Re: [PATCH] iommu/amd: Track when amd_iommu_v2 init is complete

[Oded Gabbay]

On 12/20/2014 10:12 PM, Oded Gabbay wrote:
> This patch adds a new exported function to amd_iommu_v2, which returns 1 if the
> amd_iommu_v2 initialization function has completed, and 0 otherwise.
> 
> This is necessary for the case when amd_iommu_v2 is compiled inside the kernel
> image (not as module) and another module (e.g. amdkfd), which is also compiled
> inside the kernel image, is dependent on amd_iommu_v2 functionality.
> 
> In that case, there is no mechanism in the kernel that enforces the order of
> loading between the two modules. Therefore, If the amd_iommu_v2 is loaded
> _after_ the other module, and the other module calls one of
> amd_iommu_v2 exported functions _before_ amd_iommu_v2 is loaded, than that
> function will fail, and as a result, the calling module may fail as well.
> 
> Note that when the two modules are compiled as modules, this situation can't
> occur as the kernel enforces the order of loading.
> 
> Signed-off-by: Oded Gabbay <oded.gabbay@xxxxxxx>
> ---
>  drivers/iommu/amd_iommu_v2.c | 11 +++++++++++
>  include/linux/amd-iommu.h    |  2 ++
>  2 files changed, 13 insertions(+)
> 
> diff --git a/drivers/iommu/amd_iommu_v2.c b/drivers/iommu/amd_iommu_v2.c
> index 1e6360e..2456f18 100644
> --- a/drivers/iommu/amd_iommu_v2.c
> +++ b/drivers/iommu/amd_iommu_v2.c
> @@ -895,6 +895,14 @@ out_unlock:
>  }
>  EXPORT_SYMBOL(amd_iommu_set_invalidate_ctx_cb);
>  
> +static int iommu_v2_init_completed;
> +
> +int amd_iommu_v2_is_init_completed(void)
> +{
> +	return iommu_v2_init_completed;
> +}
> +EXPORT_SYMBOL(amd_iommu_v2_is_init_completed);
> +
>  static int __init amd_iommu_v2_init(void)
>  {
>  	int ret;
> @@ -903,6 +911,7 @@ static int __init amd_iommu_v2_init(void)
>  
>  	if (!amd_iommu_v2_supported()) {
>  		pr_info("AMD IOMMUv2 functionality not available on this system\n");
> +		iommu_v2_init_completed = 1;
>  		/*
>  		 * Load anyway to provide the symbols to other modules
>  		 * which may use AMD IOMMUv2 optionally.
> @@ -919,6 +928,8 @@ static int __init amd_iommu_v2_init(void)
>  
>  	amd_iommu_register_ppr_notifier(&ppr_nb);
>  
> +	iommu_v2_init_completed = 1;
> +
>  	return 0;
>  
>  out:
> diff --git a/include/linux/amd-iommu.h b/include/linux/amd-iommu.h
> index 2b08e79..115c03a 100644
> --- a/include/linux/amd-iommu.h
> +++ b/include/linux/amd-iommu.h
> @@ -169,6 +169,8 @@ typedef void (*amd_iommu_invalidate_ctx)(struct pci_dev *pdev, int pasid);
>  extern int amd_iommu_set_invalidate_ctx_cb(struct pci_dev *pdev,
>  					   amd_iommu_invalidate_ctx cb);
>  
> +extern int amd_iommu_v2_is_init_completed(void);
> +
>  #else
>  
>  static inline int amd_iommu_detect(void) { return -ENODEV; }
> 

Hi Joerg,
This patch follows our IRC conversations from the last week.
Last update from me was that using subsys_initcall instead of module_init
does enforce amd_iommu_v2 to load _before_ amdkfd (apparently I didn't check
it correctly the first time). However, this won't help us because:
1. I don't know what effect this has when amd_iommu_v2 is compiled as module.
2. amd_iommu_v2 depends on amd_iommu which is not subsys, so amd_iommu_v2
init function fails when it is called this early, and thus, amd_iommu_v2
doesn't work.

I want to fix this situation in rc1/2, so I'm sending the complementary
patches to dri-devel in the meantime.

	Oded
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/