linux-kernel - Re: [PATCH RESEND 1/2] i2c: tegra: Add GPCDMA support

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: <cd0374f1-2c05-7e61-7187-cfc9c42edf63@gmail.com>
Date:   Tue, 23 Aug 2022 00:08:32 +0300
From:   Dmitry Osipenko <digetx@...il.com>
To:     Akhil R <akhilrajeev@...dia.com>,
        Dmitry Osipenko <dmitry.osipenko@...labora.com>,
        "robh+dt@...nel.org" <robh+dt@...nel.org>,
        "devicetree@...r.kernel.org" <devicetree@...r.kernel.org>,
        "christian.koenig@....com" <christian.koenig@....com>,
        Jonathan Hunter <jonathanh@...dia.com>,
        Laxman Dewangan <ldewangan@...dia.com>,
        "linux-i2c@...r.kernel.org" <linux-i2c@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-tegra@...r.kernel.org" <linux-tegra@...r.kernel.org>,
        "sumit.semwal@...aro.org" <sumit.semwal@...aro.org>,
        "thierry.reding@...il.com" <thierry.reding@...il.com>,
        "wsa@...nel.org" <wsa@...nel.org>
Subject: Re: [PATCH RESEND 1/2] i2c: tegra: Add GPCDMA support

22.08.2022 23:33, Dmitry Osipenko пишет:
> 22.08.2022 13:29, Akhil R пишет:
>>> On 8/22/22 09:56, Akhil R wrote:
>>>>> 19.08.2022 18:15, Dmitry Osipenko пишет:
>>>>>> 19.08.2022 15:23, Akhil R пишет:
>>>>>>>      if (of_device_is_compatible(np, "nvidia,tegra210-i2c-vi"))
>>>>>>>              i2c_dev->is_vi = true;
>>>>>>> +    else
>>>>>>> +            i2c_dev->dma_support = !!(of_find_property(np, "dmas",
>>>>>>> + NULL));
>>>>>>
>>>>>> 1. You leak the np returned by of_find_property().
>>>>>>
>>>>>> 2. There is device_property_read_bool() for this kind of
>>>>>> property-exists checks.
>>>> Okay. I went by the implementation in of_dma_request_slave_channel() to
>>>> check 'dmas'.
>>>>
>>>>>>
>>>>>> 3. If "dmas" is missing in DT, then dma_request_chan() should return
>>>>>> NULL and everything will work fine. I suppose you haven't tried to
>>>>>> test this code.
>>>>>
>>>>> Although, no. It should return ERR_PTR(-ENODEV) and then you should check
>>>>> the return code.
>>>> Yes. Agree that it is more agnostic to check for ERR_PTR(-ENODEV). But since I
>>>> call tegra_init_dma() for every large transfer until DMA is initialized, wouldn't
>>>> it be better to have a flag inside the driver so that we do not have to go
>>> through
>>>> so many functions for every attempted DMA transaction to find out that the
>>> DT
>>>> properties don't exist?
>>>>
>>>> Shall I just put i2c_dev->dma_support = true here since DMA is supported by
>>>> hardware? It would turn false if dma_request_chan() returns something other
>>>> than -EPROBE_DEFER.
>>>>
>>>>       if (of_device_is_compatible(np, "nvidia,tegra210-i2c-vi"))
>>>>               i2c_dev->is_vi = true;
>>>>  +    else
>>>>  +            i2c_dev->dma_support = true;
>>>
>>> The code already has dma_mode for that. I don't see why another variable
>>> is needed.
>>>
>>> Either add new generic dma_request_chan_optional() that will return NULL
>>> if channel is not available and make Tegra I2C driver to use it, or
>>> handle the error code returned by dma_request_chan().
>>
>> Let me elaborate my thoughts. 
>>
>> The function tegra_i2c_init_dma() is also called inside tegra_i2c_xfer_msg() if
>> DMA is not initialized before, i.e. if (!i2c_dev->dma_buf).
> 
> This is not true
> 
> i2c_dev->dma_mode=false if !i2c_dev->dma_buf and that's it
> 
> https://elixir.bootlin.com/linux/v6.0-rc2/source/drivers/i2c/busses/i2c-tegra.c#L1253
> 
> tegra_i2c_init_dma() is invoked only during probe
> 
>> So, if suppose there is no DT entry for dmas, the driver would have to go take the
>> path tegra_i2c_init_dma() -> dma_request_chan() -> of_*() apis -> ... and then figure
>> out that DMA is not supported. This would happen for each transfer of size larger than
>> I2C_PIO_MODE_PREFERRED_LEN.
>>
>> To avoid this, I am looking for a variable/flag which can indicate if the driver should attempt
>> to configure DMA or not. I didn't quite get the idea if dma_mode can be extended to support
>> this, because it is updated based on xfer_size on each transfer. My idea of i2c_dev->dma_support
>> is that it will be constant after the probe().

I see now that it's you added tegra_i2c_init_dma() to
tegra_i2c_xfer_msg(). And tegra_i2c_init_dma() already falls back to PIO
if DMA is unavailable.

I don't remember why !IS_ENABLED(CONFIG_TEGRA20_APB_DMA) was added to
tegra_i2c_init_dma(), but if dma_request_chan() returns -EPROBE_DEFER
when there is no DMA channel available at all, then you should fix it.

Trying to initialize DMA during transfer if it failed to initialize
during probe is a wrong approach. DMA must be initialized only once
during probe. Please make the probe to work properly.