linux/drivers/net/dsa/sja1105/sja1105_spi.c

643 lines
18 KiB
C
Raw Normal View History

// SPDX-License-Identifier: BSD-3-Clause
/* Copyright (c) 2016-2018, NXP Semiconductors
* Copyright (c) 2018, Sensor-Technik Wiedemann GmbH
* Copyright (c) 2018-2019, Vladimir Oltean <olteanv@gmail.com>
*/
#include <linux/spi/spi.h>
#include <linux/packing.h>
#include "sja1105.h"
#define SJA1105_SIZE_RESET_CMD 4
#define SJA1105_SIZE_SPI_MSG_HEADER 4
#define SJA1105_SIZE_SPI_MSG_MAXLEN (64 * 4)
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
struct sja1105_chunk {
u8 *buf;
size_t len;
u64 reg_addr;
};
static void
sja1105_spi_message_pack(void *buf, const struct sja1105_spi_message *msg)
{
const int size = SJA1105_SIZE_SPI_MSG_HEADER;
memset(buf, 0, size);
sja1105_pack(buf, &msg->access, 31, 31, size);
sja1105_pack(buf, &msg->read_count, 30, 25, size);
sja1105_pack(buf, &msg->address, 24, 4, size);
}
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
#define sja1105_hdr_xfer(xfers, chunk) \
((xfers) + 2 * (chunk))
#define sja1105_chunk_xfer(xfers, chunk) \
((xfers) + 2 * (chunk) + 1)
#define sja1105_hdr_buf(hdr_bufs, chunk) \
((hdr_bufs) + (chunk) * SJA1105_SIZE_SPI_MSG_HEADER)
/* If @rw is:
* - SPI_WRITE: creates and sends an SPI write message at absolute
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
* address reg_addr, taking @len bytes from *buf
* - SPI_READ: creates and sends an SPI read message from absolute
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
* address reg_addr, writing @len bytes into *buf
*/
int sja1105_xfer_buf(const struct sja1105_private *priv,
sja1105_spi_rw_mode_t rw, u64 reg_addr,
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
u8 *buf, size_t len)
{
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
struct sja1105_chunk chunk = {
.len = min_t(size_t, len, SJA1105_SIZE_SPI_MSG_MAXLEN),
.reg_addr = reg_addr,
.buf = buf,
};
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
struct spi_device *spi = priv->spidev;
struct spi_transfer *xfers;
int num_chunks;
int rc, i = 0;
u8 *hdr_bufs;
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
num_chunks = DIV_ROUND_UP(len, SJA1105_SIZE_SPI_MSG_MAXLEN);
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
/* One transfer for each message header, one for each message
* payload (chunk).
*/
xfers = kcalloc(2 * num_chunks, sizeof(struct spi_transfer),
GFP_KERNEL);
if (!xfers)
return -ENOMEM;
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
/* Packed buffers for the num_chunks SPI message headers,
* stored as a contiguous array
*/
hdr_bufs = kcalloc(num_chunks, SJA1105_SIZE_SPI_MSG_HEADER,
GFP_KERNEL);
if (!hdr_bufs) {
kfree(xfers);
return -ENOMEM;
}
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
for (i = 0; i < num_chunks; i++) {
struct spi_transfer *chunk_xfer = sja1105_chunk_xfer(xfers, i);
struct spi_transfer *hdr_xfer = sja1105_hdr_xfer(xfers, i);
u8 *hdr_buf = sja1105_hdr_buf(hdr_bufs, i);
struct sja1105_spi_message msg;
/* Populate the transfer's header buffer */
msg.address = chunk.reg_addr;
msg.access = rw;
if (rw == SPI_READ)
msg.read_count = chunk.len / 4;
else
/* Ignored */
msg.read_count = 0;
sja1105_spi_message_pack(hdr_buf, &msg);
hdr_xfer->tx_buf = hdr_buf;
hdr_xfer->len = SJA1105_SIZE_SPI_MSG_HEADER;
/* Populate the transfer's data buffer */
if (rw == SPI_READ)
chunk_xfer->rx_buf = chunk.buf;
else
chunk_xfer->tx_buf = chunk.buf;
chunk_xfer->len = chunk.len;
/* Calculate next chunk */
chunk.buf += chunk.len;
chunk.reg_addr += chunk.len / 4;
chunk.len = min_t(size_t, (ptrdiff_t)(buf + len - chunk.buf),
SJA1105_SIZE_SPI_MSG_MAXLEN);
/* De-assert the chip select after each chunk. */
if (chunk.len)
chunk_xfer->cs_change = 1;
}
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
rc = spi_sync_transfer(spi, xfers, 2 * num_chunks);
if (rc < 0)
dev_err(&spi->dev, "SPI transfer failed: %d\n", rc);
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
kfree(hdr_bufs);
kfree(xfers);
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
return rc;
}
/* If @rw is:
* - SPI_WRITE: creates and sends an SPI write message at absolute
* address reg_addr
* - SPI_READ: creates and sends an SPI read message from absolute
* address reg_addr
*
* The u64 *value is unpacked, meaning that it's stored in the native
* CPU endianness and directly usable by software running on the core.
