On 18/09/17 08:52 AM, Christian König wrote:
> Am 18.09.2017 um 14:35 schrieb Tom St Denis:
>> Signed-off-by: Tom St Denis <tom.stdenis at amd.com>
>> ---
>> Â drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 32
>> ++++++++++++++++++++++++++++++++
>> Â 1 file changed, 32 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> index 7848ffa99eb4..b4c298925e2a 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>> @@ -43,6 +43,7 @@
>> Â #include <linux/swap.h>
>> Â #include <linux/pagemap.h>
>> Â #include <linux/debugfs.h>
>> +#include <linux/iommu.h>
>> Â #include "amdgpu.h"
>> Â #include "amdgpu_trace.h"
>> Â #include "bif/bif_4_1_d.h"
>> @@ -1810,6 +1811,36 @@ static const struct file_operations
>> amdgpu_ttm_gtt_fops = {
>> Â #endif
>> +static ssize_t amdgpu_iova_to_phys_read(struct file *f, char __user
>> *buf,
>> +Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â size_t size, loff_t *pos)
>> +{
>> +Â Â Â struct amdgpu_device *adev = file_inode(f)->i_private;
>> +Â Â Â int r;
>> +Â Â Â uint64_t phys;
>> +
>> +Â Â Â // always return 8 bytes
>> +Â Â Â if (size != 8)
>> +Â Â Â Â Â Â Â return -EINVAL;
>> +
>> +Â Â Â // only accept page addresses
>> +Â Â Â if (*pos & 0xFFF)
>> +Â Â Â Â Â Â Â return -EINVAL;
>> +
>> +
>> +Â Â Â phys = iommu_iova_to_phys(iommu_get_domain_for_dev(adev->dev),
>> *pos);
>
> Well how about adding directly read/write support for the page behind
> the address?
>
> This way we won't need to fiddle with /dev/mem any more either.
Given the flak we got from requesting this from the iommu team I'm
worried that might not be appreciated by the community (even though we
maintain this part of the tree) and hurt our abilities to upstream.
I agree that adding a read/write method would be better though since we
don't need config changes of /dev/fmem anymore.
Tom
