Re: [PATCH v2] hpet: Fix missing '=' character in the __setup() code of hpet_mmap_enable

[Prarit Bhargava]

On 1/6/19 9:09 AM, Buland Singh wrote:
> On 12/21/18 6:16 PM, Greg KH wrote:
>> On Fri, Dec 21, 2018 at 05:48:38PM +0530, Buland Singh wrote:
>>> On 12/20/18 7:39 PM, Greg KH wrote:
>>>> On Thu, Dec 20, 2018 at 07:12:55PM +0530, Buland Singh wrote:
>>>>> On 12/20/18 5:59 PM, Greg KH wrote:
>>>>>> On Thu, Dec 20, 2018 at 05:35:24PM +0530, Buland Singh wrote:
>>>>>>> Commit '3d035f580699 ("drivers/char/hpet.c: allow user controlled mmap for
>>>>>>> user processes")' introduced a new kernel command line parameter hpet_mmap,
>>>>>>> that is required to expose the memory map of the HPET registers to
>>>>>>> user-space. Unfortunately the kernel command line parameter 'hpet_mmap' is
>>>>>>> broken and never takes effect due to missing '=' character in the __setup()
>>>>>>> code of hpet_mmap_enable.
>>>>>>>
>>>>>>> Before this patch:
>>>>>>>
>>>>>>> dmesg output with the kernel command line parameter hpet_mmap=1
>>>>>>>
>>>>>>> [ÂÂÂ 0.204152] HPET mmap disabled
>>>>>>>
>>>>>>> dmesg output with the kernel command line parameter hpet_mmap=0
>>>>>>>
>>>>>>> [ÂÂÂ 0.204192] HPET mmap disabled
>>>>>>>
>>>>>>> After this patch:
>>>>>>>
>>>>>>> dmesg output with the kernel command line parameter hpet_mmap=1
>>>>>>>
>>>>>>> [ÂÂÂ 0.203945] HPET mmap enabled
>>>>>>>
>>>>>>> dmesg output with the kernel command line parameter hpet_mmap=0
>>>>>>>
>>>>>>> [ÂÂÂ 0.204652] HPET mmap disabled
>>>>>>>
>>>>>>> Fixes: 3d035f580699 ("drivers/char/hpet.c: allow user controlled mmap for
>>>>>>> user processes")
>>>>>>> Signed-off-by: Buland Singh <bsingh@xxxxxxxxxx>
>>>>>>> ---
>>>>>>> ÂÂÂ drivers/char/hpet.c | 2 +-
>>>>>>> ÂÂÂ 1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/drivers/char/hpet.c b/drivers/char/hpet.c
>>>>>>> index 4a22b4b41aef..9bffcd37cc7b 100644
>>>>>>> --- a/drivers/char/hpet.c
>>>>>>> +++ b/drivers/char/hpet.c
>>>>>>> @@ -377,7 +377,7 @@ static __init int hpet_mmap_enable(char *str)
>>>>>>> ÂÂÂÂÂÂÂ pr_info("HPET mmap %s\n", hpet_mmap_enabled ? "enabled" :
>>>>>>> "disabled");
>>>>>>> ÂÂÂÂÂÂÂ return 1;
>>>>>>> ÂÂÂ }
>>>>>>> -__setup("hpet_mmap", hpet_mmap_enable);
>>>>>>> +__setup("hpet_mmap=", hpet_mmap_enable);
>>>>>
>>>>> Hello Greag,
>>>>>
>>>>>> This has _never_ worked? Since 3.13?
>>>>>
>>>>> Yes, that's true :)
>>>>>
>>>>>> Why not just remove the thing as it is obvious no one actually has ever
>>>>>> used it. > That would make the code even simpler :)
>>>>>
>>>>> Data Plane Development Kit (DPDK)[1] provides API that requires the
>>>>> CONFIG_HPET_MMAP
>>>>> kernel configuration option to be enabled[2]. Some end users might want to
>>>>> use the
>>>>> HPET MMAP functionality within the application.
>>>>
>>>> But, obviously, they really don't need to do that from the kernel
>>>> command line as no one has ever noticed this didn't work :)
>>>>
>>>> Also, that page:
>>>>
>>>>> [2] https://doc.dpdk.org/guides-18.08/linux_gsg/enable_func.html
>>>>
>>>> Does not say to use this command line option either. So if no one has
>>>> ever used it, please, let us just delete it.
>>>>
>>>> thanks,
>>>>
>>>> greg k-h
>>>>
>>>
>>> Hello Greg,
>>> It is better to allow a user to 'enable/disable' the HPET mmap from the
>>> kernel command line as per the requirement rather than recompiling the
>>> kernel to 'enable/disable' this functionality.
>>
>> It might be "nice" but given that no one has noticed that this has never
>> worked, for all of the years that this has been present, that means to
>> me that no one has ever tried it because they really do not "need" it.
>>
>>> Also as per the description in the initial patch (commit 3d035f58), the
>>> 'CONFIG_HPET_MMAP' Kconfig option has a security risk involved. Hence,
>>> keeping the CONFIG_HPET_MMAP_DEFAULT (disabled) and allowing a user to
>>> alter the default behavior using the kernel command line parameter
>>> hpet_mmap is a better solution.
>>
>> Again, because no one has ever noticed that this was not working, why
>> not just rip it out until someone speaks up as to why they have to have
>> this feature and the other alternatives do not work for them?
>>

The feature came about when two customers asked for it.  The customers are
willing to live with the increased security risk with fast HPET MMAP access.

The only place I've seen a use request for this is in RHEL, and the bug was
likely uncovered during RHEL8 beta testing which explains why no one has used
the upstream code. ie) RHEL7 does contain an "=" sign, upstream/RHEL8 do not.

P.

>> thanks,
>>
>> greg k-h
>>
> 
> Hello,
> 
> Adding Prarit to cc list for his opinion.
> 
> Regards,
> -- 
> Buland Singh
> 
> 
>