Re: [PATCH v6 2/2] mfd: loongson-se: Fix miscellaneous issues
From: Huacai Chen
Date: Mon Sep 07 2026 - 10:12:27 EST
Hi, Qunqin,
On Mon, Sep 7, 2026 at 10:38 AM Qunqin Zhao <zhaoqunqin@xxxxxxx> wrote:
>
> From: Qunqin Zhao <zhaoqunqin@xxxxxxxxxxx>
>
> Address multiple historical driver issues discovered by the Sashiko
> Automation system within the loongson_se_probe() initialization flow
> and the driver's interrupt service routines [1].
>
> - Add an explicit bounds check in se_irq_handler() before accessing
> the engines array to prevent an out-of-bounds memory access.
>
> - Switch from devm_kmalloc() to devm_kzalloc() and explicitly
> initialize all engine completion structures in probe() to avoid a
> kernel panic from complete() dereferencing a NULL wait head when a
> spurious interrupt fires before child drivers call
> loongson_se_init_engine().
>
> - Serialize command submission with a new poll_lock spinlock and
> protect the completion reset with the interrupt lock. Before
> issuing a command, mask the target interrupt, clear any pending
> status, and reinitialize the completion under dev_lock, then
> re-enable the interrupt after the command has been triggered. This
> closes the race where a stale interrupt from a previously
> interrupted command could complete the current command's
> completion. The spin_lock_irq() around the 10 ms poll is dropped so
> interrupts are not disabled for the full busy-wait.
>
> - Fix EPROBE_DEFER handling: propagate the error directly from
> platform_irq_count() instead of overwriting it with ENODEV so that
> probe deferral works when the interrupt provider is not yet ready.
>
> - Validate dmam_size from firmware against the minimum required size
> to keep the command buffers of all engines within engine 0's data
> region and prevent overlapping DMA buffers.
>
> - Return the error code from devm_request_irq() instead of silently
> continuing to prevent an indefinite hang.
>
> - Add a loongson_se_stop() cleanup handler registered with
> devm_add_action_or_reset() before loongson_se_init() so that a
> failed init still stops the controller and masks all interrupts via
> devres, preventing DMA access to freed memory.
>
> - Zero-initialize the local controller command structures in
> loongson_se_init() and loongson_se_init_engine() to prevent
> uninitialized stack data from being written to device registers.
>
> - Add the SE_CMD_STOP command definition.
>
> Link: https://lore.kernel.org/all/20260618095949.GB1672911@xxxxxxxxxx/ [1]
> Fixes: e551fa3159e3 ("mfd: Add support for Loongson Security Engine chip controller")
> Signed-off-by: Qunqin Zhao <zhaoqunqin@xxxxxxxxxxx>
> ---
> drivers/mfd/loongson-se.c | 90 +++++++++++++++++++++++++++++----
> include/linux/mfd/loongson-se.h | 1 +
> 2 files changed, 82 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/mfd/loongson-se.c b/drivers/mfd/loongson-se.c
> index 7f552a8ee..1d332c418 100644
> --- a/drivers/mfd/loongson-se.c
> +++ b/drivers/mfd/loongson-se.c
> @@ -23,6 +23,8 @@
> struct loongson_se {
> void __iomem *base;
> spinlock_t dev_lock;
> + /* Synchronizes command submission between users of different engines */
> + spinlock_t poll_lock;
Can't we reuse dev_lock?
Huacai
> struct completion cmd_completion;
>
> void *dmam_base;
> @@ -42,7 +44,7 @@ static int loongson_se_poll(struct loongson_se *se, u32 int_bit)
> u32 status;
> int err;
>
> - spin_lock_irq(&se->dev_lock);
> + spin_lock(&se->poll_lock);
>
> /* Notify the controller that the engine needs to be started */
> writel(int_bit, se->base + SE_L2SINT_SET);
> @@ -52,17 +54,48 @@ static int loongson_se_poll(struct loongson_se *se, u32 int_bit)
> !(status & int_bit),
> 1, LOONGSON_ENGINE_CMD_TIMEOUT_US);
>
> - spin_unlock_irq(&se->dev_lock);
> + /*
> + * Re-enable the interrupt that loongson_se_reinit_completion() masked.
> + * The hardware guarantees that once the interrupt is re-enabled, only
> + * interrupts for the command just issued can arrive, so a stale
> + * interrupt from a previously interrupted command can never complete
> + * this command's completion.
> + */
> + writel(int_bit | readl(se->base + SE_S2LINT_EN), se->base + SE_S2LINT_EN);
> +
> + spin_unlock(&se->poll_lock);
>
> return err;
> }
>
> +/*
> + * Prepare a completion for a new command: mask the corresponding interrupt,
> + * clear any pending interrupt status, and reset the completion. This runs
> + * under dev_lock so that the IRQ handler cannot race with it. The interrupt
> + * is re-enabled in loongson_se_poll() after the command has been issued.
> + */
> +static void loongson_se_reinit_completion(struct loongson_se *se,
> + struct completion *completion, u32 int_bit)
> +{
> + spin_lock_irq(&se->dev_lock);
> +
> + writel(readl(se->base + SE_S2LINT_EN) & ~int_bit, se->base + SE_S2LINT_EN);
> +
> + writel(int_bit, se->base + SE_S2LINT_CL);
> +
> + reinit_completion(completion);
> +
> + spin_unlock_irq(&se->dev_lock);
> +}
> +
> static int loongson_se_send_controller_cmd(struct loongson_se *se,
> struct loongson_se_controller_cmd *cmd)
> {
> u32 *send_cmd = (u32 *)cmd;
> int err, i;
>
> + loongson_se_reinit_completion(se, &se->cmd_completion, SE_INT_CONTROLLER);
> +
> for (i = 0; i < SE_SEND_CMD_REG_LEN; i++)
> writel(send_cmd[i], se->base + SE_SEND_CMD_REG + i * 4);
>
> @@ -75,12 +108,16 @@ static int loongson_se_send_controller_cmd(struct loongson_se *se,
>
> int loongson_se_send_engine_cmd(struct loongson_se_engine *engine)
> {
> + int err;
> +
> + loongson_se_reinit_completion(engine->se, &engine->completion, BIT(engine->id));
> +
> /*
> * After engine initialization, the controller already knows
> * where to obtain engine commands from. Now all we need to
> * do is notify the controller that the engine needs to be started.
> */
> - int err = loongson_se_poll(engine->se, BIT(engine->id));
> + err = loongson_se_poll(engine->se, BIT(engine->id));
>
> if (err)
> return err;
> @@ -93,7 +130,7 @@ struct loongson_se_engine *loongson_se_init_engine(struct device *dev, int id)
> {
> struct loongson_se *se = dev_get_drvdata(dev);
> struct loongson_se_engine *engine = &se->engines[id];
> - struct loongson_se_controller_cmd cmd;
> + struct loongson_se_controller_cmd cmd = {0};
>
> engine->se = se;
> engine->id = id;
> @@ -155,7 +192,8 @@ static irqreturn_t se_irq_handler(int irq, void *dev_id)
> /* For engines */
> while (int_status) {
> id = __ffs(int_status);
> - complete(&se->engines[id].completion);
> + if (id < SE_ENGINE_MAX)
> + complete(&se->engines[id].completion);
> int_status &= ~BIT(id);
> writel(BIT(id), se->base + SE_S2LINT_CL);
> }
> @@ -167,7 +205,7 @@ static irqreturn_t se_irq_handler(int irq, void *dev_id)
>
> static int loongson_se_init(struct loongson_se *se, dma_addr_t addr, int size)
> {
> - struct loongson_se_controller_cmd cmd;
> + struct loongson_se_controller_cmd cmd = {0};
