Message-ID: <4c40fbe4-3492-ec76-e452-3a3ecb0f2433@ti.com>
Date:   Fri, 11 Aug 2023 10:24:18 -0500
From:   Andrew Davis <afd@...com>
To:     Md Danish Anwar <a0501179@...com>,
        MD Danish Anwar <danishanwar@...com>,
        Randy Dunlap <rdunlap@...radead.org>,
        Roger Quadros <rogerq@...nel.org>,
        Simon Horman <simon.horman@...igine.com>,
        Vignesh Raghavendra <vigneshr@...com>,
        Andrew Lunn <andrew@...n.ch>,
        Richard Cochran <richardcochran@...il.com>,
        Conor Dooley <conor+dt@...nel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>,
        Rob Herring <robh+dt@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        "David S. Miller" <davem@...emloft.net>
CC:     <nm@...com>, <srk@...com>, <linux-kernel@...r.kernel.org>,
        <devicetree@...r.kernel.org>, <netdev@...r.kernel.org>,
        <linux-omap@...r.kernel.org>,
        <linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH v3 3/5] net: ti: icss-iep: Add IEP driver

On 8/10/23 6:50 AM, Md Danish Anwar wrote:
> Hi Andrew,
> 
> On 09/08/23 8:30 pm, Andrew Davis wrote:
>> On 8/9/23 6:49 AM, MD Danish Anwar wrote:
>>> From: Roger Quadros <rogerq@...com>
>>>
>>> Add a driver for Industrial Ethernet Peripheral (IEP) block of PRUSS to
>>> support timestamping of ethernet packets and thus support PTP and PPS
>>> for PRU ethernet ports.
>>>
>>> Signed-off-by: Roger Quadros <rogerq@...com>
>>> Signed-off-by: Lokesh Vutla <lokeshvutla@...com>
>>> Signed-off-by: Murali Karicheri <m-karicheri2@...com>
>>> Signed-off-by: Vignesh Raghavendra <vigneshr@...com>
>>> Signed-off-by: MD Danish Anwar <danishanwar@...com>
>>> ---
>>>    drivers/net/ethernet/ti/Kconfig          |  12 +
>>>    drivers/net/ethernet/ti/Makefile         |   1 +
>>>    drivers/net/ethernet/ti/icssg/icss_iep.c | 935 +++++++++++++++++++++++
>>>    drivers/net/ethernet/ti/icssg/icss_iep.h |  38 +
>>>    4 files changed, 986 insertions(+)
>>>    create mode 100644 drivers/net/ethernet/ti/icssg/icss_iep.c
>>>    create mode 100644 drivers/net/ethernet/ti/icssg/icss_iep.h
>>>
>>> diff --git a/drivers/net/ethernet/ti/Kconfig b/drivers/net/ethernet/ti/Kconfig
>>> index 63e510b6860f..88b5b1b47779 100644
>>> --- a/drivers/net/ethernet/ti/Kconfig
>>> +++ b/drivers/net/ethernet/ti/Kconfig
>>> @@ -186,6 +186,7 @@ config CPMAC
>>>    config TI_ICSSG_PRUETH
>>>        tristate "TI Gigabit PRU Ethernet driver"
>>>        select PHYLIB
>>> +    select TI_ICSS_IEP
>>
>> Why not save selecting this until you add its use in the ICSSG_PRUETH driver in
>> the next patch.
>>
> 
> The next patch is only adding changes to icssg-prueth .c /.h files. This patch
> is adding changes to Kconfig and the Makefile. To keep it that way selecting
> this is added in this patch. No worries, I will move this to next patch.
> 
>> [...]
>>
>>> +
>>> +static u32 icss_iep_readl(struct icss_iep *iep, int reg)
>>> +{
>>> +    return readl(iep->base + iep->plat_data->reg_offs[reg]);
>>> +}
>>
>> Do these one line functions really add anything? Actually why
>> not use the regmap you have here.
> 
> These one line functions are not really adding anything but they are acting as
> a wrapper around readl /writel and providing some sort of encapsulation as
> directly calling readl will result in a little complicated code.
> 
> /* WIth One line function */
> ts_lo = icss_iep_readl(iep, ICSS_IEP_COUNT_REG0);
> 
> /* Without one line function */
> ts_lo = readl(iep->base, iep->plat_data->reg_offs[ICSS_IEP_COUNT_REG0]);
> 
> Previously regmap was used in this driver. But in older commit [1] in
> 5.10-ti-linux-kernel (Before I picked the driver for upstream) it got changed
> to readl / writel stating that regmap_read / write is too slow. IEP is time
> sensitive and needs faster read and write, probably because of this they
> changed it.
> 

Sounds like you only need direct register access for time sensitive
gettime/settime functions, if that is the only place writel()/readl() is
needed just drop the helper and use directly in that one spot.

