[PATCH] drm/amd/amdgpu: add support for iova_to_phys to replace TTM trace (v4)

[Christian König]

Am 19.09.2017 um 13:20 schrieb Tom St Denis:
> Signed-off-by: Tom St Denis <tom.stdenis at amd.com>
>
> (v2): Add domain to iova debugfs
> (v3): Add true read/write methods to access system memory of pages
>        mapped to the device
> (v4): Move get_domain call out of loop and return on error
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 114 ++++++++++++++++++++++++++++++++
>   1 file changed, 114 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 50d20903de4f..71a2fdc91a85 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -43,6 +43,7 @@
>   #include <linux/swap.h>
>   #include <linux/pagemap.h>
>   #include <linux/debugfs.h>
> +#include <linux/iommu.h>
>   #include "amdgpu.h"
>   #include "amdgpu_trace.h"
>   #include "bif/bif_4_1_d.h"
> @@ -1810,6 +1811,118 @@ static const struct file_operations amdgpu_ttm_gtt_fops = {
>   #endif
>   
>   
> +static void *transform_page(uint64_t phys)
> +{
> +	if (PageHighMem(pfn_to_page(PFN_DOWN(phys))))
> +		return kmap(pfn_to_page(PFN_DOWN(phys)));
> +	else
> +		return __va(phys);
> +}
> +
> +static void untransform_page(uint64_t phys)
> +{
> +	if (PageHighMem(pfn_to_page(PFN_DOWN(phys))))
> +		return kunmap(pfn_to_page(PFN_DOWN(phys)));
> +}

I mentioned that in the previous patch as well, why not using 
kmap()/kunmap() here directly?

See kmap() for x86 for example:
> void  *kmap <http://elixir.free-electrons.com/linux/v4.2/ident/kmap>(struct  page  *page)
> {
> 	might_sleep 
> <http://elixir.free-electrons.com/linux/v4.2/ident/might_sleep>();
> 	if  (!PageHighMem 
> <http://elixir.free-electrons.com/linux/v4.2/ident/PageHighMem>(page))
> 		return  page_address 
> <http://elixir.free-electrons.com/linux/v4.2/ident/page_address>(page);
> 	return  kmap_high 
> <http://elixir.free-electrons.com/linux/v4.2/ident/kmap_high>(page);
> }

Both normal as well as highmem are correctly handled here.

Regards,
Christian.

> +
> +static ssize_t amdgpu_iova_to_phys_read(struct file *f, char __user *buf,
> +				   size_t size, loff_t *pos)
> +{
> +	struct amdgpu_device *adev = file_inode(f)->i_private;
> +	ssize_t result, n;
> +	int r;
> +	uint64_t phys;
> +	void *ptr;
> +	struct iommu_domain *dom;
> +
> +	dom = iommu_get_domain_for_dev(adev->dev);
> +	if (!dom)
> +		return -EFAULT;
> +
> +	result = 0;
> +	while (size) {
> +		// get physical address and map
> +		phys = iommu_iova_to_phys(dom, *pos);
> +
> +		// copy upto one page
> +		if (size > PAGE_SIZE)
> +			n = PAGE_SIZE;
> +		else
> +			n = size;
> +
> +		// to end of the page
> +		if (((*pos & (PAGE_SIZE - 1)) + n) >= PAGE_SIZE)
> +			n = PAGE_SIZE - (*pos & (PAGE_SIZE - 1));
> +
> +		ptr = transform_page(phys);
> +		if (!ptr)
> +			return -EFAULT;
> +
> +		r = copy_to_user(buf, ptr, n);
> +		untransform_page(phys);
> +		if (r)
> +			return -EFAULT;
> +
> +		*pos += n;
> +		size -= n;
> +		result += n;
> +	}
> +
> +	return result;
> +}
> +
> +static ssize_t amdgpu_iova_to_phys_write(struct file *f, const char __user *buf,
> +				   size_t size, loff_t *pos)
> +{
> +	struct amdgpu_device *adev = file_inode(f)->i_private;
> +	ssize_t result, n;
> +	int r;
> +	uint64_t phys;
> +	void *ptr;
> +	struct iommu_domain *dom;
> +
> +	dom = iommu_get_domain_for_dev(adev->dev);
> +	if (!dom)
> +		return -EFAULT;
> +
> +	result = 0;
> +	while (size) {
> +		// get physical address and map
> +		phys = iommu_iova_to_phys(dom, *pos);
> +
> +		// copy upto one page
> +		if (size > PAGE_SIZE)
> +			n = PAGE_SIZE;
> +		else
> +			n = size;
> +
> +		// to end of the page
> +		if (((*pos & (PAGE_SIZE - 1)) + n) >= PAGE_SIZE)
> +			n = PAGE_SIZE - (*pos & (PAGE_SIZE - 1));
> +
> +		ptr = transform_page(phys);
> +		if (!ptr)
> +			return -EFAULT;
> +
> +		r = copy_from_user(ptr, buf, n);
> +		untransform_page(phys);
> +		if (r)
> +			return -EFAULT;
> +
> +		*pos += n;
> +		size -= n;
> +		result += n;
> +	}
> +
> +	return result;
> +}
> +
> +static const struct file_operations amdgpu_ttm_iova_fops = {
> +	.owner = THIS_MODULE,
> +	.read = amdgpu_iova_to_phys_read,
> +	.write = amdgpu_iova_to_phys_write,
> +	.llseek = default_llseek
> +};
>   
>   static const struct {
>   	char *name;
> @@ -1820,6 +1933,7 @@ static const struct {
>   #ifdef CONFIG_DRM_AMDGPU_GART_DEBUGFS
>   	{ "amdgpu_gtt", &amdgpu_ttm_gtt_fops, TTM_PL_TT },
>   #endif
> +	{ "amdgpu_iova", &amdgpu_ttm_iova_fops, TTM_PL_SYSTEM },
>   };
>   
>   #endif


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20170919/77232c4e/attachment.html>