*/
int sja1105_xfer_u64(const struct sja1105_private *priv,
sja1105_spi_rw_mode_t rw, u64 reg_addr, u64 *value)
{
u8 packed_buf[8];
int rc;
if (rw == SPI_WRITE)
sja1105_pack(packed_buf, value, 63, 0, 8);
rc = sja1105_xfer_buf(priv, rw, reg_addr, packed_buf, 8);
if (rw == SPI_READ)
sja1105_unpack(packed_buf, value, 63, 0, 8);
return rc;
}
/* Same as above, but transfers only a 4 byte word */
int sja1105_xfer_u32(const struct sja1105_private *priv,
sja1105_spi_rw_mode_t rw, u64 reg_addr, u32 *value)
{
u8 packed_buf[4];
u64 tmp;
int rc;
if (rw == SPI_WRITE) {
/* The packing API only supports u64 as CPU word size,
* so we need to convert.
*/
tmp = *value;
sja1105_pack(packed_buf, &tmp, 31, 0, 4);
}
rc = sja1105_xfer_buf(priv, rw, reg_addr, packed_buf, 4);
if (rw == SPI_READ) {
sja1105_unpack(packed_buf, &tmp, 31, 0, 4);
*value = tmp;
}
return rc;
}
/* Back-ported structure from UM11040 Table 112.
* Reset control register (addr. 100440h)
* In the SJA1105 E/T, only warm_rst and cold_rst are
* supported (exposed in UM10944 as rst_ctrl), but the bit
* offsets of warm_rst and cold_rst are actually reversed.
*/
struct sja1105_reset_cmd {
u64 switch_rst;
u64 cfg_rst;
u64 car_rst;
u64 otp_rst;
u64 warm_rst;
u64 cold_rst;
u64 por_rst;
};
static void
sja1105et_reset_cmd_pack(void *buf, const struct sja1105_reset_cmd *reset)
{
const int size = SJA1105_SIZE_RESET_CMD;
memset(buf, 0, size);
sja1105_pack(buf, &reset->cold_rst, 3, 3, size);
sja1105_pack(buf, &reset->warm_rst, 2, 2, size);
}
static void
sja1105pqrs_reset_cmd_pack(void *buf, const struct sja1105_reset_cmd *reset)
{
const int size = SJA1105_SIZE_RESET_CMD;
memset(buf, 0, size);
sja1105_pack(buf, &reset->switch_rst, 8, 8, size);
sja1105_pack(buf, &reset->cfg_rst, 7, 7, size);
sja1105_pack(buf, &reset->car_rst, 5, 5, size);
sja1105_pack(buf, &reset->otp_rst, 4, 4, size);
sja1105_pack(buf, &reset->warm_rst, 3, 3, size);
sja1105_pack(buf, &reset->cold_rst, 2, 2, size);
sja1105_pack(buf, &reset->por_rst, 1, 1, size);
}
static int sja1105et_reset_cmd(const void *ctx, const void *data)
{
const struct sja1105_private *priv = ctx;
const struct sja1105_reset_cmd *reset = data;
const struct sja1105_regs *regs = priv->info->regs;
struct device *dev = priv->ds->dev;
u8 packed_buf[SJA1105_SIZE_RESET_CMD];
if (reset->switch_rst ||
reset->cfg_rst ||
reset->car_rst ||
reset->otp_rst ||
reset->por_rst) {
dev_err(dev, "Only warm and cold reset is supported "
"for SJA1105 E/T!\n");
return -EINVAL;
}
if (reset->warm_rst)
dev_dbg(dev, "Warm reset requested\n");
if (reset->cold_rst)
dev_dbg(dev, "Cold reset requested\n");
sja1105et_reset_cmd_pack(packed_buf, reset);
return sja1105_xfer_buf(priv, SPI_WRITE, regs->rgu, packed_buf,
SJA1105_SIZE_RESET_CMD);
}
static int sja1105pqrs_reset_cmd(const void *ctx, const void *data)
{
const struct sja1105_private *priv = ctx;
const struct sja1105_reset_cmd *reset = data;
const struct sja1105_regs *regs = priv->info->regs;
struct device *dev = priv->ds->dev;
u8 packed_buf[SJA1105_SIZE_RESET_CMD];
if (reset->switch_rst)
dev_dbg(dev, "Main reset for all functional modules requested\n");
if (reset->cfg_rst)
