Linux kernel and device drivers for NXP i.MX platforms
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Dario Binacchi" <dario.binacchi@amarulasolutions.com>
Cc: imx@lists.linux.dev, Frank.Li@kernel.org
Subject: Re: [PATCH v2 3/4] pmdomain: imx8m-blk-ctrl: Keep the domains left on by the bootloader
Date: Sun, 22 Nov 2026 10:20:33 +0000	[thread overview]
Message-ID: <sashiko-outbox-227889@kernel.org> (raw)
In-Reply-To: <20261122095927.2631909-4-dario.binacchi@amarulasolutions.com>

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] pmdomain: imx8m-blk-ctrl: unlocked bus domain check in
  imx8m_blk_ctrl_is_off()
- [Medium] pmdomain: imx8m-blk-ctrl: put_boot_on() races with genpd
  power_off_work
--

commit c08d94b261d674a0f3a8ba6bcb7abe4442f21fa6
Author: Dario Binacchi <dario.binacchi@amarulasolutions.com>

pmdomain: imx8m-blk-ctrl: Keep the domains left on by the bootloader

This commit reads the blk-ctrl registers at probe time on supported SoCs
(i.MX8MN DISP) to keep power domains left on by the bootloader active
until their consumers release them.

> diff --git a/drivers/pmdomain/imx/imx8m-blk-ctrl.c b/drivers/pmdomain/imx/imx8m-blk-ctrl.c
> index 7d6735728d8c6..5ab201eefeb38 100644
> --- a/drivers/pmdomain/imx/imx8m-blk-ctrl.c
> +++ b/drivers/pmdomain/imx/imx8m-blk-ctrl.c
[ ... ]
> @@ -187,11 +190,80 @@ static int imx8m_blk_ctrl_power_off(struct generic_pm_domain *genpd)
>  	/* allow bus domain to suspend */
>  	pm_runtime_put(bc->bus_power_dev);
>  
> +	domain->boot_on = false;
> +
>  	return 0;
>  }
>  
>  static struct lock_class_key blk_ctrl_genpd_lock_class;
>  
> +/*
> + * A domain left on by the bootloader has its resets deasserted and its
> + * clocks enabled. The registers can only be read if the bus domain is
> + * on, so a domain is considered off whenever the bus domain is off.
> + */
> +static bool imx8m_blk_ctrl_is_off(struct imx8m_blk_ctrl *bc,
> +				  const struct imx8m_blk_ctrl_data *bc_data,
> +				  const struct imx8m_blk_ctrl_domain_data *data)
> +{
> +	u32 rst, clk, mipi = 0;
> +
> +	if (!bc_data->boot_state_readable ||
> +	    !dev_pm_genpd_is_on(bc->bus_power_dev))
> +		return true;
> +
> +	if (regmap_read(bc->regmap, BLK_SFT_RSTN, &rst) ||
> +	    regmap_read(bc->regmap, BLK_CLK_EN, &clk))
> +		return true;
> +
> +	if (data->mipi_phy_rst_mask &&
> +	    regmap_read(bc->regmap, BLK_MIPI_RESET_DIV, &mipi))
> +		return true;
> +
> +	return (rst & data->rst_mask) != data->rst_mask ||
> +	       (clk & data->clk_mask) != data->clk_mask ||
> +	       (mipi & data->mipi_phy_rst_mask) != data->mipi_phy_rst_mask;
> +}

[Severity: High]
Can the upstream bus domain (pgc_dispmix) power down asynchronously on
pm_wq between dev_pm_genpd_is_on(bc->bus_power_dev) and the regmap_read()
calls?

dev_pm_genpd_is_on() drops the genpd lock before returning, and no
runtime PM reference is held on bc->bus_power_dev during the domain loop
in imx8m_blk_ctrl_probe().  If DISPMIX was off at boot, an earlier probe
of a child domain such as pgc_mipi powers pgc_dispmix on with
PD_FLAG_ATTACH_POWER_ON, and dev_pm_domain_attach_by_name() in
imx8m_blk_ctrl_probe() queues pgc_dispmix->power_off_work on pm_wq right
before imx8m_blk_ctrl_is_off() runs:

CPU0                                  kworker (pm_wq)
imx8m_blk_ctrl_probe()
  dev_pm_domain_attach_by_name()
    genpd_queue_power_off_work()
  imx8m_blk_ctrl_is_off()
    dev_pm_genpd_is_on() -> true
                                      genpd_power_off_work_fn()
                                        imx_pgc_power_down(pgc_dispmix)
    regmap_read(bc->regmap, ...)  <-- MMIO read with DISPMIX powered off

Reading the disp_blk_ctrl registers after DISPMIX and its
DISP_AXI_ROOT/DISP_APB_ROOT clocks are disabled causes a bus hang or
synchronous external abort during boot.

> +
> +/* Take the references of imx8m_blk_ctrl_power_on() for a domain left on */
> +static int imx8m_blk_ctrl_get_boot_on(struct imx8m_blk_ctrl_domain *domain)
> +{
> +	struct imx8m_blk_ctrl *bc = domain->bc;
> +	int ret;
> +
> +	if (!domain->boot_on)
> +		return 0;
> +
> +	ret = pm_runtime_resume_and_get(bc->bus_power_dev);
> +	if (ret < 0) {
> +		dev_err(bc->dev, "failed to power up bus domain\n");
> +		return ret;
> +	}
> +
> +	ret = pm_runtime_resume_and_get(domain->power_dev);
> +	if (ret < 0) {
> +		dev_err(bc->dev, "failed to power up peripheral domain\n");
> +		pm_runtime_put(bc->bus_power_dev);
> +		return ret;
> +	}
> +
> +	if (icc_bulk_set_bw(domain->num_paths, domain->paths))
> +		dev_err(bc->dev, "failed to set icc bw\n");
> +
> +	return 0;
> +}
> +
> +static void imx8m_blk_ctrl_put_boot_on(struct imx8m_blk_ctrl_domain *domain)
> +{
> +	if (!domain->boot_on)
> +		return;
> +
> +	/* Sync, to power down the bus domain before its notifier is removed */
> +	pm_runtime_put_sync(domain->power_dev);
> +	pm_runtime_put_sync(domain->bc->bus_power_dev);
> +	domain->boot_on = false;
> +}
[ ... ]
> @@ -353,6 +439,10 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
>  	dev_pm_genpd_remove_notifier(bc->bus_power_dev);
>  cleanup_provider:
>  	of_genpd_del_provider(dev->of_node);
> +cleanup_boot_on:
> +	while (i--)
> +		imx8m_blk_ctrl_put_boot_on(&bc->domains[i]);
> +	i = bc_data->num_domains;
>  cleanup_pds:
>  	for (i--; i >= 0; i--) {
>  		pm_genpd_remove(&bc->domains[i].genpd);
> @@ -374,6 +464,7 @@ static void imx8m_blk_ctrl_remove(struct platform_device *pdev)
>  	for (i = 0; i < bc->onecell_data.num_domains; i++) {
>  		struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
>  
> +		imx8m_blk_ctrl_put_boot_on(domain);
>  		pm_genpd_remove(&domain->genpd);
>  		dev_pm_domain_detach(domain->power_dev, true);
>  	}

[Severity: Medium]
Can imx8m_blk_ctrl_put_boot_on() race with domain->genpd.power_off_work
when called before pm_genpd_remove()?

Both imx8m_blk_ctrl_remove() and the cleanup_boot_on path in
imx8m_blk_ctrl_probe() call imx8m_blk_ctrl_put_boot_on() before
pm_genpd_remove() cancels power_off_work, without taking genpd_lock or
transitioning domain->genpd.status to GENPD_STATE_OFF.  If power_off_work
was queued on pm_wq (for example by genpd_power_off_unused() or when a
consumer detaches), both paths drop the same runtime PM references:

CPU0                                  kworker (pm_wq)
imx8m_blk_ctrl_remove()
  imx8m_blk_ctrl_put_boot_on(domain)
    pm_runtime_put_sync(power_dev)
    pm_runtime_put_sync(bus_power_dev)
    domain->boot_on = false
  pm_genpd_remove(&domain->genpd)
    cancel_work_sync(&power_off_work)
                                      genpd_power_off_work_fn()
                                        genpd_power_off()
                                          imx8m_blk_ctrl_power_off()
                                            regmap_clear_bits(...)
                                            pm_runtime_put(power_dev)
                                            pm_runtime_put(bus_power_dev)
                                              <-- double put & MMIO while
                                                  bus_power_dev is off

Calling pm_genpd_remove(&domain->genpd) before
imx8m_blk_ctrl_put_boot_on(domain) (matching imx_pgc_domain_remove() in
drivers/pmdomain/imx/gpcv2.c) ensures power_off_work is flushed before
checking domain->boot_on.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261122095927.2631909-1-dario.binacchi@amarulasolutions.com?part=3

  reply	other threads:[~2026-11-22 10:20 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-22  9:59 [PATCH v2 0/4] pmdomain: imx: Keep the domains left on by the bootloader Dario Binacchi
2026-11-22  9:59 ` [PATCH v2 1/4] pmdomain: imx8m-blk-ctrl: Remove the power notifier on probe failure Dario Binacchi
2026-11-22  9:59 ` [PATCH v2 2/4] pmdomain: imx: gpcv2: Keep the domains left on by the bootloader Dario Binacchi
2026-11-22 10:12   ` sashiko-bot
2026-11-22  9:59 ` [PATCH v2 3/4] pmdomain: imx8m-blk-ctrl: " Dario Binacchi
2026-11-22 10:20   ` sashiko-bot [this message]
2026-11-22  9:59 ` [PATCH v2 4/4] pmdomain: imx8m-blk-ctrl: Hold boot-on domain clocks until sync_state Dario Binacchi

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=sashiko-outbox-227889@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=Frank.Li@kernel.org \
    --cc=dario.binacchi@amarulasolutions.com \
    --cc=imx@lists.linux.dev \
    --cc=sashiko-reviews@lists.linux.dev \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox