linux-kernel - Re: [PATCH v1 1/1] platform/x86: intel_pmc

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <8c0966ba-0b83-839a-9198-51aae2ecd005@linux.intel.com>
Date:   Thu, 16 Mar 2017 14:05:10 -0700
From:   sathyanarayanan kuppuswamy 
        <sathyanarayanan.kuppuswamy@...ux.intel.com>
To:     Rajneesh Bhardwaj <rajneesh.bhardwaj@...el.com>
Cc:     andy@...radead.org, qipeng.zha@...el.com, dvhart@...radead.org,
        platform-driver-x86@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1 1/1] platform/x86: intel_pmc_ipc: fix io mem mapping
 size



On 03/16/2017 12:20 PM, Rajneesh Bhardwaj wrote:
> On Thu, Mar 16, 2017 at 11:50:16AM -0700, sathyanarayanan kuppuswamy wrote:
>> Hi,
>>
>>
>> On 03/16/2017 07:52 AM, Rajneesh Bhardwaj wrote:
>>> On Wed, Mar 15, 2017 at 08:32:53PM -0700, Kuppuswamy Sathyanarayanan wrote:
>>>> Mapping entire GCR mem region in this driver creates
>>>> mem region request conflict in sub devices that depend
>>>> on PMC. This creates driver probe failure in devices like
>>>> iTC0_wdt and telemetry device.
>>> While this patch might fix the issue for now but IMHO its not taking the
>>> right approch. I guess we need some guidance here from the maintainers but
>> Agreed. I thought about this problem and I know this solution does not
>> scale well. Other way is to expose an API for GCR access and use it on
>> PMC dependent drivers. In this case, we should also add GCR register
>> address macros to PMC header file and this might need change to PMC header
>> file each time when some one wants to add access new register address.
>>
>> Since S0ix counter access is only GCR access use case in PMC driver, I
>> thought  both solutions has some pros and cons.
>>> please do consider the below explaination for why we shoud not take this
>>> approch to fix WDT issue. Telemetry driver has no issues while loading since
>>> its not using any register in the GCR region.
>>> PMC on BXT/APL platforms has contiguous IPC and GCR regions. PMC_IPC driver
>>> maps the entire IPC and GCR region. It would be inefficient to map and unmap
>>> each time we want to use another register present in IPC or GCR spaces.
>> But I am wondering whether there will be any future GCR access
>> changes in PMC driver?
> You never know!
>
>> If its going to be just S0ix counter access then I don't think we
>> need to worry about performance here.
> Ditto, its not about performance.
>
>>> iTCO_WDT driver needs to check the BIT4 (NO_REBOOT) of PMC_CFG register
>>> (Offset: 0x1008) and this falls in GCR space. The iTCO_WDT driver fails to
>>> load because it can't request mem region for the resources already claimed
>>> by PMC_IPC driver. However, ioremap would still work here and WDT driver
>>> would load just fine.
>>>
>>> So, IMHO the problem lies elsewhere and we should find a way to handle this
>>> better in iTCO_WDT driver.
>>>
>>> The IPC and GCR resources belong to PMC and should be claimed by the PMC
>>> driver rightfully and should not be reclaimed by iTCO_WDT or any other
>>> driver.
>> iTCO_WDT , Telemtry and Punit are PMC dependent devices right ? And they
>> share the resources among them right ?
> Resources belong to PMC and hence it should own them, others share. Other
> drivers request and map those resources only when PMC driver does not
> already do so.
I see your point.
Correct way is to use MFD driver architecture for this device driver and 
use regmap for GCR registers.
We can get the regmap pointer in dependent device driver by getting the 
private data of parent device.
>
>>>
>>>> Currently this driver only need memory mapping for
>>>> s0ix counter registers. So this patch fixes this issue
>>>> by requesting memory mapping for only the s0ix counter mem
>>>> region.
>>> How about exposing a new API in PMC_IPC driver which can be used for reading
>>> the desired GCR register and it can be used by iTCO_WDT instead of
>>> requesting mem regions and remapping?
>> If you think in future if we might need access to more GCR space in
>> PMC driver, then we need to change this solution.
>>
>> Please let me know. If yes, I will add an API as you mentioned.
> Yes, I think so. Hope Andy and Darren are fine with that?
>
>

-- 
Sathyanarayanan Kuppuswamy
Android kernel developer