dev_dbg(dev, "Chip configuration reset requested\n");
if (reset->car_rst)
dev_dbg(dev, "Clock and reset control logic reset requested\n");
if (reset->otp_rst)
dev_dbg(dev, "OTP read cycle for reading product "
"config settings requested\n");
if (reset->warm_rst)
dev_dbg(dev, "Warm reset requested\n");
if (reset->cold_rst)
dev_dbg(dev, "Cold reset requested\n");
if (reset->por_rst)
dev_dbg(dev, "Power-on reset requested\n");
sja1105pqrs_reset_cmd_pack(packed_buf, reset);
return sja1105_xfer_buf(priv, SPI_WRITE, regs->rgu, packed_buf,
SJA1105_SIZE_RESET_CMD);
}
static int sja1105_cold_reset(const struct sja1105_private *priv)
{
struct sja1105_reset_cmd reset = {0};
reset.cold_rst = 1;
return priv->info->reset_cmd(priv, &reset);
}
int sja1105_inhibit_tx(const struct sja1105_private *priv,
unsigned long port_bitmap, bool tx_inhibited)
{
const struct sja1105_regs *regs = priv->info->regs;
u32 inhibit_cmd;
int rc;
rc = sja1105_xfer_u32(priv, SPI_READ, regs->port_control,
&inhibit_cmd);
if (rc < 0)
return rc;
if (tx_inhibited)
inhibit_cmd |= port_bitmap;
else
inhibit_cmd &= ~port_bitmap;
return sja1105_xfer_u32(priv, SPI_WRITE, regs->port_control,
&inhibit_cmd);
}
struct sja1105_status {
u64 configs;
u64 crcchkl;
u64 ids;
u64 crcchkg;
};
/* This is not reading the entire General Status area, which is also
* divergent between E/T and P/Q/R/S, but only the relevant bits for
* ensuring that the static config upload procedure was successful.
*/
static void sja1105_status_unpack(void *buf, struct sja1105_status *status)
{
/* So that addition translates to 4 bytes */
u32 *p = buf;
/* device_id is missing from the buffer, but we don't
* want to diverge from the manual definition of the
* register addresses, so we'll back off one step with
* the register pointer, and never access p[0].
*/
p--;
sja1105_unpack(p + 0x1, &status->configs, 31, 31, 4);
sja1105_unpack(p + 0x1, &status->crcchkl, 30, 30, 4);
sja1105_unpack(p + 0x1, &status->ids, 29, 29, 4);
sja1105_unpack(p + 0x1, &status->crcchkg, 28, 28, 4);
}
static int sja1105_status_get(struct sja1105_private *priv,
struct sja1105_status *status)
{
const struct sja1105_regs *regs = priv->info->regs;
u8 packed_buf[4];
int rc;
rc = sja1105_xfer_buf(priv, SPI_READ, regs->status, packed_buf, 4);
if (rc < 0)
return rc;
sja1105_status_unpack(packed_buf, status);
return 0;
}
/* Not const because unpacking priv->static_config into buffers and preparing
* for upload requires the recalculation of table CRCs and updating the
* structures with these.
*/
static int
static_config_buf_prepare_for_upload(struct sja1105_private *priv,
void *config_buf, int buf_len)
{
struct sja1105_static_config *config = &priv->static_config;
struct sja1105_table_header final_header;
sja1105_config_valid_t valid;
char *final_header_ptr;
int crc_len;
valid = sja1105_static_config_check_valid(config);
if (valid != SJA1105_CONFIG_OK) {
dev_err(&priv->spidev->dev,
sja1105_static_config_error_msg[valid]);
return -EINVAL;
}
/* Write Device ID and config tables to config_buf */
sja1105_static_config_pack(config_buf, config);
/* Recalculate CRC of the last header (right now 0xDEADBEEF).
* Don't include the CRC field itself.
