Re: [PATCH 11/11] firmware: arm_scmi: Use asynchronous CLOCK_RATE_SET when possible
From: Stephen Boyd
Date: Mon Jul 22 2019 - 17:29:57 EST
Quoting Sudeep Holla (2019-07-08 08:47:30)
> CLOCK_PROTOCOL_ATTRIBUTES provides attributes to indicate the maximum
> number of pending asynchronous clock rate changes supported by the
> platform. If it's non-zero, then we should be able to use asynchronous
> clock rate set for any clocks until the maximum limit is reached.
>
> Keeping the current count of pending asynchronous clock set rate
> requests, we can decide if we can you asynchronous request for the
This last part of the sentence doesn't read properly. Please rewrite.
> incoming/new request.
>
> Signed-off-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> ---
> drivers/firmware/arm_scmi/clock.c | 21 ++++++++++++++++++---
> 1 file changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/firmware/arm_scmi/clock.c b/drivers/firmware/arm_scmi/clock.c
> index dd215bd11a58..70044b7c812e 100644
> --- a/drivers/firmware/arm_scmi/clock.c
> +++ b/drivers/firmware/arm_scmi/clock.c
> @@ -221,21 +222,35 @@ static int scmi_clock_rate_set(const struct scmi_handle *handle, u32 clk_id,
> u64 rate)
> {
> int ret;
> + u32 flags = 0;
> struct scmi_xfer *t;
> struct scmi_clock_set_rate *cfg;
> + struct clock_info *ci = handle->clk_priv;
>
> ret = scmi_xfer_get_init(handle, CLOCK_RATE_SET, SCMI_PROTOCOL_CLOCK,
> sizeof(*cfg), 0, &t);
> if (ret)
> return ret;
>
> + if (ci->max_async_req) {
> + if (atomic_inc_return(&ci->cur_async_req) < ci->max_async_req)
> + flags |= CLOCK_SET_ASYNC;
> + else
> + atomic_dec(&ci->cur_async_req);
Can this be combined with the atomic_dec() below and done after either
transfer?
> + }
> +
> cfg = t->tx.buf;
> - cfg->flags = cpu_to_le32(0);
> + cfg->flags = cpu_to_le32(flags);
> cfg->id = cpu_to_le32(clk_id);
> cfg->value_low = cpu_to_le32(rate & 0xffffffff);
> cfg->value_high = cpu_to_le32(rate >> 32);
>
> - ret = scmi_do_xfer(handle, t);
> + if (flags & CLOCK_SET_ASYNC) {
> + ret = scmi_do_xfer_with_response(handle, t);
> + atomic_dec(&ci->cur_async_req);
> + } else {
> + ret = scmi_do_xfer(handle, t);
> + }
I mean putting the atomic_dec() here.
>
> scmi_xfer_put(handle, t);
> return ret;