> int err;
>
> cmd.command_id = SE_CMD_START;
> @@ -188,6 +226,17 @@ static const struct mfd_cell engines[] = {
> { .name = "tpm_loongson" },
> };
>
> +static void loongson_se_stop(void *data)
> +{
> + struct loongson_se *se = data;
> + struct loongson_se_controller_cmd cmd = {0};
> +
> + cmd.command_id = SE_CMD_STOP;
> + loongson_se_send_controller_cmd(se, &cmd);
> +
> + writel(0, se->base + SE_S2LINT_EN);
> +}
> +
> static int loongson_se_probe(struct platform_device *pdev)
> {
> struct device *dev = &pdev->dev;
> @@ -195,19 +244,34 @@ static int loongson_se_probe(struct platform_device *pdev)
> int nr_irq, irq, err, i;
> dma_addr_t paddr;
>
> - se = devm_kmalloc(dev, sizeof(*se), GFP_KERNEL);
> + se = devm_kzalloc(dev, sizeof(*se), GFP_KERNEL);
> if (!se)
> return -ENOMEM;
>
> dev_set_drvdata(dev, se);
> init_completion(&se->cmd_completion);
> spin_lock_init(&se->dev_lock);
> + spin_lock_init(&se->poll_lock);
> mutex_init(&se->engine_init_lock);
>
> + for (i = 0; i < SE_ENGINE_MAX; i++)
> + init_completion(&se->engines[i].completion);
> +
> dma_set_mask_and_coherent(dev, DMA_BIT_MASK(64));
> if (device_property_read_u32(dev, "dmam_size", &se->dmam_size))
> return -ENODEV;
>
> + /*
> + * Engine 0 does not exist and its data buffer is reused as the command
> + * buffer for the other engines. The command buffers of all engines
> + * therefore must fit within engine 0's data region, i.e.
> + * dmam_size / SE_ENGINE_MAX >= SE_ENGINE_MAX * 2 * SE_ENGINE_CMD_SIZE.
> + * Enforce this to keep each engine's data and command buffers from
> + * overlapping.
> + */
> + if ((se->dmam_size / SE_ENGINE_MAX) < (SE_ENGINE_MAX * 2 * SE_ENGINE_CMD_SIZE))
> + return -EINVAL;
> +
> se->dmam_base = dmam_alloc_coherent(dev, se->dmam_size, &paddr, GFP_KERNEL);
> if (!se->dmam_base)
> return -ENOMEM;
> @@ -217,20 +281,28 @@ static int loongson_se_probe(struct platform_device *pdev)
> return PTR_ERR(se->base);
>
> nr_irq = platform_irq_count(pdev);
> - if (nr_irq <= 0)
> + if (nr_irq == 0)
> return -ENODEV;
> + if (nr_irq < 0)
> + return nr_irq;
>
> writel(SE_INT_ALL, se->base + SE_S2LINT_CL);
>
> for (i = 0; i < nr_irq; i++) {
> irq = platform_get_irq(pdev, i);
> err = devm_request_irq(dev, irq, se_irq_handler, IRQF_SHARED, "loongson-se", se);
> - if (err)
> + if (err) {
> dev_err(dev, "failed to request IRQ: %d\n", irq);
> + return err;
> + }
> }
>
> writel(SE_INT_ALL, se->base + SE_S2LINT_EN);
>
> + err = devm_add_action_or_reset(dev, loongson_se_stop, se);
> + if (err)
> + return err;
> +
> err = loongson_se_init(se, paddr, se->dmam_size);
> if (err)
> return err;
> diff --git a/include/linux/mfd/loongson-se.h b/include/linux/mfd/loongson-se.h
> index 07afa0c25..8237ccab7 100644
> --- a/include/linux/mfd/loongson-se.h
> +++ b/include/linux/mfd/loongson-se.h
> @@ -9,6 +9,7 @@
> #define SE_SEND_CMD_REG_LEN 0x8
> /* Controller command ID */
> #define SE_CMD_START 0x0
> +#define SE_CMD_STOP 0x1
> #define SE_CMD_SET_DMA 0x3
> #define SE_CMD_SET_ENGINE_CMDBUF 0x4
>
> --
> 2.47.2
>