Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <1582694833-9407-3-git-send-email-mkshah@codeaurora.org>
References: <1582694833-9407-1-git-send-email-mkshah@codeaurora.org> <1582694833-9407-3-git-send-email-mkshah@codeaurora.org>
Subject: Re: [PATCH v7 2/3] soc: qcom: rpmh: Update dirty flag only when data changes
From:   Stephen Boyd <swboyd@chromium.org>
Cc:     linux-kernel@vger.kernel.org, linux-arm-msm@vger.kernel.org,
        agross@kernel.org, dianders@chromium.org, rnayak@codeaurora.org,
        ilina@codeaurora.org, lsrao@codeaurora.org,
        Maulik Shah <mkshah@codeaurora.org>
To:     Maulik Shah <mkshah@codeaurora.org>, bjorn.andersson@linaro.org,
        evgreen@chromium.org, mka@chromium.org
Date:   Wed, 26 Feb 2020 14:43:23 -0800
Message-ID: <158275700389.177367.5843608826404724304@swboyd.mtv.corp.google.com>
User-Agent: alot/0.9
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Quoting Maulik Shah (2020-02-25 21:27:12)
> Currently rpmh ctrlr dirty flag is set for all cases regardless
> of data is really changed or not. Add changes to update it when
> data is updated to newer values.
>=20
> Also move dirty flag updates to happen from within cache_lock.
>=20
> Signed-off-by: Maulik Shah <mkshah@codeaurora.org>
> Reviewed-by: Srinivas Rao L <lsrao@codeaurora.org>

Probably worth adding a Fixes tag here? Doesn't make sense to mark
something dirty when it isn't changed.

> ---
>  drivers/soc/qcom/rpmh.c | 21 ++++++++++++++++-----
>  1 file changed, 16 insertions(+), 5 deletions(-)
>=20
> diff --git a/drivers/soc/qcom/rpmh.c b/drivers/soc/qcom/rpmh.c
> index eb0ded0..83ba4e0 100644
> --- a/drivers/soc/qcom/rpmh.c
> +++ b/drivers/soc/qcom/rpmh.c
> @@ -139,20 +139,27 @@ static struct cache_req *cache_rpm_request(struct r=
pmh_ctrlr *ctrlr,
>  existing:
>         switch (state) {
>         case RPMH_ACTIVE_ONLY_STATE:
> -               if (req->sleep_val !=3D UINT_MAX)
> +               if (req->sleep_val !=3D UINT_MAX) {
>                         req->wake_val =3D cmd->data;
> +                       ctrlr->dirty =3D true;
> +               }
>                 break;
>         case RPMH_WAKE_ONLY_STATE:
> -               req->wake_val =3D cmd->data;
> +               if (req->wake_val !=3D cmd->data) {
> +                       req->wake_val =3D cmd->data;
> +                       ctrlr->dirty =3D true;
> +               }
>                 break;
>         case RPMH_SLEEP_STATE:
> -               req->sleep_val =3D cmd->data;
> +               if (req->sleep_val !=3D cmd->data) {
> +                       req->sleep_val =3D cmd->data;
> +                       ctrlr->dirty =3D true;
> +               }
>                 break;
>         default:
>                 break;

Please remove the default case. There are only three states in the enum. The
compiler will warn if a switch statement doesn't cover all cases and
we'll know to add something here if another enum value is added in the
future.

>         }
> =20
> -       ctrlr->dirty =3D true;
>  unlock:
>         spin_unlock_irqrestore(&ctrlr->cache_lock, flags);
> =20
> @@ -323,6 +331,7 @@ static void invalidate_batch(struct rpmh_ctrlr *ctrlr)
>         list_for_each_entry_safe(req, tmp, &ctrlr->batch_cache, list)
>                 kfree(req);
>         INIT_LIST_HEAD(&ctrlr->batch_cache);
> +       ctrlr->dirty =3D true;
>         spin_unlock_irqrestore(&ctrlr->cache_lock, flags);
>  }
> =20
> @@ -456,6 +465,7 @@ static int send_single(struct rpmh_ctrlr *ctrlr, enum=
 rpmh_state state,
>  int rpmh_flush(struct rpmh_ctrlr *ctrlr)
>  {
>         struct cache_req *p;
> +       unsigned long flags;
>         int ret;
> =20
>         if (!ctrlr->dirty) {
> @@ -488,7 +498,9 @@ int rpmh_flush(struct rpmh_ctrlr *ctrlr)
>                         return ret;
>         }
> =20
> +       spin_lock_irqsave(&ctrlr->cache_lock, flags);
>         ctrlr->dirty =3D false;
> +       spin_unlock_irqrestore(&ctrlr->cache_lock, flags);

So we take the spinlock to update it here. But we don't hold the
spinlock to test for !dirty up above. Seems like either rpmh_flush() can
only be called sequentially, or the lock added here needs to be held
during the whole flush. Which way is it?