Re: [PATCH v3 1/4] VFIO: platform: add reset struct and lookup table

[Eric Auger <eric.auger@linaro.org>]

Hi Alex,
On 06/12/2015 04:19 PM, Alex Williamson wrote:
> On Fri, 2015-06-12 at 15:41 +0200, Eric Auger wrote:
>> Hi Alex,
>> On 06/11/2015 11:11 PM, Alex Williamson wrote:
>>> On Thu, 2015-06-11 at 14:08 +0200, Eric Auger wrote:
>>>> This patch introduces the vfio_platform_reset_combo struct that
>>>> stores all the information useful to handle the reset modality:
>>>> compat string, name of the reset function, name of the module that
>>>> implements the reset function. A lookup table of such structures
>>>> is added, currently containing a single sentinel element. A new
>>>> type field is added in vfio_platform_device to store what kind of
>>>> reset is associated to the device, if any.
>>>
>>> The commit log no longer matches the code.
>> yes indeed missed that.
>>>
>>> The only other thing I'm struggling with in this series is that in 0/4
>>> you suggest that the reset modules can be in-kernel or external, but
>>> we're making a static list here, so there's really no support for
>>> random, user-provided reset modules.  So are we missing the mark on the
>>> requirements?
>>
>> Well personally I do not have a MUST HAVE requirement for that feature.
>> My main requirement was to find a way to stop DMA/IRQ transfers
>> programmed by a previous VM potentially jeopardizing the integrity of a
>> second VM.
>>>
>>> One way I thought you could achieve your requirement would be if we did
>>> away with the lookup table and looked for the module and function using
>>> a pre-defined transform on the compat ID.  For instance, a compat ID of
>>> "calxeda,hb-xgmac" would automatically request a module named
>>> "vfio-platform-reset-calxedahb-xgmac" and look for a symbol of the same
>>> name for the reset function (I wonder if we can actually have a module
>>> and symbol of the same name).
>> I just tried and it works
>>   It seems fairly safe since an external
>>> module would need to be explicitly placed in the search path for the
>>> userspace module loader.
>> we talked together with Christoffer about such a technique at the very
>> beginning and he was not very fond of it, advising an approach using a
>> defined API.
>>>
>>> Otherwise the table would need to become a list, the external module
>>> would need to be manually loaded, and the module_init() would need to
>>> register an entry on that list.
>> Isn't it what I attempted to do in v2 or do you mean something else?
> 
> IIRC, v2 required all modules to be manually loaded, which I think was
> bound to be a support issue.
  In-kernel modules should definitely have
> the benefit of auto-loading,
yes you're right! thanks for your clarification.

 but if we want to support external modules,
> I'm just hypothesizing that we either need to standardize naming so that
> we can ask for them or require registration.  In-kernel modules could
> still effectively be pre-registered and requested on-demand.
>> Having a whitelist brings the benefit to know which devices really are
>> used with vfio-platform (besides some devices may not need any reset
>> module). I don't know yet whether this is something that may slow down
>> the adoption.
>>
>> Personally I would keep it as is for now and see how it gets used and
>> whether users complain, ...
>>
>> If you agree I will respin just reworking the commit message.
> 
> If there's really no external module requirement, then your current
> approach is fine.  We're not creating anything here that can't later be
> modified.  Thanks,
OK. I will send v4 with updated commit messages early next week.

Have a nice week end.

Eric

> 
> Alex
> 
> 
>>>> ---
>>>>
>>>> v2 -> v3:
>>>> - add const in struct vfio_platform_reset_combo
>>>> - remove enum vfio_platform_reset_type
>>>>
>>>> v2: creation
>>>> ---
>>>>  drivers/vfio/platform/vfio_platform_common.c  | 3 +++
>>>>  drivers/vfio/platform/vfio_platform_private.h | 6 ++++++
>>>>  2 files changed, 9 insertions(+)
>>>>
>>>> diff --git a/drivers/vfio/platform/vfio_platform_common.c b/drivers/vfio/platform/vfio_platform_common.c
>>>> index abcff7a..611597e 100644
>>>> --- a/drivers/vfio/platform/vfio_platform_common.c
>>>> +++ b/drivers/vfio/platform/vfio_platform_common.c
>>>> @@ -25,6 +25,9 @@
>>>>  
>>>>  static DEFINE_MUTEX(driver_lock);
>>>>  
>>>> +static const struct vfio_platform_reset_combo reset_lookup_table[] = {
>>>> +};
>>>> +
>>>>  static int vfio_platform_regions_init(struct vfio_platform_device *vdev)
>>>>  {
>>>>  	int cnt = 0, i;
>>>> diff --git a/drivers/vfio/platform/vfio_platform_private.h b/drivers/vfio/platform/vfio_platform_private.h
>>>> index 5d31e04..9e37b9f 100644
>>>> --- a/drivers/vfio/platform/vfio_platform_private.h
>>>> +++ b/drivers/vfio/platform/vfio_platform_private.h
>>>> @@ -69,6 +69,12 @@ struct vfio_platform_device {
>>>>  	int	(*get_irq)(struct vfio_platform_device *vdev, int i);
>>>>  };
>>>>  
>>>> +struct vfio_platform_reset_combo {
>>>> +	const char *compat;
>>>> +	const char *reset_function_name;
>>>> +	const char *module_name;
>>>> +};
>>>> +
>>>>  extern int vfio_platform_probe_common(struct vfio_platform_device *vdev,
>>>>  				      struct device *dev);
>>>>  extern struct vfio_platform_device *vfio_platform_remove_common
>>>
>>>
>>>
>>
> 
> 
>