11.01.2021 22:31, Dmitry Osipenko пишет:

11.01.2021 19:08, Mikko Perttunen пишет:

quoted

Upon a communication error, the interrupt handler can call
tegra_i2c_disable_packet_mode. This causes a sleeping poll to happen
unless the current transaction was marked atomic. Fix this by
making the poll happen atomically if we are in an IRQ.

This matches the behavior prior to the patch mentioned
in the Fixes tag.

Fixes: ede2299f7101 ("i2c: tegra: Support atomic transfers")
Cc: stable@vger.kernel.org
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
---
v2:
* Use in_irq() instead of passing a flag from the ISR.
  Thanks to Dmitry for the suggestion.
* Update commit message.
---
 drivers/i2c/busses/i2c-tegra.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
index 6f08c0c3238d..0727383f4940 100644
--- a/drivers/i2c/busses/i2c-tegra.c
+++ b/drivers/i2c/busses/i2c-tegra.c

@@ -533,7 +533,7 @@ static int tegra_i2c_poll_register(struct tegra_i2c_dev *i2c_dev,
 	void __iomem *addr = i2c_dev->base + tegra_i2c_reg_addr(i2c_dev, reg);
 	u32 val;
 
-	if (!i2c_dev->atomic_mode)
+	if (!i2c_dev->atomic_mode && !in_irq())
 		return readl_relaxed_poll_timeout(addr, val, !(val & mask),
 						  delay_us, timeout_us);
 

Reviewed-by: Dmitry Osipenko <digetx@gmail.com>

Perhaps a follow up change could be to use a threaded interrupt context,
I'll type a patch for that.