All of lore.kernel.org
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Paul Cercueil" <paul.cercueil@bootlin.com>
Cc: devicetree@vger.kernel.org, robh@kernel.org, conor+dt@kernel.org
Subject: Re: [PATCH 2/3] mailbox: marvell-mhu: add Marvell MHU driver
Date: Thu, 19 Nov 2026 15:49:56 +0000	[thread overview]
Message-ID: <sashiko-outbox-223424@kernel.org> (raw)
In-Reply-To: <20261119152944.276763-3-paul.cercueil@bootlin.com>

Thank you for your contribution! Sashiko AI review found 5 potential
issues to consider:
- [High] mailbox: marvell-mhu: manual pci_free_irq_vectors() conflicts
  with devres
- [High] mailbox: marvell-mhu: missing shutdown op causes NULL deref in
  IRQ
- [Medium] mailbox: marvell-mhu: missing PCI dependency in Kconfig
- [Medium] mailbox: marvell-mhu: wrong register and polarity in
  mrvl_mhu_last_tx_done()
- [Low] mailbox: marvell-mhu: duplicate and unused register macro
  definitions
--

commit d88d0044b75f722104999e16b9dd2c9f50a997d0
Author: George Cherian <gcherian@marvell.com>

mailbox: marvell-mhu: add Marvell MHU driver

This commit adds a PCI mailbox controller driver for the Marvell Message
Handling Unit on OcteonTX and OcteonTX2 platforms.

> diff --git a/drivers/mailbox/Kconfig b/drivers/mailbox/Kconfig
> index ab869769e2789..ce3b70749a182 100644
> --- a/drivers/mailbox/Kconfig
> +++ b/drivers/mailbox/Kconfig
> @@ -195,6 +195,17 @@ config MAILBOX_TEST
>  	  Test client to help with testing new Controller driver
>  	  implementations.
>  
> +config MARVELL_MHU
> +	tristate "Marvell Message Handling Unit"
> +	depends on (ARM64 || COMPILE_TEST) && 64BIT
> +	depends on OF
> +	help

[Severity: Medium]
Should MARVELL_MHU also depend on PCI?  Without CONFIG_PCI, building this
driver fails because module_pci_driver() is only defined under CONFIG_PCI
and pcim_request_region() and pcim_iomap() have no !CONFIG_PCI stubs.

[ ... ]

> diff --git a/drivers/mailbox/marvell_mhu.c b/drivers/mailbox/marvell_mhu.c
> new file mode 100644
> index 0000000000000..a037fe104195b
> --- /dev/null
> +++ b/drivers/mailbox/marvell_mhu.c

[ ... ]

> +/*  Register offset: Enable interrupt from SCP to AP */
> +#define XCP0_XCP_DEV0_MBOX_RINT_ENA_W1S	0x000D1C40
> +#define XCP0_XCP_DEV1_MBOX_RINT_ENA_W1S	0x000D1C50
> +#define XCP0_XCP_DEV2_MBOX_RINT_ENA_W1S	0x000D1C60
> +#define XCP0_XCP_DEV3_MBOX_RINT_ENA_W1S	0x000D1C70
> +
> +/* Rx interrupt from SCP to Non-secure AP (linux kernel) */
> +#define XCPX_XCP_DEVY_MBOX_RINT_OFFSET 0x000D1C00
> +#define XCPX_XCP_DEVY_MBOX_RINT(xcp_core, device_id) \
> +	(XCPX_XCP_DEVY_MBOX_RINT_OFFSET | \
> +	((uint64_t)(xcp_core) << 36) | \
> +	((uint64_t)(device_id) << 4))
> +
> +/* The interrupt status register */
> +#define SCP_TO_AP0_MBOX_RINT  XCPX_XCP_DEVY_MBOX_RINT(SCP_INDEX, DEV_AP0)
> +
> +#define XCPX_XCP_DEVY_MBOX_RINT_OFFSET	0x000D1C00
> +#define XCPX_XCP_DEVY_MBOX_RINT(xcp_core, device_id) \
> +	(XCPX_XCP_DEVY_MBOX_RINT_OFFSET | \
> +	((uint64_t)(xcp_core) << 36) | \
> +	((uint64_t)(device_id) << 4))
> +
> +#define SCP_TO_AP0_MBOX_RINT  XCPX_XCP_DEVY_MBOX_RINT(SCP_INDEX, DEV_AP0)
> +#define SCP_TO_DEV0 XCPX_XCP_DEVY_MBOX_RINT(0, 0)
> +#define SCP_TO_DEV1 XCPX_XCP_DEVY_MBOX_RINT(0, 1)
> +#define SCP_TO_DEV2 XCPX_XCP_DEVY_MBOX_RINT(0, 2)
> +#define SCP_TO_DEV3 XCPX_XCP_DEVY_MBOX_RINT(0, 3)

[Severity: Low]
This isn't a bug, but XCPX_XCP_DEVY_MBOX_RINT_OFFSET,
XCPX_XCP_DEVY_MBOX_RINT, and SCP_TO_AP0_MBOX_RINT are defined twice here,
and XCP0_XCP_DEV0/1/3_MBOX_RINT_ENA_W1S and SCP_TO_DEV0..3 are unused.

> +
> +struct mrvl_mhu {
> +	/* SCP link information */
> +	void __iomem *base; /* tx_reg, rx_reg */
> +
> +	struct mbox_controller ctrl;
> +	struct mbox_chan chan;
> +};
> +
> +static irqreturn_t mrvl_mhu_rx_interrupt(int irq, void *p)
> +{
> +	struct mrvl_mhu *mhu = p;
> +	u64 val;
> +
> +	/* Read interrupt status register */
> +	val = readq_relaxed(mhu->base + SCP_TO_AP0_MBOX_RINT);
> +	if (val) {
> +		/* Clear the interrupt : Write on clear */
> +		writeq_relaxed(1ul, mhu->base + SCP_TO_AP0_MBOX_RINT);
> +		/* Poke mbox */
> +		mbox_chan_received_data(&mhu->chan, (void *)&val);
> +		return IRQ_HANDLED;
> +	}
> +
> +	return IRQ_NONE;
> +}
> +
> +static int mrvl_mhu_send_data(struct mbox_chan *chan, void *data)
> +{
> +	const u64 doorbell = 0xff;  /* Arbitrary value, SCP requires a value */
> +	struct mrvl_mhu *mhu = chan->con_priv;
> +
> +	/* signal SCP */
> +	iowrite64(doorbell, mhu->base + AP0_TO_SCP_MBOX);
> +
> +	return 0;
> +}
> +
> +static bool mrvl_mhu_last_tx_done(struct mbox_chan *chan)
> +{
> +	struct mrvl_mhu *mhu = chan->con_priv;
> +	u64 status;
> +
> +	status = ioread64(mhu->base + XCPX_XCP_DEVY_MBOX_RINT(0, 2));
> +
> +	return status != 0;
> +}

[Severity: Medium]
Does mrvl_mhu_last_tx_done() check the wrong register and polarity?

XCPX_XCP_DEVY_MBOX_RINT(0, 2) is SCP_TO_AP0_MBOX_RINT, the SCP-to-AP0 RX
interrupt status register, which mrvl_mhu_rx_interrupt() immediately clears
to 0 in hardirq context when an RX interrupt fires.  Consequently, any
client relying on the controller's txdone_poll = true will read 0 and time
out.  Meanwhile, the AP0-to-SCP TX interrupt status macro
AP0_TO_SCP_MBOX_LINT defined earlier in this file is unused, and its status
bit is cleared to 0 when the SCP consumes the doorbell (status == 0).

