Re: [PATCHv3 4/5] mtd: mxc_nand fixups

2 messages, 2 authors, 2010-06-23 · open the first message on its own page

Re: [PATCHv3 4/5] mtd: mxc_nand fixups

From: John Ogness <john.ogness@linutronix.de>
Date: 2010-06-23 08:48:58

On 2010-06-23, Ivo Clarysse [off-list ref] wrote:
Unfortunately, this breaks i.MX21 support.

With this patch, the driver gets stuck at the first send_addr()
invocation:

[...]

Setting NFC_CONFIG1:NFC_INT_MSK in the interrupt handler causes
NFC_CONFIG2:NFC_INT to always read out 0 (on i.MX21), causing

    wait_event(host->irq_waitq,
               readw(host->regs + NFC_CONFIG2) & NFC_INT);

to wait forever.
OK. Now I understand the problem. Here is a new patch that introduces a
flag that is set by the interrupt handler. This way we do not rely on
the i.MX21 being able to read NFC_INT when the interrupt is masked.

I have also added a lot of comments so that it is (hopefully) clear why
we are enabling and disabling the irq line rather than only masking and
unmasking the interrupt.

The patch is against linux-next 20100618.

Signed-off-by: John Ogness <john.ogness@linutronix.de>
---
 drivers/mtd/nand/mxc_nand.c |   49 +++++++++++++++++++++++++++++-----
 1 file changed, 42 insertions(+), 7 deletions(-)

Index: linux-next-20100618/drivers/mtd/nand/mxc_nand.c
===================================================================
--- linux-next-20100618.orig/drivers/mtd/nand/mxc_nand.c
+++ linux-next-20100618/drivers/mtd/nand/mxc_nand.c
@@ -30,6 +30,7 @@
 #include <linux/clk.h>
 #include <linux/err.h>
 #include <linux/io.h>
+#include <linux/irq.h>
 
 #include <asm/mach/flash.h>
 #include <mach/mxc_nand.h>
@@ -117,6 +118,7 @@ struct mxc_nand_host {
 	int			clk_act;
 	int			irq;
 
+	int			nfc_int;
 	wait_queue_head_t	irq_waitq;
 
 	uint8_t			*data_buf;
@@ -173,8 +175,21 @@ static const char *part_probes[] = { "Re
 static irqreturn_t mxc_nfc_irq(int irq, void *dev_id)
 {
 	struct mxc_nand_host *host = dev_id;
+	uint16_t tmp;
+
+	/* If NFC_INT is not set, we have a spurious interrupt! */
+	if ((readw(host->regs + NFC_CONFIG2) & NFC_INT) == 0)
+		return IRQ_NONE;
+
+	/* We use the host->nfc_int flag because the i.MX21 cannot
+	 * read the CONFIG2:NFC_INT bit if interrupts are masked. */
+	host->nfc_int = 1;
 
-	disable_irq_nosync(irq);
+	/* Mask the interrupts so that we avoid an interrupt
+	 * flood until the irq line is disabled. */
+	tmp = readw(host->regs + NFC_CONFIG1);
+	tmp |= NFC_INT_MSK;
+	writew(tmp, host->regs + NFC_CONFIG1);
 
 	wake_up(&host->irq_waitq);
 
@@ -192,15 +207,33 @@ static void wait_op_done(struct mxc_nand
 	if (useirq) {
 		if ((readw(host->regs + NFC_CONFIG2) & NFC_INT) == 0) {
 
+			/* Clear the interrupt flag. It will be set
+			 * by the interrupt handler. */
+			host->nfc_int = 0;
+
+			/* We enable the irq line and wait for an interrupt.
+			 * We must use irq line enable/disable rather than
+			 * simply masking of the interrupt because the i.MX21
+			 * cannot read the CONFIG2:NFC_INT bit if the
+			 * interrupt is masked. This is also the reason we
+			 * use the host->nfc_int flag. */
 			enable_irq(host->irq);
 
-			wait_event(host->irq_waitq,
-				readw(host->regs + NFC_CONFIG2) & NFC_INT);
+			wait_event(host->irq_waitq, host->nfc_int);
+
+			disable_irq(host->irq);
 
-			tmp = readw(host->regs + NFC_CONFIG2);
-			tmp  &= ~NFC_INT;
-			writew(tmp, host->regs + NFC_CONFIG2);
+			/* The irq line has been disabled. We can now
+			 * unmask the interrupts. */
+			tmp = readw(host->regs + NFC_CONFIG1);
+			tmp &= ~NFC_INT_MSK;
+			writew(tmp, host->regs + NFC_CONFIG1);
 		}
+
+		/* clear interrupt flag */
+		tmp = readw(host->regs + NFC_CONFIG2);
+		tmp &= ~NFC_INT;
+		writew(tmp, host->regs + NFC_CONFIG2);
 	} else {
 		while (max_retries-- > 0) {
 			if (readw(host->regs + NFC_CONFIG2) & NFC_INT) {
@@ -846,7 +879,9 @@ static int __init mxcnd_probe(struct pla
 
 	host->irq = platform_get_irq(pdev, 0);
 
-	err = request_irq(host->irq, mxc_nfc_irq, IRQF_DISABLED, DRIVER_NAME, host);
+	/* request irq as disabled */
+	set_irq_flags(host->irq, IRQF_VALID | IRQF_NOAUTOEN);
+	err = request_irq(host->irq, mxc_nfc_irq, 0, DRIVER_NAME, host);
 	if (err)
 		goto eirq;
 

Re: [PATCHv3 4/5] mtd: mxc_nand fixups

From: Ivo Clarysse <hidden>
Date: 2010-06-23 09:23:59

On Wed, Jun 23, 2010 at 10:48 AM, John Ogness [off-list ref] wrote:
[...]
OK. Now I understand the problem. Here is a new patch that introduces a
flag that is set by the interrupt handler. This way we do not rely on
the i.MX21 being able to read NFC_INT when the interrupt is masked.
Yes, this works on i.MX21 (tested on an MX21ADS board).

[...]
quoted hunk
@@ -117,6 +118,7 @@ struct mxc_nand_host {
? ? ? ?int ? ? ? ? ? ? ? ? ? ? clk_act;
? ? ? ?int ? ? ? ? ? ? ? ? ? ? irq;

+ ? ? ? int ? ? ? ? ? ? ? ? ? ? nfc_int;
But is it OK to use a regular (non-volatile) variable to communicate
between interrupt context and the non-interrupt context ?

My original patch for i.MX21 used completions instead:

http://lists.infradead.org/pipermail/linux-arm-kernel/2010-April/012694.html
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help