[PATCH libdrm] amdgpu: add VM test to exercise max/min address space

[Zhang, Jerry]

在 2018年10月29日，16:40，Koenig, Christian <Christian.Koenig at amd.com> 写道：
> 
> Am 29.10.18 um 09:13 schrieb Zhang, Jerry:
>> 在 2018年10月29日，15:32，Christian König <ckoenig.leichtzumerken at gmail.com> 写道：
>>> Am 29.10.18 um 06:50 schrieb Zhang, Jerry:
>>>> 在 2018年10月26日，18:59，Christian König <ckoenig.leichtzumerken at gmail.com> 写道：
>>>>> Make sure the kernel doesn't crash if we map something at the minimum/maximum address.
>>>>> 
>>>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>>>> ---
>>>>> tests/amdgpu/vm_tests.c | 45 ++++++++++++++++++++++++++++++++++++++++-
>>>>> 1 file changed, 44 insertions(+), 1 deletion(-)
>>>>> 
>>>>> diff --git a/tests/amdgpu/vm_tests.c b/tests/amdgpu/vm_tests.c
>>>>> index 7b6dc5d6..bbdeef4d 100644
>>>>> --- a/tests/amdgpu/vm_tests.c
>>>>> +++ b/tests/amdgpu/vm_tests.c
>>>>> @@ -31,8 +31,8 @@ static  amdgpu_device_handle device_handle;
>>>>> static  uint32_t  major_version;
>>>>> static  uint32_t  minor_version;
>>>>> 
>>>>> -
>>>>> static void amdgpu_vmid_reserve_test(void);
>>>>> +static void amdgpu_vm_mapping_test(void);
>>>>> 
>>>>> CU_BOOL suite_vm_tests_enable(void)
>>>>> {
>>>>> @@ -84,6 +84,7 @@ int suite_vm_tests_clean(void)
>>>>> 
>>>>> CU_TestInfo vm_tests[] = {
>>>>> 	{ "resere vmid test",  amdgpu_vmid_reserve_test },
>>>>> +	{ "vm mapping test",  amdgpu_vm_mapping_test },
>>>>> 	CU_TEST_INFO_NULL,
>>>>> };
>>>>> 
>>>>> @@ -167,3 +168,45 @@ static void amdgpu_vmid_reserve_test(void)
>>>>> 	r = amdgpu_cs_ctx_free(context_handle);
>>>>> 	CU_ASSERT_EQUAL(r, 0);
>>>>> }
>>>>> +
>>>>> +static void amdgpu_vm_mapping_test(void)
>>>>> +{
>>>>> +	struct amdgpu_bo_alloc_request req = {0};
>>>>> +	struct drm_amdgpu_info_device dev_info;
>>>>> +	const uint64_t size = 4096;
>>>>> +	amdgpu_bo_handle buf;
>>>>> +	uint64_t addr;
>>>>> +	int r;
>>>>> +
>>>>> +	req.alloc_size = size;
>>>>> +	req.phys_alignment = 0;
>>>>> +	req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
>>>>> +	req.flags = 0;
>>>>> +
>>>>> +	r = amdgpu_bo_alloc(device_handle, &req, &buf);
>>>>> +	CU_ASSERT_EQUAL(r, 0);
>>>>> +
>>>>> +	r = amdgpu_query_info(device_handle, AMDGPU_INFO_DEV_INFO,
>>>>> +			      sizeof(dev_info), &dev_info);
>>>>> +	CU_ASSERT_EQUAL(r, 0);
>>>>> +
>>>>> +	addr = dev_info.virtual_address_offset;
>>>>> +	r = amdgpu_bo_va_op(buf, 0, size, addr, 0, AMDGPU_VA_OP_MAP);
>>>>> +	CU_ASSERT_EQUAL(r, 0);
>>>> Please confirm:
>>>> 
>>>> We may need to unmap the VA before bo free, although this VA range is unlikely to be used by other test cases.
>>> Unmapping is unnecessary, VA mappings of BOs are automatically removed when the BO is freed.
>> Ah, correct. Feel free to add
>> Reviewed-by: Junwei Zhang <Jerry.Zhang at amd.com>
> 
> Thanks.
> 
>> 
>>>> BTW, is it a chance in practice that a process may map different VA ranges to the same bo?
>>> Yeah, that is used quite extensively with PRTs.
>> PRT may use different ranges in entire BO.
>> What I actually mean is the same bo range, e.g. 1 page of BO starting from offset 0, mapping to different VA ranges, like what the test does now.
> 
> Good point, but that should also be used by PRT mappings.
> 
> E.g. what we do with PRT is to have a BO filled with parts of a texture. 
> This BO is then mapped once completely form start till end

It sounds MAP with map op

> and once page by page into the sparse populated PRT texture.

It sounds REPLACE with map op

Is there a case that map(MAP) the same bo range by different VA ranges?
work as a physical ram range is mapped by different virtual address.


Jerry

> 
> Christian.
> 
>> 
>> Jerry
>> 
>>> Regards,
>>> Christian.
>>> 
>>>> Regards,
>>>> Jerry
>>>> 
>>>>> +
>>>>> +	addr = dev_info.virtual_address_max - size;
>>>>> +	r = amdgpu_bo_va_op(buf, 0, size, addr, 0, AMDGPU_VA_OP_MAP);
>>>>> +	CU_ASSERT_EQUAL(r, 0);
>>>>> +
>>>>> +	if (dev_info.high_va_offset) {
>>>>> +		addr = dev_info.high_va_offset;
>>>>> +		r = amdgpu_bo_va_op(buf, 0, size, addr, 0, AMDGPU_VA_OP_MAP);
>>>>> +		CU_ASSERT_EQUAL(r, 0);
>>>>> +
>>>>> +		addr = dev_info.high_va_max - size;
>>>>> +		r = amdgpu_bo_va_op(buf, 0, size, addr, 0, AMDGPU_VA_OP_MAP);
>>>>> +		CU_ASSERT_EQUAL(r, 0);
>>>>> +	}
>>>>> +
>>>>> +	amdgpu_bo_free(buf);
>>>>> +}
>>>>> -- 
>>>>> 2.17.1
>>>>> 
>>>>> _______________________________________________
>>>>> amd-gfx mailing list
>>>>> amd-gfx at lists.freedesktop.org
>>>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx