Message-ID: <c47b05ee-5e0a-08ee-318a-52e370619d2c@amd.com>
Date:   Mon, 23 Aug 2021 12:58:18 -0500
From:   Terry Bowman <Terry.Bowman@....com>
To:     Guenter Roeck <linux@...ck-us.net>, linux-watchdog@...r.kernel.org
Cc:     wim@...ux-watchdog.org, linux-kernel@...r.kernel.org,
        rrichter@....com, thomas.lendacky@....com
Subject: Re: [PATCH] Watchdog: sp5100_tco: Replace watchdog cd6h/cd7h port I/O
 accesses with MMIO accesses



On 8/18/21 3:34 PM, Guenter Roeck wrote:
> On 8/16/21 2:29 PM, Terry Bowman wrote:
>>
>>
>> On 8/13/21 5:37 PM, Guenter Roeck wrote:
>>> On 8/13/21 2:32 PM, Terry Bowman wrote:
>>>> Use MMIO instead of port I/O during SMBus controller address discovery.
>>>> Also, update how EFCH capability is determined by replacing a family check
>>>> with a PCI revision ID check.
>>>>
>>>> cd6h/cd7h port I/O can be disabled on recent AMD hardware. Read accesses to
>>>> disabled cd6h/cd7h port I/O will return F's and written data is dropped.
>>>> The recommended workaround to handle disabled cd6h/cd7h port I/O is
>>>> replacing port I/O with MMIO accesses. The MMIO access method has been
>>>> available since at least SMBus controllers using PCI revision 0x59.
>>>>
>>>> The sp5100_tco driver uses a CPU family match of 17h to determine
>>>> EFCH_PM_DECODEEN_WDT_TMREN register support. Using a family check requires
>>>> driver updates for each new AMD CPU family following 17h. This patch
>>>> replaces the family check with a check for SMBus PCI revision ID 0x59 and
>>>> later. Note: Family 17h processors use SMBus PCI revision ID 0x59. The
>>>> intent is to use the PCI revision ID check to support future AMD processors
>>>> while minimizing required driver changes. The caveat with this change is
>>>> the sp5100_tco driver must be updated if a new AMD processor family changes
>>>> the EFCH design or the SMBus PCI ID value doesn't follow this pattern.
>>>>
>>>> Tested with forced WDT reset using `cat >> /dev/watchdog`.
>>>>
>>>
>>> I am sorry, I don't understand why the new code can not use devm functions,
>>> why the new data structure is necessary in the first place, and why it is
>>> not possible to improve alignment with the existing code. This will require
>>> a substantial amount of time to review to ensure that the changes are not
>>> excessive (at first glance it for sure looks like that to me).
>>>
>>> Guenter
>>>
>>
>> Hi Guenter,
>>
>> I can change the patch to use devm functions as you mentioned. My
>> understanding is the patch's reservation and mapping related functions
>> are the focus. I originally chose not to use devm functions because the
>> patch's MMIO reserved and mapped resources are not held for the driver
>> lifetime as is the case for most device managed resources. The
>> sp5100_tco driver must only hold these MMIO resources briefly because
>> other drivers use the same EFCH MMIO registers. An example of another
>> driver using the same registers is the piix4_smbus driver (drivers/i2c
>> /busses/i2c-piix4.c). This patch can be changed to use the devm
>> functions but the driver may not benefit from the device management.
>>
>> The 'struct efch_cfg' addition is needed for MMIO reads/writes as well
>> as during cleanup when leaving sp5100_region_setup(). This structure was
>> chosen to contain the data instead of passing multiple parameters to
>> each EFCH function called.
>>
>> Do you have any recommendations for how to best improve the alignment?
>>
> 
> Overall it seems to me that it might make more sense to implement this
> as new driver instead of messing with the existing driver. Have you
> thought about that ?
> 
> Guenter

A new driver was initially considered for this patch. I decided to update
the sp5100_tco driver instead because the changes are limited to
initialization and post-initialization code is not modified. Also, updating
the existing sp5100_tco driver allows for code reuse.

This patch can be split into 2 or more patches to simplify review and
testing. Would this help?

Regards,
Terry

Powered by blists - more mailing lists