diff mbox series

[5/7] Input: synaptics-rmi4 - don't do unaligned reads in IRQ context

Message ID 20230929-caleb-rmi4-quirks-v1-5-cc3c703f022d@linaro.org
State Superseded
Headers show
Series [1/7] dt-bindings: input: syna,rmi4: document syna,pdt-fallback-desc | expand

Commit Message

Caleb Connolly Sept. 30, 2023, 5:08 p.m. UTC 
From: methanal <baclofen@tuta.io>

Some replacement displays include third-party touch ICs which incur a
significant penalty (1-2 seconds) when doing certain unaligned reads.
This is enough to break functionality when it happens in the hot path,
so adjust the interrupt handler to not read from an unaligned address.

Signed-off-by: methanal <baclofen@tuta.io>
Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
This solution does wind up doing the unaligned reads on the CPU instead,
although I'm not sure how significant of a penalty this is in practise.
---
 drivers/input/rmi4/rmi_driver.c | 20 +++++++++++++-------
 1 file changed, 13 insertions(+), 7 deletions(-)
diff mbox series

Patch

diff --git a/drivers/input/rmi4/rmi_driver.c b/drivers/input/rmi4/rmi_driver.c
index cd3e4e77ab9b..58bf3dfbf81c 100644
--- a/drivers/input/rmi4/rmi_driver.c
+++ b/drivers/input/rmi4/rmi_driver.c
@@ -136,9 +136,14 @@  static int rmi_process_interrupt_requests(struct rmi_device *rmi_dev)
 		return 0;
 
 	if (!data->attn_data.data) {
+		/*
+		 * Read the device status register as well and ignore it.
+		 * Some aftermarket ICs have issues with interrupt requests
+		 * otherwise.
+		 */
 		error = rmi_read_block(rmi_dev,
-				data->f01_container->fd.data_base_addr + 1,
-				data->irq_status, data->num_of_irq_regs);
+				data->f01_container->fd.data_base_addr,
+				data->irq_status - 1, data->num_of_irq_regs + 1);
 		if (error < 0) {
 			dev_err(dev, "Failed to read irqs, code=%d\n", error);
 			return error;
@@ -1083,16 +1088,17 @@  int rmi_probe_interrupts(struct rmi_driver_data *data)
 	data->num_of_irq_regs = (data->irq_count + 7) / 8;
 
 	size = BITS_TO_LONGS(data->irq_count) * sizeof(unsigned long);
-	data->irq_memory = devm_kcalloc(dev, size, 4, GFP_KERNEL);
+	data->irq_memory = devm_kzalloc(dev, size * 4 + 1, GFP_KERNEL);
 	if (!data->irq_memory) {
 		dev_err(dev, "Failed to allocate memory for irq masks.\n");
 		return -ENOMEM;
 	}
 
-	data->irq_status	= data->irq_memory + size * 0;
-	data->fn_irq_bits	= data->irq_memory + size * 1;
-	data->current_irq_mask	= data->irq_memory + size * 2;
-	data->new_irq_mask	= data->irq_memory + size * 3;
+	/* The first byte is reserved for the device status register */
+	data->irq_status	= data->irq_memory + size * 0 + 1;
+	data->fn_irq_bits	= data->irq_memory + size * 1 + 1;
+	data->current_irq_mask	= data->irq_memory + size * 2 + 1;
+	data->new_irq_mask	= data->irq_memory + size * 3 + 1;
 
 	return retval;
 }