>>
>> [...]
>>
>>> +static void icss_iep_enable(struct icss_iep *iep)
>>> +{
>>> +    regmap_update_bits(iep->map, ICSS_IEP_GLOBAL_CFG_REG,
>>> +               IEP_GLOBAL_CFG_CNT_ENABLE,
>>> +               IEP_GLOBAL_CFG_CNT_ENABLE);
>>
>> Have you looked into regmap_fields?
>>
> 
> No I hadn't. But now I looked into regmap_fields, seems to be another way to
> update the bits, instead of passing mask and value, regmap_filed_read / write
> only takes the value. But for that we will need to create a regmap field. If
> you want me to switch to regmap_fields instead of regmap_update_bits I can make
> the changes. But I am fine with regmap_update_bits().
> 

I'm suggesting regmap fields as I have used it several times and it resulted
in greatly improved readability. Yes you will need a regmap field table, but
that is the best part, it lets you put all your bit definitions in one spot
that can match 1:1 with the datasheet, much easier to check for correctness
than if the bit usages are all spread out in the driver.

I won't insist on you converting this driver to use it today, but I do recommend
you at least give it a shot for your own learning.

Andrew

>> [...]
>>
>>> +
>>> +    if (!!(iep->latch_enable & BIT(index)) == !!on)
>>> +        goto exit;
>>> +
>>
>> There has to be a better way to write this logic..
>>
>> [...]
>>
>>> +
>>> +static const struct of_device_id icss_iep_of_match[];
>>> +
>>
>> Why the forward declaration?
> 
> I will remove this, I don't see any reason for this.
> 
>>
>>> +static int icss_iep_probe(struct platform_device *pdev)
>>> +{
>>> +    struct device *dev = &pdev->dev;
>>> +    struct icss_iep *iep;
>>> +    struct clk *iep_clk;
>>> +
>>> +    iep = devm_kzalloc(dev, sizeof(*iep), GFP_KERNEL);
>>> +    if (!iep)
>>> +        return -ENOMEM;
>>> +
>>> +    iep->dev = dev;
>>> +    iep->base = devm_platform_ioremap_resource(pdev, 0);
>>> +    if (IS_ERR(iep->base))
>>> +        return -ENODEV;
>>> +
>>> +    iep_clk = devm_clk_get(dev, NULL);
>>> +    if (IS_ERR(iep_clk))
>>> +        return PTR_ERR(iep_clk);
>>> +
>>> +    iep->refclk_freq = clk_get_rate(iep_clk);
>>> +
>>> +    iep->def_inc = NSEC_PER_SEC / iep->refclk_freq;    /* ns per clock tick */
>>> +    if (iep->def_inc > IEP_MAX_DEF_INC) {
>>> +        dev_err(dev, "Failed to set def_inc %d.  IEP_clock is too slow to be
>>> supported\n",
>>> +            iep->def_inc);
>>> +        return -EINVAL;
>>> +    }
>>> +
>>> +    iep->plat_data = of_device_get_match_data(dev);
>>
>> Directly using of_*() functions is often wrong, try just device_get_match_data().
>>
> 
> Sure. I will change to device_get_match_data().
> 
>> [...]
>>
>>> +static struct platform_driver icss_iep_driver = {
>>> +    .driver = {
>>> +        .name = "icss-iep",
>>> +        .of_match_table = of_match_ptr(icss_iep_of_match),
>>
>> This driver cannot work without OF, using of_match_ptr() is not needed.
>>
> 
> Sure, I will drop of_match_ptr().
> 
>> Andrew
> 
> 
> For reading and updating registers, we can have
> 	1. icss_iep_readl / writel and regmap_update_bits() OR
> 	2. regmap_read / write and regmap_update_bits() OR
> 	3. icss_iep_readl / writel and regmap_fields OR
> 	4. regmap_read / write and regmap_fields
> 	
> 
> Currently we are using 1. Please let me know if you are fine with this and I
> can continue using 1. If not, please let me know your recommendation out of this 4.
> 
> [1]
> https://git.ti.com/cgit/ti-linux-kernel/ti-linux-kernel/commit/?h=linux-5.10.y&id=f4f45bf71cad5be232536d63a0557d13a7eed162
> 

Powered by blists - more mailing lists