
1.7.9.5

Re: [PATCH] modify the IO_TLB_SEGSIZE to configtable as flexible requirement about IOMMU/SW-IOMMU.

[Paul Bolle]

(For some reason the "Date:" header of this message is 16 hours into the
future. Ie, it claims to be for (UTC) 2015-01-26, 19.30 but actually was
probably sent around (UTC) 2015-01-26, 03.30. This made this message
stick at the top of the mail folder it ended up in. And that was about
the only reason why I even scanned it.)

On Tue, 2015-01-27 at 03:30 +0800, xiaomin1 wrote:
> The maximum of SW-IOMMU is limited to 2^11*128 = 256K.
> While in different platform and different requirements this seems improper.
> So modify the IO_TLB_SEGSIZE to configtable is make sense.
                                  configurable?

Ditto for the summary.

> Example:
> If 1M bytes are requied. There has an error like.
                  required one will run into an error like:

or something similar.

> [ 33.293120] DMA: Out of SW-IOMMU space for 1048576 bytes at device gadget

So in order to solve that and use 1M this new option needs to be set at
512?

> Signed-off-by: Chuansheng Liu <chuansheng.liu@intel.com>
> Signed-off-by: Zhang Dongxing <dongxing.zhang@intel.com>
> Signed-off-by: xiaomin1 <xiaoming.wang@intel.com>
> ---
>  include/linux/swiotlb.h |    2 +-
>  lib/Kconfig             |    4 ++++
>  2 files changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/swiotlb.h b/include/linux/swiotlb.h
> index e7a018e..5ab10b9 100644
> --- a/include/linux/swiotlb.h
> +++ b/include/linux/swiotlb.h
> @@ -14,7 +14,7 @@ extern int swiotlb_force;
>   * must be a power of 2.  What is the appropriate value ?
>   * The complexity of {map,unmap}_single is linearly dependent on this value.
>   */
> -#define IO_TLB_SEGSIZE	128
> +#define IO_TLB_SEGSIZE	CONFIG_DEFAULT_IO_TLB_SEGSIZE
>  
>  /*
>   * log of the size of each IO TLB slab.  The number of slabs is command line
> diff --git a/lib/Kconfig b/lib/Kconfig
> index 54cf309..76f1108 100644
> --- a/lib/Kconfig
> +++ b/lib/Kconfig
> @@ -518,4 +518,8 @@ source "lib/fonts/Kconfig"
>  config ARCH_HAS_SG_CHAIN
>  	def_bool n
>  
> +config DEFAULT_IO_TLB_SEGSIZE
> +	int

No "prompt", so to change this value one needs to edit the .config
manually. Is that intended?

> +	default 128

According to include/linux/swiotlb.h this needs to be a power of 2.
Shouldn't that be explained here too? Ditto the 2K unit size. And should
a range of useful values be specified?

(These are not rhetorical questions: I know nothing about all this. This
also means I might just not be part of the intended audience for this
option.)

>  endmenu


Paul Bolle