2023-02-24 15:22:42

by Martin Povišer

Subject: [PATCH 1/3] dmaengine: apple-admac: Handle 'global' interrupt flags

In addition to TX channel and RX channel interrupt flags there's
another class of 'global' interrupt flags with unknown semantics. Those
weren't being handled up to now, and they are the suspected cause of
stuck IRQ states that have been sporadically occurring. Check the global
flags and clear them if raised.

Fixes: b127315d9a78 ("dmaengine: apple-admac: Add Apple ADMAC driver")
Signed-off-by: Martin Povišer <[email protected]>
---
drivers/dma/apple-admac.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/dma/apple-admac.c b/drivers/dma/apple-admac.c
index 90f28bda29c8..00cbfafe0ed9 100644
--- a/drivers/dma/apple-admac.c
+++ b/drivers/dma/apple-admac.c
@@ -75,6 +75,7 @@

#define REG_TX_INTSTATE(idx) (0x0030 + (idx) * 4)
#define REG_RX_INTSTATE(idx) (0x0040 + (idx) * 4)
+#define REG_GLOBAL_INTSTATE(idx) (0x0050 + (idx) * 4)
#define REG_CHAN_INTSTATUS(ch, idx) (0x8010 + (ch) * 0x200 + (idx) * 4)
#define REG_CHAN_INTMASK(ch, idx) (0x8020 + (ch) * 0x200 + (idx) * 4)

@@ -672,13 +673,14 @@ static void admac_handle_chan_int(struct admac_data *ad, int no)
static irqreturn_t admac_interrupt(int irq, void *devid)
{
struct admac_data *ad = devid;
- u32 rx_intstate, tx_intstate;
+ u32 rx_intstate, tx_intstate, global_intstate;
int i;

rx_intstate = readl_relaxed(ad->base + REG_RX_INTSTATE(ad->irq_index));
tx_intstate = readl_relaxed(ad->base + REG_TX_INTSTATE(ad->irq_index));
+ global_intstate = readl_relaxed(ad->base + REG_GLOBAL_INTSTATE(ad->irq_index));

- if (!tx_intstate && !rx_intstate)
+ if (!tx_intstate && !rx_intstate && !global_intstate)
return IRQ_NONE;

for (i = 0; i < ad->nchannels; i += 2) {
@@ -693,6 +695,12 @@ static irqreturn_t admac_interrupt(int irq, void *devid)
rx_intstate >>= 1;
}

+ if (global_intstate) {
+ dev_warn(ad->dev, "clearing unknown global interrupt flag: %x\n",
+ global_intstate);
+ writel_relaxed(~(u32) 0, ad->base + REG_GLOBAL_INTSTATE(ad->irq_index));
+ }
+
return IRQ_HANDLED;
}

--
2.33.0

2023-02-24 15:22:49

by Martin Povišer

Subject: [PATCH 2/3] dmaengine: apple-admac: Fix 'current_tx' not getting freed

In terminate_all we should queue up all submitted descriptors to be
freed. We do that for the content of the 'issued' and 'submitted' lists,
but the 'current_tx' descriptor falls through the cracks as it's
removed from the 'issued' list once it gets assigned to be the current
descriptor. Explicitly queue up freeing of the 'current_tx' descriptor
to address a memory leak that is otherwise present.

Fixes: b127315d9a78 ("dmaengine: apple-admac: Add Apple ADMAC driver")
Signed-off-by: Martin Povišer <[email protected]>
---
drivers/dma/apple-admac.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/dma/apple-admac.c b/drivers/dma/apple-admac.c
index 00cbfafe0ed9..497bb65b7d6d 100644
--- a/drivers/dma/apple-admac.c
+++ b/drivers/dma/apple-admac.c
@@ -512,7 +512,10 @@ static int admac_terminate_all(struct dma_chan *chan)
admac_stop_chan(adchan);
admac_reset_rings(adchan);

- adchan->current_tx = NULL;
+ if (adchan->current_tx) {
+ list_add_tail(&adchan->current_tx->node, &adchan->to_free);
+ adchan->current_tx = NULL;
+ }
/*
* Descriptors can only be freed after the tasklet
* has been killed (in admac_synchronize).
--
2.33.0

2023-03-31 12:52:26

by Vinod Koul

Subject: Re: [PATCH 1/3] dmaengine: apple-admac: Handle 'global' interrupt flags

On 24-02-23, 16:22, Martin Povišer wrote:
> In addition to TX channel and RX channel interrupt flags there's
> another class of 'global' interrupt flags with unknown semantics. Those
> weren't being handled up to now, and they are the suspected cause of
> stuck IRQ states that have been sporadically occurring. Check the global
> flags and clear them if raised.

Applied, thanks

--
~Vinod