[PATCH] ADS5121 Fix: put dummy byte to enable fixing EOF bug to first place in order not to break some drivers\nUse ALARM interrupt to avoid waiting for data to come in
From d500e922b750a2bea554d32d8f12937f4da9c80a Mon Sep 17 00:00:00 2001
From: Matteo Fortini <redacted>
Date: Wed, 10 Dec 2008 19:33:16 +0100
Subject: [PATCH] Fix: put dummy byte to enable fixing EOF bug to first
place in order not to break some drivers
Use ALARM interrupt to avoid waiting for data to come in
Signed-off-by: Matteo Fortini <redacted>
---
drivers/spi/mpc512x_psc_spi.c | 53
++++++++++++++++++++++++-----------------
1 files changed, 31 insertions(+), 22 deletions(-)
spi_device *spi)
(spi->mode & SPI_CS_HIGH) ? 1 : 0);
}
+
/*
* Current MPC5121's have a bug in the SS logic that requires setting
* the EOF flag on the next to last byte instead of the last
@@ -155,7 +156,7 @@ static int mpc512x_psc_spi_transfer_rxtx(struct
spi_device *spi,
if (!tx_buf && !rx_buf && t->len)
return -EINVAL;
-
+
/*
* zero out Mode register 2
* From the ref man:
@@ -175,6 +176,7 @@ static int mpc512x_psc_spi_transfer_rxtx(struct
spi_device *spi,
u8 data;
size_t fifosz;
int rxcount;
+ int txcount;
/*
* The number of bytes that can be sent at a time
@@ -183,41 +185,49 @@ static int mpc512x_psc_spi_transfer_rxtx(struct
spi_device *spi,
fifosz = MPC512x_PSC_FIFO_SZ(in_be32(&fifo->txsz));
count = min(fifosz, len);
+ txcount = 0;
+ /*
+ * Insert a dummy byte before a message of len 1 to make it at
least 2 bytes long
+ * to be able to set EOF correctly
+ */
+ if (t->len == 1) {
+ out_8(&fifo->txdata_8, 0);
+ txcount++;
+ }
for (i = count; i > 0; i--) {
if (len == EOFBYTE || t->len == 1)
setbits32(&fifo->txcmd, MPC512x_PSC_FIFO_EOF);
data = tx_buf ? *tx_buf++ : 0;
out_8(&fifo->txdata_8, data);
- if (t->len == 1)
- out_8(&fifo->txdata_8, 0);
+ txcount++;
len--;
}
INIT_COMPLETION(mps->done);
- /* interrupt on tx fifo empty */
- out_be32(&fifo->txisr, MPC512x_PSC_FIFO_EMPTY);
- out_be32(&fifo->tximr, MPC512x_PSC_FIFO_EMPTY);
+ /* Enable FIFO_ALARM interrupts for rx_fifo */
+ out_be32(&fifo->rxalarm, txcount);
+ out_be32(&fifo->rxisr, MPC512x_PSC_FIFO_ALARM);
+ out_be32(&fifo->rximr, MPC512x_PSC_FIFO_ALARM);
- /* enable transmiter/receiver */
- out_8(&psc->command, MPC52xx_PSC_TX_ENABLE |
MPC52xx_PSC_RX_ENABLE);
-
- wait_for_completion(&mps->done);
+ /* Disable tx_fifo interrupts */
+ out_be32(&fifo->txisr, 0xffffffff);
+ out_be32(&fifo->tximr, 0);
- mdelay(1);
+ out_8(&psc->command, MPC52xx_PSC_TX_ENABLE |
MPC52xx_PSC_RX_ENABLE);
- /* rx fifo should have count bytes in it */
- rxcount = in_be32(&fifo->rxcnt);
- if (rxcount != count)
- mdelay(1);
+ wait_for_completion (&mps->done);
rxcount = in_be32(&fifo->rxcnt);
- if (rxcount != count && t->len != 1)
+ if (rxcount != txcount)
printk(KERN_WARNING "expected %d bytes in rx fifo "
- "but got %d\n", count, rxcount);
-
+ "but got %d\n", txcount, rxcount);
rxcount = min(rxcount, count);
{
+ /* Throw away possible initial dummy byte */
+ if (t->len == 1) {
+ (void)in_8(&fifo->rxdata_8);
+ }
for (i = rxcount; i > 0; i--) {
data = in_8(&fifo->rxdata_8);
if (rx_buf)
From: Stephen Rothwell <hidden> Date: 2008-12-10 23:06:50
Hi Matteo,
On Wed, 10 Dec 2008 19:49:58 +0100 Matteo Fortini [off-list ref] wrote:
From d500e922b750a2bea554d32d8f12937f4da9c80a Mon Sep 17 00:00:00 2001
From: Matteo Fortini <redacted>
Date: Wed, 10 Dec 2008 19:33:16 +0100
Subject: [PATCH] Fix: put dummy byte to enable fixing EOF bug to first
place in order not to break some drivers
Use ALARM interrupt to avoid waiting for data to come in
Please don't make unrelated cleanups in patches - it makes it very hard
for the reviewers to do their work.