*/
crc_len = buf_len - 4;
/* Read the whole table header */
final_header_ptr = config_buf + buf_len - SJA1105_SIZE_TABLE_HEADER;
sja1105_table_header_packing(final_header_ptr, &final_header, UNPACK);
/* Modify */
final_header.crc = sja1105_crc32(config_buf, crc_len);
/* Rewrite */
sja1105_table_header_packing(final_header_ptr, &final_header, PACK);
return 0;
}
#define RETRIES 10
int sja1105_static_config_upload(struct sja1105_private *priv)
{
unsigned long port_bitmap = GENMASK_ULL(SJA1105_NUM_PORTS - 1, 0);
struct sja1105_static_config *config = &priv->static_config;
const struct sja1105_regs *regs = priv->info->regs;
struct device *dev = &priv->spidev->dev;
struct sja1105_status status;
int rc, retries = RETRIES;
u8 *config_buf;
int buf_len;
buf_len = sja1105_static_config_get_length(config);
config_buf = kcalloc(buf_len, sizeof(char), GFP_KERNEL);
if (!config_buf)
return -ENOMEM;
rc = static_config_buf_prepare_for_upload(priv, config_buf, buf_len);
if (rc < 0) {
dev_err(dev, "Invalid config, cannot upload\n");
rc = -EINVAL;
goto out;
}
/* Prevent PHY jabbering during switch reset by inhibiting
* Tx on all ports and waiting for current packet to drain.
* Otherwise, the PHY will see an unterminated Ethernet packet.
*/
rc = sja1105_inhibit_tx(priv, port_bitmap, true);
if (rc < 0) {
dev_err(dev, "Failed to inhibit Tx on ports\n");
rc = -ENXIO;
goto out;
}
/* Wait for an eventual egress packet to finish transmission
* (reach IFG). It is guaranteed that a second one will not
* follow, and that switch cold reset is thus safe
*/
usleep_range(500, 1000);
do {
/* Put the SJA1105 in programming mode */
rc = sja1105_cold_reset(priv);
if (rc < 0) {
dev_err(dev, "Failed to reset switch, retrying...\n");
continue;
}
/* Wait for the switch to come out of reset */
usleep_range(1000, 5000);
/* Upload the static config to the device */
net: dsa: sja1105: Switch to scatter/gather API for SPI This reworks the SPI transfer implementation to make use of more of the SPI core features. The main benefit is to avoid the memcpy in sja1105_xfer_buf(). The memcpy was only needed because the function was transferring a single buffer at a time. So it needed to copy the caller-provided buffer at buf + 4, to store the SPI message header in the "headroom" area. But the SPI core supports scatter-gather messages, comprised of multiple transfers. We can actually use those to break apart every SPI message into 2 transfers: one for the header and one for the actual payload. To keep the behavior the same regarding the chip select signal, it is necessary to tell the SPI core to de-assert the chip select after each chunk. This was not needed before, because each spi_message contained only 1 single transfer. The meaning of the per-transfer cs_change=1 is: - If the transfer is the last one of the message, keep CS asserted - Otherwise, deassert CS We need to deassert CS in the "otherwise" case, which was implicit before. Avoiding the memcpy creates yet another opportunity. The device can't process more than 256 bytes of SPI payload at a time, so the sja1105_xfer_long_buf() function used to exist, to split the larger caller buffer into chunks. But these chunks couldn't be used as scatter/gather buffers for spi_message until now, because of that memcpy (we would have needed more memory for each chunk). So we can now remove the sja1105_xfer_long_buf() function and have a single implementation for long and short buffers. Another benefit is lower usage of stack memory. Previously we had to store 2 SPI buffers for each chunk. Due to the elimination of the memcpy, we can now send pointers to the actual chunks from the caller-supplied buffer to the SPI core. Since the patch merges two functions into a rewritten implementation, the function prototype was also changed, mainly for cosmetic consistency with the structures used within it. Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-10-12 01:31:15 +03:00
