[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <x4y44ajcdi2y2dieaa6oohrptpzyiono3fruvwcdelmtzsh4ne@cgqxsz45ohcy>
Date: Tue, 30 Jul 2024 15:01:13 +0200
From: Markus Schneider-Pargmann <msp@...libre.com>
To: Nishanth Menon <nm@...com>
Cc: Tero Kristo <kristo@...nel.org>,
Santosh Shilimkar <ssantosh@...nel.org>, Rob Herring <robh@...nel.org>,
Krzysztof Kozlowski <krzk+dt@...nel.org>, Conor Dooley <conor+dt@...nel.org>,
Vignesh Raghavendra <vigneshr@...com>, Vibhore Vardhan <vibhore@...com>,
Kevin Hilman <khilman@...libre.com>, Dhruva Gole <d-gole@...com>,
linux-arm-kernel@...ts.infradead.org, devicetree@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 2/6] firmware: ti_sci: Partial-IO support
On Tue, Jul 30, 2024 at 07:28:01AM GMT, Nishanth Menon wrote:
> On 10:00-20240729, Markus Schneider-Pargmann wrote:
> > Add support for Partial-IO poweroff. In Partial-IO pins of a few modules
> > can generate system wakeups while DDR memory is not powered resulting in
> > a fresh boot of the system. The modules that can be wakeup sources are
> > defined by the devicetree.
> >
> > Only wakeup sources that are actually enabled by the user will be
> > considered as a an active wakeup source. If none of the wakeup sources
> > are enabled the system will do a normal poweroff. If at least one wakeup
> > source is enabled it will instead send a TI_SCI_MSG_PREPARE_SLEEP
> > message from the sys_off handler. Sending this message will result in an
> > immediate shutdown of the system. No execution is expected after this
> > point. The code will enter an infinite loop.
> >
> > The wakeup source device nodes are gathered during probe. But they are
> > only resolved to the actual devices in the sys_off handler, if they
> > exist. If they do not exist, they are ignored.
> >
> > A short documentation about Partial-IO can be found in section 6.2.4.5
> > of the TRM at
> > https://www.ti.com/lit/pdf/spruiv7
> >
> > Signed-off-by: Markus Schneider-Pargmann <msp@...libre.com>
> > ---
> > drivers/firmware/ti_sci.c | 160 +++++++++++++++++++++++++++++++++-----
> > drivers/firmware/ti_sci.h | 34 ++++++++
> > 2 files changed, 175 insertions(+), 19 deletions(-)
> >
> > diff --git a/drivers/firmware/ti_sci.c b/drivers/firmware/ti_sci.c
> > index 160968301b1f..ba2e56da0215 100644
> > --- a/drivers/firmware/ti_sci.c
> > +++ b/drivers/firmware/ti_sci.c
> > @@ -99,6 +99,9 @@ struct ti_sci_desc {
> > * @node: list head
> > * @host_id: Host ID
> > * @users: Number of users of this instance
> > + * @nr_wakeup_sources: Number of device nodes in wakeup_source_nodes
> > + * @wakeup_source_nodes: Array of all device_nodes listed as wakeup sources in
> > + * the devicetree
> > */
> > struct ti_sci_info {
> > struct device *dev;
> > @@ -116,6 +119,9 @@ struct ti_sci_info {
> > u8 host_id;
> > /* protected by ti_sci_list_mutex */
> > int users;
> > +
> > + int nr_wakeup_sources;
> > + struct device_node **wakeup_source_nodes;
> > };
> >
> > #define cl_to_ti_sci_info(c) container_of(c, struct ti_sci_info, cl)
> > @@ -392,10 +398,13 @@ static void ti_sci_put_one_xfer(struct ti_sci_xfers_info *minfo,
> > static inline int ti_sci_do_xfer(struct ti_sci_info *info,
> > struct ti_sci_xfer *xfer)
> > {
> > + struct ti_sci_msg_hdr *hdr = (struct ti_sci_msg_hdr *)xfer->tx_message.buf;
> > int ret;
> > int timeout;
> > struct device *dev = info->dev;
> > bool done_state = true;
> > + bool response_expected = !!(hdr->flags & (TI_SCI_FLAG_REQ_ACK_ON_PROCESSED |
> > + TI_SCI_FLAG_REQ_ACK_ON_RECEIVED));
>
> I think a separate patch to introduce a no_response expected patch would
> make sense on which we build tisci_sys_off_handler in the next patch?
>
> >
> > ret = mbox_send_message(info->chan_tx, &xfer->tx_message);
> > if (ret < 0)
> > @@ -403,25 +412,27 @@ static inline int ti_sci_do_xfer(struct ti_sci_info *info,
> >
> > ret = 0;
> >
> > - if (system_state <= SYSTEM_RUNNING) {
> > - /* And we wait for the response. */
> > - timeout = msecs_to_jiffies(info->desc->max_rx_timeout_ms);
> > - if (!wait_for_completion_timeout(&xfer->done, timeout))
> > - ret = -ETIMEDOUT;
> > - } else {
> > - /*
> > - * If we are !running, we cannot use wait_for_completion_timeout
> > - * during noirq phase, so we must manually poll the completion.
> > - */
> > - ret = read_poll_timeout_atomic(try_wait_for_completion, done_state,
> > - done_state, 1,
> > - info->desc->max_rx_timeout_ms * 1000,
> > - false, &xfer->done);
> > - }
> > + if (response_expected) {
>
> How about a goto?
Yes, thanks, looks cleaner.
>
> if (!response_expected)
> goto no_response;
> > + if (system_state <= SYSTEM_RUNNING) {
> > + /* And we wait for the response. */
> > + timeout = msecs_to_jiffies(info->desc->max_rx_timeout_ms);
> > + if (!wait_for_completion_timeout(&xfer->done, timeout))
> > + ret = -ETIMEDOUT;
> > + } else {
> > + /*
> > + * If we are !running, we cannot use wait_for_completion_timeout
> > + * during noirq phase, so we must manually poll the completion.
> > + */
> > + ret = read_poll_timeout_atomic(try_wait_for_completion, done_state,
> > + done_state, 1,
> > + info->desc->max_rx_timeout_ms * 1000,
> > + false, &xfer->done);
> > + }
> >
> > - if (ret == -ETIMEDOUT)
> > - dev_err(dev, "Mbox timedout in resp(caller: %pS)\n",
> > - (void *)_RET_IP_);
> > + if (ret == -ETIMEDOUT)
> > + dev_err(dev, "Mbox timedout in resp(caller: %pS)\n",
> > + (void *)_RET_IP_);
> > + }
> >
> no_response:
>
> > /*
> > * NOTE: we might prefer not to need the mailbox ticker to manage the
> > @@ -3262,6 +3273,82 @@ static int tisci_reboot_handler(struct sys_off_data *data)
> > return NOTIFY_BAD;
> > }
> >
> [...]
>
> > +static int tisci_sys_off_handler(struct sys_off_data *data)
> > +{
> > + struct ti_sci_info *info = data->cb_data;
> > + int i;
> > + int ret;
> > + bool enter_partial_io = false;
> > +
> > + for (i = 0; i != info->nr_wakeup_sources; ++i) {
> > + struct platform_device *pdev =
> > + of_find_device_by_node(info->wakeup_source_nodes[i]);
> > +
> > + if (!pdev)
> > + continue;
> > +
> > + if (device_may_wakeup(&pdev->dev)) {
> > + dev_dbg(info->dev, "%pOFp identified as wakeup source\n",
> > + info->wakeup_source_nodes[i]);
> > + enter_partial_io = true;
> > + }
> > + }
> > +
> > + if (!enter_partial_io)
> > + return NOTIFY_DONE;
> > +
> > + ret = tisci_enter_partial_io(info);
> > +
> > + if (ret) {
> > + dev_err(info->dev,
> > + "Failed to enter Partial-IO %pe, trying to do an emergency restart\n",
> > + ERR_PTR(ret));
> > + emergency_restart();
> > + }
> > +
> > + while (1);
>
> Why not fall through OR go through emergency_restart (since there is
> no fall through for shutdown path) if it acks, but actually fails to
> enter LPM state after a dt described or a default timeout period?
>
> > +
> > + return NOTIFY_DONE;
> > +}
> > +
> > /* Description for K2G */
> > static const struct ti_sci_desc ti_sci_pmmc_k2g_desc = {
> > .default_host_id = 2,
> > @@ -3398,6 +3485,35 @@ static int ti_sci_probe(struct platform_device *pdev)
> > goto out;
> > }
> >
> > + if (of_property_read_bool(dev->of_node, "ti,partial-io-wakeup-sources")) {
>
> You should probably check on TISCI_MSG_QUERY_FW_CAPS[1] if
> Partial IO on low power mode is supported as well? if there is a
> mismatch, report so?
I actually have another series in my queue that introduces this check. I
just implemented this check for Partial-IO yesterday in the patch that
introduces fw capabilities. If you like I can switch these series
around.
>
> > + info->nr_wakeup_sources =
> > + of_count_phandle_with_args(dev->of_node,
> > + "ti,partial-io-wakeup-sources",
> > + NULL);
> > + info->wakeup_source_nodes =
> > + devm_kzalloc(dev, sizeof(*info->wakeup_source_nodes),
> > + GFP_KERNEL);
> > +
> > + for (i = 0; i != info->nr_wakeup_sources; ++i) {
> > + struct device_node *devnode =
> > + of_parse_phandle(dev->of_node,
> > + "ti,partial-io-wakeup-sources",
> > + i);
> > + info->wakeup_source_nodes[i] = devnode;
>
> Curious: Don't we need to maintain reference counting for the devnode
> if CONFIG_OF_DYNAMIC?
In case you mean I missed of_node_put(), yes, I did, thank you. I added
it in a ti_sci_remove().
Best
Markus
Powered by blists - more mailing lists