Hi Stephen,
On 2021-02-24 12:36, Stephen Boyd wrote:
> Quoting [email protected] (2021-02-18 06:15:17)
>> Hi Stephen,
>>
>> On 2021-01-13 12:24, Stephen Boyd wrote:
>> > Quoting Roja Rani Yarubandi (2021-01-08 07:05:45)
>> >> diff --git a/drivers/i2c/busses/i2c-qcom-geni.c
>> >> b/drivers/i2c/busses/i2c-qcom-geni.c
>> >> index 214b4c913a13..c3f584795911 100644
>> >> --- a/drivers/i2c/busses/i2c-qcom-geni.c
>> >> +++ b/drivers/i2c/busses/i2c-qcom-geni.c
>> >> @@ -375,6 +375,32 @@ static void geni_i2c_tx_msg_cleanup(struct
>> >> geni_i2c_dev *gi2c,
>> >> }
>> >> }
>> >>
>> >> +static void geni_i2c_stop_xfer(struct geni_i2c_dev *gi2c)
>> >> +{
>> >> + int ret;
>> >> + u32 geni_status;
>> >> + struct i2c_msg *cur;
>> >> +
>> >> + /* Resume device, as runtime suspend can happen anytime during
>> >> transfer */
>> >> + ret = pm_runtime_get_sync(gi2c->se.dev);
>> >> + if (ret < 0) {
>> >> + dev_err(gi2c->se.dev, "Failed to resume device: %d\n",
>> >> ret);
>> >> + return;
>> >> + }
>> >> +
>> >> + geni_status = readl_relaxed(gi2c->se.base + SE_GENI_STATUS);
>> >> + if (geni_status & M_GENI_CMD_ACTIVE) {
>> >> + cur = gi2c->cur;
>> >
>> > Why don't we need to hold the spinlock gi2c::lock here?
>> >
>>
>> I am not seeing any race here. May I know which race are you
>> suspecting
>> here?
>
> Sorry there are long delays between posting and replies to my review
> comments. It takes me some time to remember what we're talking about
> because this patch has dragged on for many months.
>
Sorry for the delayed responses.
> So my understanding is that gi2c::lock protects the 'cur' pointer. I
> imagine this scenario might go bad
>
> CPU0 CPU1
> ---- ----
> geni_i2c_stop_xfer()
> ... geni_i2c_rx_one_msg()
> gi2c->cur = cur1;
> cur = gi2c->cur;
> ... geni_i2c_tx_one_msg()
> gi2c->cur = cur2;
> geni_i2c_abort_xfer()
> <uses cur2>
> if (cur->flags & I2C_M_RD)
> <uses cur1 for the condition and call; oops that's bad>
>
> It's almost like we should combine the geni_i2c_abort_xfer() logic with
> the rx/tx message cleanup functions so that it's all done under one
> lock. Unfortunately it's complicated by the fact that there are various
> completion waiting timeouts involved. Fun!
>
Thanks for the explanation. Fixed this possible race by protecting
gi2c->cur
and calling geni_i2c_abort_xfer() with adding another parameter to
differentiate
from which sequence is the geni_i2c_abort_xfer() called and handle the
spin_lock/spin_unlock accordingly inside geni_i2c_abort_xfer()
> But even after all that, I don't see how the geni_i2c_stop_xfer() puts
> a
> stop to future calls to geni_i2c_rx_one_msg() or geni_i2c_tx_one_msg().
Now handled this by adding a bool variable "stop_xfer" in geni_i2c_dev
struct,
used to put stop to upcoming geni_i2c_rx_one_msg() and
geni_i2c_tx_one_msg() calls
once we receive the shutdown call.
> The hardware isn't disabled from what I can tell. The irq isn't
> disabled, the clks aren't turned off, etc. What is to stop an i2c
> device
> from trying to use the bus after this shutdown function is called? If
> anything, this function looks like a "flush", where we flush out any
> pending transfer. Where's the "plug" operation that prevents any future
> operations from following this call?
>
We are turning off clocks and disabling irq in
geni_i2c_runtime_suspend().
IIUC about shutdown sequence, during "remove" we will unplug the device
with opposite
calls to "probe's" plug operations example i2c_del_adapter(). For
"shutdown", as system
is going to shutdown, there is no need of unplug operations to be done.
> BTW, I see this is merged upstream. That's great, but it seems broken.
> Please fix it or revert it out.
>
>>
>> >> + geni_i2c_abort_xfer(gi2c);
>> >> + if (cur->flags & I2C_M_RD)
>> >> + geni_i2c_rx_msg_cleanup(gi2c, cur);
>> >> + else
>> >> + geni_i2c_tx_msg_cleanup(gi2c, cur);
>> >> + }
>> >> +
>> >> + pm_runtime_put_sync_suspend(gi2c->se.dev);
>> >> +}
>> >> +
>> >> static int geni_i2c_rx_one_msg(struct geni_i2c_dev *gi2c, struct
>> >> i2c_msg *msg,
>> >> u32 m_param)
>> >> {
Thanks,
Roja