--
Cheers,
Stephen Rothwell sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/
On Wed, 10 Dec 2008 19:49:58 +0100
Matteo Fortini [off-list ref] wrote:
From d500e922b750a2bea554d32d8f12937f4da9c80a Mon Sep 17 00:00:00 2001
From: Matteo Fortini <redacted>
Date: Wed, 10 Dec 2008 19:33:16 +0100
Subject: [PATCH] Fix: put dummy byte to enable fixing EOF bug to first
place in order not to break some drivers
Use ALARM interrupt to avoid waiting for data to come in
Signed-off-by: Matteo Fortini <redacted>
---
drivers/spi/mpc512x_psc_spi.c | 53
++++++++++++++++++++++++-----------------
1 files changed, 31 insertions(+), 22 deletions(-)
This patch is to avoid breaking some drivers, in my case the ADS7846
touchscreen one, which use 1 char messages.
If you put the dummy byte after the 1 char message, you get part of the
answer to the message in the rxbuf of the message, which is thrown away.
The solution is to put the dummy byte before the message, so that the
slave doesn't respond.
It also optimizes the interrupt handling, by using the alarm function of
the FIFO, to wait until the rx FIFO has received enough bytes, instead
of waiting until the tx FIFO is empty.
Signed-off-by: Matteo Fortini <redacted>
---
drivers/spi/mpc512x_psc_spi.c | 50
+++++++++++++++++++++++-----------------
1 files changed, 29 insertions(+), 21 deletions(-)
On Thu, 11 Dec 2008 19:39:21 +0100
Matteo Fortini [off-list ref] wrote:
This patch is to avoid breaking some drivers, in my case the ADS7846
touchscreen one, which use 1 char messages.
If you put the dummy byte after the 1 char message, you get part of the
answer to the message in the rxbuf of the message, which is thrown away.
The solution is to put the dummy byte before the message, so that the
slave doesn't respond.
It also optimizes the interrupt handling, by using the alarm function of
the FIFO, to wait until the rx FIFO has received enough bytes, instead
of waiting until the tx FIFO is empty.
Signed-off-by: Matteo Fortini <redacted>
---
drivers/spi/mpc512x_psc_spi.c | 50
+++++++++++++++++++++++-----------------
1 files changed, 29 insertions(+), 21 deletions(-)
This still seems to be whitespace corrupted. Also, that file doesn't
even exist in any upstream kernel I can find. What tree did you diff
this from?
josh
I rechecked it and I should have taken away all the only-whitespace changes.
The diffs are from the ads5121 branch from Denx.
Regards,
M
Josh Boyer ha scritto:
On Thu, 11 Dec 2008 19:39:21 +0100
Matteo Fortini [off-list ref] wrote:
quoted
This patch is to avoid breaking some drivers, in my case the ADS7846
touchscreen one, which use 1 char messages.
If you put the dummy byte after the 1 char message, you get part of the
answer to the message in the rxbuf of the message, which is thrown away.
The solution is to put the dummy byte before the message, so that the
slave doesn't respond.
It also optimizes the interrupt handling, by using the alarm function of
the FIFO, to wait until the rx FIFO has received enough bytes, instead
of waiting until the tx FIFO is empty.
Signed-off-by: Matteo Fortini <redacted>
---
drivers/spi/mpc512x_psc_spi.c | 50
+++++++++++++++++++++++-----------------
1 files changed, 29 insertions(+), 21 deletions(-)
This still seems to be whitespace corrupted. Also, that file doesn't
even exist in any upstream kernel I can find. What tree did you diff
this from?
josh
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev