[PATCH v2] mmc: meson-gx: fix SDIO interrupt handling
Heiner Kallweit
hkallweit1 at gmail.com
Tue Nov 22 13:41:30 PST 2022
On 22.11.2022 14:23, Peter Suti wrote:
> With the interrupt support introduced in commit 066ecde sometimes the
> Marvell-8987 wifi chip entered a deadlock using the marvell-sd-uapsta-8987
> vendor driver. The cause seems to be that sometimes the interrupt handler
> handles 2 IRQs and one of them disables the interrupts which are not reenabled
> when all interrupts are finished. To work around this, disable all interrupts
> when we are in the IRQ context and reenable them when the current IRQ is handled.
>
To me it's still not clear what you mean with "handles 2 IRQs".
Hard irq handlers aren't re-entrant. Can you provide the exact call sequence
for the issue you're facing?
Are SDIO interrupts handled on more than one CPU in your case?
Are you concerned about the hard irq handler running in parallel on more than one CPU?
The proposed patch looks hacky and somewhat bypasses the SDIO core logic
by partially re-enabling SDIO interrupts in the hard irq handler.
In the first review round I wrote the following but didn't see a feedback yet.
Did you check the linked discussion whether it may be related to your issue?
-> IIRC I had a similar/same problem in mind when discussing the following:
-> https://lore.kernel.org/linux-arm-kernel/CAPDyKFoameOb7d3cn8_ki1O6DbMEAFvkQh1uUsYp4S-Lkq41oQ@mail.gmail.com/
-> Not sure though whether it's related to the issue you're facing.
> Fixes: 066ecde ("mmc: meson-gx: add SDIO interrupt support")
>
> Signed-off-by: Peter Suti <peter.suti at streamunlimited.com>
> ---
> Changes in v2:
> - use spin_lock instead of spin_lock_irqsave
> - only reenable interrupts if they were enabled already
>
> drivers/mmc/host/meson-gx-mmc.c | 30 +++++++++++++++++++++++-------
> 1 file changed, 23 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
> index 6e5ea0213b47..0c95f8640b34 100644
> --- a/drivers/mmc/host/meson-gx-mmc.c
> +++ b/drivers/mmc/host/meson-gx-mmc.c
> @@ -934,6 +934,13 @@ static void meson_mmc_read_resp(struct mmc_host *mmc, struct mmc_command *cmd)
> }
> }
>
> +static bool __meson_mmc_sdio_irq_is_enabled(struct mmc_host *mmc)
> +{
> + struct meson_host *host = mmc_priv(mmc);
> +
> + return readl(host->regs + SD_EMMC_IRQ_EN) & IRQ_SDIO;
> +}
> +
> static void __meson_mmc_enable_sdio_irq(struct mmc_host *mmc, int enable)
> {
> struct meson_host *host = mmc_priv(mmc);
> @@ -950,6 +957,11 @@ static irqreturn_t meson_mmc_irq(int irq, void *dev_id)
> struct mmc_command *cmd;
> u32 status, raw_status;
> irqreturn_t ret = IRQ_NONE;
> + bool irq_enabled;
> +
> + spin_lock(&host->lock);
> + irq_enabled = __meson_mmc_sdio_irq_is_enabled(host->mmc);
> + __meson_mmc_enable_sdio_irq(host->mmc, 0);
>
> raw_status = readl(host->regs + SD_EMMC_STATUS);
> status = raw_status & (IRQ_EN_MASK | IRQ_SDIO);
> @@ -958,11 +970,11 @@ static irqreturn_t meson_mmc_irq(int irq, void *dev_id)
> dev_dbg(host->dev,
> "Unexpected IRQ! irq_en 0x%08lx - status 0x%08x\n",
> IRQ_EN_MASK | IRQ_SDIO, raw_status);
> - return IRQ_NONE;
> + goto out_unlock;
> }
>
> if (WARN_ON(!host))
> - return IRQ_NONE;
> + goto out_unlock;
>
> /* ack all raised interrupts */
> writel(status, host->regs + SD_EMMC_STATUS);
> @@ -970,17 +982,16 @@ static irqreturn_t meson_mmc_irq(int irq, void *dev_id)
> cmd = host->cmd;
>
> if (status & IRQ_SDIO) {
> - spin_lock(&host->lock);
> - __meson_mmc_enable_sdio_irq(host->mmc, 0);
> sdio_signal_irq(host->mmc);
> - spin_unlock(&host->lock);
> status &= ~IRQ_SDIO;
> - if (!status)
> + if (!status) {
> + spin_unlock(&host->lock);
> return IRQ_HANDLED;
> + }
> }
>
> if (WARN_ON(!cmd))
> - return IRQ_NONE;
> + goto out_unlock;
>
> cmd->error = 0;
> if (status & IRQ_CRC_ERR) {
> @@ -1023,6 +1034,11 @@ static irqreturn_t meson_mmc_irq(int irq, void *dev_id)
> if (ret == IRQ_HANDLED)
> meson_mmc_request_done(host->mmc, cmd->mrq);
>
> +out_unlock:
> + if (irq_enabled)
> + __meson_mmc_enable_sdio_irq(host->mmc, 1);
> + spin_unlock(&host->lock);
> +
> return ret;
> }
>
More information about the linux-amlogic
mailing list