> +
> +static int mrvl_mhu_startup(struct mbox_chan *chan)
> +{
> +	struct mrvl_mhu *mhu = chan->con_priv;
> +
> +	/* Enable interrupts only if there is client for data */
> +	writeq_relaxed(1ul, mhu->base + XCP0_XCP_DEV2_MBOX_RINT_ENA_W1S);
> +
> +	return 0;
> +}
> +
> +/* Define operations related to mbox */
> +static const struct mbox_chan_ops mrvl_mhu_chan_ops = {
> +	.startup = mrvl_mhu_startup,
> +	.send_data = mrvl_mhu_send_data,
> +	.last_tx_done = mrvl_mhu_last_tx_done,
> +};

[Severity: High]
Can an SCP interrupt after mbox_free_channel() cause a NULL pointer
dereference in mrvl_mhu_rx_interrupt() -> mbox_chan_received_data()?

mrvl_mhu_startup() enables the hardware interrupt via
XCP0_XCP_DEV2_MBOX_RINT_ENA_W1S, but mrvl_mhu_chan_ops provides no
.shutdown callback to disable it when a client releases the channel (or if
the interrupt was already left enabled before a client binds).  Once
mbox_free_channel() sets chan->cl = NULL, an incoming interrupt calls
mbox_chan_received_data(&mhu->chan, ...), which unconditionally
dereferences chan->cl->rx_callback.  In addition, mrvl_mhu_rx_interrupt()
clears SCP_TO_AP0_MBOX_RINT before mbox_chan_received_data() finishes
processing the shared memory payload.

[ ... ]

> +	pci_set_drvdata(pdev, mhu);
> +	ret = pcim_enable_device(pdev);
> +	if (ret)
> +		return ret;
> +
> +	ret = pcim_request_region(pdev, BAR0, DRIVER_NAME);
> +	if (ret)
> +		return ret;
> +
> +	mhu->base = pcim_iomap(pdev, BAR0, pci_resource_len(pdev, BAR0));
> +	if (!mhu->base)
> +		return -EINVAL;
> +
> +	/* Configure IRQ from SCP to AP */
> +	ret = mrvl_mhu_setup_irq(pdev);
> +	if (ret)
> +		return ret;
> +
> +	ret = mbox_controller_register(&mhu->ctrl);
> +	if (ret) {
> +		pci_free_irq_vectors(pdev);
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static void mrvl_mhu_remove(struct pci_dev *pdev)
> +{
> +	struct mrvl_mhu *mhu = pci_get_drvdata(pdev);
> +
> +	mbox_controller_unregister(&mhu->ctrl);
> +	pci_free_irq_vectors(pdev);
> +}

[Severity: High]
Does calling pci_free_irq_vectors() here tear down the MSI-X IRQ descriptor
while the devm_request_irq() handler is still registered?

Because mrvl_mhu_probe() calls pcim_enable_device(), pci_alloc_irq_vectors()
is already managed by devres (via pcim_msi_release()).  Calling
pci_free_irq_vectors() manually in mrvl_mhu_remove() and on the
mbox_controller_register() failure path frees the IRQ descriptors before
devres runs devm_irq_release(), which triggers a warning when removing the
populated procfs IRQ directory, causes free_irq() to fail to find the
descriptor and leak struct irqaction, and then runs pci_free_irq_vectors()
a second time during devres teardown.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261119152944.276763-1-paul.cercueil@bootlin.com?part=2

  reply	other threads:[~2026-11-19 15:49 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-19 15:29 [PATCH 0/3] mailbox: Add Marvell MHU driver Paul Cercueil
2026-11-19 15:29 ` [PATCH 1/3] dt-bindings: mailbox: add Marvell MHU Paul Cercueil
2026-11-19 15:38   ` sashiko-bot
2026-11-19 15:29 ` [PATCH 2/3] mailbox: marvell-mhu: add Marvell MHU driver Paul Cercueil
2026-11-19 15:49   ` sashiko-bot [this message]
2026-11-19 15:29 ` [PATCH 3/3] scmi: driver: Increase timeout for rx response Paul Cercueil
2026-11-19 15:38   ` sashiko-bot
2026-11-19 17:32   ` Sudeep Holla
2026-11-19 15:54 ` [PATCH 0/3] mailbox: Add Marvell MHU driver Paul Cercueil

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-223424@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=conor+dt@kernel.org \
    --cc=devicetree@vger.kernel.org \
    --cc=paul.cercueil@bootlin.com \
    --cc=robh@kernel.org \
    --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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.