Re: [PATCH v2 0/4] i2c-hid: Save power by reducing i2c xfers with block reads

[Hans de Goede]

Hi,

On 9/22/20 11:19 AM, Jiri Kosina wrote:
> On Wed, 16 Sep 2020, Sultan Alsawaf wrote:
> 
>> From: Sultan Alsawaf <sultan@xxxxxxxxxxxxxxx>
>>
>> This is a fixed resubmission of "[PATCH 0/2] i2c-hid: Save power by reducing i2c
>> xfers with block reads". That original patchset did not have enough fixes for
>> the designware i2c adapter's I2C_M_RECV_LEN feature, which is documented
>> extensively in the original email thread.
>>
>> Here is the original cover letter, which still applies:
>> "I noticed on my Dell Precision 15 5540 with an i9-9880H that simply putting my
>> finger on the touchpad would increase my system's power consumption by 4W, which
>> is quite considerable. Resting my finger on the touchpad would generate roughly
>> 4000 i2c irqs per second, or roughly 20 i2c irqs per touchpad irq.
>>
>> Upon closer inspection, I noticed that the i2c-hid driver would always transfer
>> the maximum report size over i2c (which is 60 bytes for my touchpad), but all of
>> my touchpad's normal touch events are only 32 bytes long according to the length
>> byte contained in the buffer sequence.
>>
>> Therefore, I was able to save about 2W of power by passing the I2C_M_RECV_LEN
>> flag in i2c-hid, which says to look for the payload length in the first byte of
>> the transfer buffer and adjust the i2c transaction accordingly. The only problem
>> though is that my i2c controller's driver allows bytes other than the first one
>> to be used to retrieve the payload length, which is incorrect according to the
>> SMBus spec, and would break my i2c-hid change since not *all* of the reports
>> from my touchpad are conforming SMBus block reads.
>>
>> This patchset fixes the I2C_M_RECV_LEN behavior in the designware i2c driver and
>> modifies i2c-hid to use I2C_M_RECV_LEN to save quite a bit of power. Even if the
>> peripheral controlled by i2c-hid doesn't support block reads, the i2c controller
>> drivers should cope with this and proceed with the i2c transfer using the
>> original requested length."
>>
>> Sultan
>>
>> Sultan Alsawaf (4):
>>   i2c: designware: Fix transfer failures for invalid SMBus block reads
>>   i2c: designware: Ensure tx_buf_len is nonzero for SMBus block reads
>>   i2c: designware: Allow SMBus block reads up to 255 bytes in length
>>   HID: i2c-hid: Use block reads when possible to save power
>>
>>  drivers/hid/i2c-hid/i2c-hid-core.c         |  5 ++++-
>>  drivers/i2c/busses/i2c-designware-master.c | 15 +++++++++------
>>  2 files changed, 13 insertions(+), 7 deletions(-)
> 
> Hans, Benjamin, could you please give this patchset some smoke-testing? It 
> looks good to me, but I'd like it to get some testing from your testing 
> machinery before merging.

Sorry for being slow to respond to this. I have not gotten around to testing
this, but I saw another email that this breaks things on at least AMD
platforms, so I guess that this is on hold for now ?

Regards,

Hans