Date:   Wed, 22 Feb 2023 19:43:35 +0100
From:   Marek Vasut <marex@...x.de>
To:     "Russell King (Oracle)" <linux@...linux.org.uk>
Cc:     netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>,
        Andrew Lunn <andrew@...n.ch>,
        Arun Ramadoss <arun.ramadoss@...rochip.com>,
        Eric Dumazet <edumazet@...gle.com>,
        Florian Fainelli <f.fainelli@...il.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>, UNGLinuxDriver@...rochip.com,
        Vladimir Oltean <olteanv@...il.com>,
        Woojung Huh <woojung.huh@...rochip.com>, stable@...r.kernel.org
Subject: Re: [PATCH] net: dsa: microchip: Fix gigabit set and get function for
 KSZ87xx

On 2/22/23 17:39, Russell King (Oracle) wrote:
> On Wed, Feb 22, 2023 at 05:30:43PM +0100, Marek Vasut wrote:
>> On 2/22/23 16:56, Russell King (Oracle) wrote:
>>> On Wed, Feb 22, 2023 at 04:10:33PM +0100, Marek Vasut wrote:
>>>> On 2/22/23 14:03, Russell King (Oracle) wrote:
>>>>> On Wed, Feb 22, 2023 at 12:50:07PM +0000, Russell King (Oracle) wrote:
>>>>>> On Wed, Feb 22, 2023 at 04:17:38AM +0100, Marek Vasut wrote:
>>>>>>> diff --git a/drivers/net/dsa/microchip/ksz_common.c b/drivers/net/dsa/microchip/ksz_common.c
>>>>>>> index 729b36eeb2c46..7fc2155d93d6e 100644
>>>>>>> --- a/drivers/net/dsa/microchip/ksz_common.c
>>>>>>> +++ b/drivers/net/dsa/microchip/ksz_common.c
>>>>>>> @@ -319,7 +319,7 @@ static const u16 ksz8795_regs[] = {
>>>>>>>     	[S_BROADCAST_CTRL]		= 0x06,
>>>>>>>     	[S_MULTICAST_CTRL]		= 0x04,
>>>>>>>     	[P_XMII_CTRL_0]			= 0x06,
>>>>>>> -	[P_XMII_CTRL_1]			= 0x56,
>>>>>>> +	[P_XMII_CTRL_1]			= 0x06,
>>>>>>
>>>>>> Looking at this driver, I have to say that it looks utterly vile
>>>>>> from the point of view of being sure that it is correct, and I
>>>>>> think this patch illustrates why.
>>>>>>
>>>>>> You mention you're using a KSZ8794. This uses the ksz8795_regs
>>>>>> array, and ksz8_dev_ops. You claim this is about the P_GMII_1GBIT_M
>>>>>> bit, which is bit 6.
>>>>>>
>>>>>> This bit is accessed only by ksz_get_gbit() and ksz_set_gbit().
>>>>>>
>>>>>> Firstly, ksz_set_gbit() is only called from ksz_port_set_xmii_speed(),
>>>>>> which is only called from ksz9477_phylink_mac_link_up(). This is only
>>>>>> referenced by ksz9477_dev_ops and lan937x_dev_ops, but not ksz8_dev_ops.
>>>>>> Therefore, ksz_set_gbit() is not called for KSZ8794.
>>>>>>
>>>>>> ksz_get_gbit() is only referenced by ksz9477.c in
>>>>>> ksz9477_get_interface(), called only by ksz9477_config_cpu_port().
>>>>>> This is only referenced by ksz9477_dev_ops, but not ksz8_dev_ops.
>>>>>>
>>>>>> Therefore, my conclusion is that neither of the ksz_*_gbit()
>>>>>> functions are called on KSZ8794, and thus your change has no effect
>>>>>> on the driver's use of P_GMII_1GBIT_M - I think if you put some
>>>>>> debugging printk()s into both ksz_*_gbit() functions, it'll prove
>>>>>> that.
>>>>>>
>>>>>> There's other places that P_XMII_CTRL_1 is accessed - ksz_set_xmii()
>>>>>> and ksz_get_xmii(). These look at the P_MII_SEL_M, P_RGMII_ID_IG_ENABLE
>>>>>> and P_RGMII_ID_EG_ENABLE bits - bits 0, 1, 3 and 4.
>>>>>>
>>>>>> ksz_get_xmii() is only called by ksz9477_get_interface(), which we've
>>>>>> already looked at above as not being called.
>>>>>>
>>>>>> ksz_set_xmii() is only called by ksz_phylink_mac_config(), which is
>>>>>> always called irrespective of the KSZ chip.
>>>>>>
>>>>>> Now, let's look at functions that access P_XMII_CTRL_0. These are
>>>>>> ksz_set_100_10mbit() and ksz_duplex_flowctrl(). The former
>>>>>> accesses bit P_MII_100MBIT_M, which is bit 4. The latter looks at
>>>>>> bits 6, bit 5, and possibly bit 3 depending on the masks being used.
>>>>>> KSZ8795 uses ksz8795_masks, which omits bit 3, so bits 5 and 6.
>>>>>> Note... bit 6 is also P_GMII_1GBIT_M. So if ksz_duplex_flowctrl()
>>>>>> is ever called for the KSZ8795, then we have a situation where
>>>>>> the P_GMII_1GBIT_M will be manipulated.
>>>>>>
>>>>>> ksz_set_100_10mbit() is only called from ksz_port_set_xmii_speed(),
>>>>>> which we've established won't be called.
>>>>>>
>>>>>> ksz_duplex_flowctrl() is only called from ksz9477_phylink_mac_link_up()
>>>>>> which we've also established won't be called.
>>>>>>
>>>>>> So, as far as I can see, P_XMII_CTRL_0 won't be accessed on this
>>>>>> device.
>>>>>>
>>>>>> Now, what about other KSZ devices - I've analysed this for the KSZ8795,
>>>>>> but what about any of the others which use this register table? It
>>>>>> looks to me like those that use ksz8795_regs[] all use ksz8_dev_ops
>>>>>> and the same masks and bitvals, so they should be the same.
>>>>>>
>>>>>> That is a hell of a lot of work to prove that setting both
>>>>>> P_XMII_CTRL_0 and P_XMII_CTRL_1 to point at the same register is
>>>>>> in fact safe. Given the number of registers, the masks, and bitval
>>>>>> arrays, doing this to prove every combination and then analysing
>>>>>> the code is utterly impractical - and thus why I label this driver
>>>>>> as "vile". Is there really no better option to these register
>>>>>> arrays, bitval arrays and mask arrays - something that makes it
>>>>>> easier to review and prove correctness?
>>>>>>
>>>>>> I'm not going to give a reviewed-by for this, because... I could
>>>>>> have made a mistake in the above analysis given the vile nature
>>>>>> of this driver.
>>>>>
>>>>> However, I should add that - as a result of neither ksz_*_gbit()
>>>>> functions being used, I consider at least the subject line to be
>>>>> rather misleading! While it may be something that you spotted,
>>>>> I suspect the other bits that are actually written are more the
>>>>> issue you're fixing.
>>>>
>>>> Thank you for the lengthy review, I agree the driver and the register offset
>>>> calculation are hideous.
>>>>
>>>> However, I did spent quite a bit of time on it already and checked both
>>>> P_XMII_CTRL_0 and P_XMII_CTRL_1 mappings with printks and by dumping the
>>>> register values via regmap debugfs interface.
>>>>
>>>> Also note that KSZ8794 and KSZ8795 seem to be the same chip die, just
>>>> different package (the former has fewer ports) and different chip ID.
>>>
>>> It's not clear what you think of my review and whether you are going to
>>> take any action at all... So, let me try again...
>>>
>>> The fundamental question that my review raises was whether this gigabit
>>> bit is actually used, and your response remains silent on that point.
>>>
>>> As the gigabit bit is not actually used given the code structure, it
>>> is irrelevant for this commit, despite Is_Gbit being the thing that
>>> lead to the patch.
>>>
>>> Therefore, I believe that the patch description needs to be updated
>>> to state what the effective fix for this change is (which is to fix
>>> ksz_set_xmii()) rather than making it sound like it's fixing a wrong
>>> access for Is_Gbit.
>>>
>>> The reason I think this is important is that if we need to look back
>>> at the history, current description leads one to think that this
>>> change is about fixing the Is_Gbit bit - but that isn't used as the
>>> code stands. The effective change that this patch makes is to the
>>> only access the driver makes to this register in ksz_set_xmii(),
>>> and I think that needs to be explained as the primary reason for
>>> this patch. Fixing Is_Gbit seems to be merely incidental.
>>
>> On the hardware I use here, the P_XMII_CTRL_1 register ends up being
>> populated with all bits set, 0xff. Without this change, the driver writes to
>> non-existent register when it attempts to access P_XMII_CTRL_1 .
> 
> Why is this so difficult?
> 
> I'm *not* disagreeing with the patch. I'm disagreeing with your
> commit description.

Why not comment on the commit message part which you disagree with, and 
suggest an improved wording you do agree with ? Then I can send a V2 
with that part changed, and be done with it.

Powered by blists - more mailing lists