rc = sja1105_xfer_buf(priv, SPI_WRITE, regs->config,
config_buf, buf_len);
if (rc < 0) {
dev_err(dev, "Failed to upload config, retrying...\n");
continue;
}
/* Check that SJA1105 responded well to the config upload */
rc = sja1105_status_get(priv, &status);
if (rc < 0)
continue;
if (status.ids == 1) {
dev_err(dev, "Mismatch between hardware and static config "
"device id. Wrote 0x%llx, wants 0x%llx\n",
config->device_id, priv->info->device_id);
continue;
}
if (status.crcchkl == 1) {
dev_err(dev, "Switch reported invalid local CRC on "
"the uploaded config, retrying...\n");
continue;
}
if (status.crcchkg == 1) {
dev_err(dev, "Switch reported invalid global CRC on "
"the uploaded config, retrying...\n");
continue;
}
if (status.configs == 0) {
dev_err(dev, "Switch reported that configuration is "
"invalid, retrying...\n");
continue;
}
/* Success! */
break;
} while (--retries);
if (!retries) {
rc = -EIO;
dev_err(dev, "Failed to upload config to device, giving up\n");
goto out;
} else if (retries != RETRIES) {
dev_info(dev, "Succeeded after %d tried\n", RETRIES - retries);
}
rc = sja1105_ptp_reset(priv->ds);
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
if (rc < 0)
dev_err(dev, "Failed to reset PTP clock: %d\n", rc);
dev_info(dev, "Reset switch and programmed static config\n");
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
out:
kfree(config_buf);
return rc;
}
static struct sja1105_regs sja1105et_regs = {
.device_id = 0x0,
.prod_id = 0x100BC3,
.status = 0x1,
.port_control = 0x11,
.config = 0x020000,
.rgu = 0x100440,
/* UM10944.pdf, Table 86, ACU Register overview */
.pad_mii_tx = {0x100800, 0x100802, 0x100804, 0x100806, 0x100808},
.rmii_pll1 = 0x10000A,
.cgu_idiv = {0x10000B, 0x10000C, 0x10000D, 0x10000E, 0x10000F},
.mac = {0x200, 0x202, 0x204, 0x206, 0x208},
.mac_hl1 = {0x400, 0x410, 0x420, 0x430, 0x440},
.mac_hl2 = {0x600, 0x610, 0x620, 0x630, 0x640},
/* UM10944.pdf, Table 78, CGU Register overview */
.mii_tx_clk = {0x100013, 0x10001A, 0x100021, 0x100028, 0x10002F},
.mii_rx_clk = {0x100014, 0x10001B, 0x100022, 0x100029, 0x100030},
.mii_ext_tx_clk = {0x100018, 0x10001F, 0x100026, 0x10002D, 0x100034},
.mii_ext_rx_clk = {0x100019, 0x100020, 0x100027, 0x10002E, 0x100035},
.rgmii_tx_clk = {0x100016, 0x10001D, 0x100024, 0x10002B, 0x100032},
.rmii_ref_clk = {0x100015, 0x10001C, 0x100023, 0x10002A, 0x100031},
.rmii_ext_tx_clk = {0x100018, 0x10001F, 0x100026, 0x10002D, 0x100034},
.ptpegr_ts = {0xC0, 0xC2, 0xC4, 0xC6, 0xC8},
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_control = 0x17,
.ptpclk = 0x18, /* Spans 0x18 to 0x19 */
.ptpclkrate = 0x1A,
.ptptsclk = 0x1B, /* Spans 0x1B to 0x1C */
};
static struct sja1105_regs sja1105pqrs_regs = {
.device_id = 0x0,
.prod_id = 0x100BC3,
.status = 0x1,
.port_control = 0x12,
.config = 0x020000,
.rgu = 0x100440,
/* UM10944.pdf, Table 86, ACU Register overview */
.pad_mii_tx = {0x100800, 0x100802, 0x100804, 0x100806, 0x100808},
.pad_mii_id = {0x100810, 0x100811, 0x100812, 0x100813, 0x100814},
.rmii_pll1 = 0x10000A,
.cgu_idiv = {0x10000B, 0x10000C, 0x10000D, 0x10000E, 0x10000F},
.mac = {0x200, 0x202, 0x204, 0x206, 0x208},
.mac_hl1 = {0x400, 0x410, 0x420, 0x430, 0x440},
.mac_hl2 = {0x600, 0x610, 0x620, 0x630, 0x640},
/* UM11040.pdf, Table 114 */
.mii_tx_clk = {0x100013, 0x100019, 0x10001F, 0x100025, 0x10002B},
.mii_rx_clk = {0x100014, 0x10001A, 0x100020, 0x100026, 0x10002C},
.mii_ext_tx_clk = {0x100017, 0x10001D, 0x100023, 0x100029, 0x10002F},
.mii_ext_rx_clk = {0x100018, 0x10001E, 0x100024, 0x10002A, 0x100030},
.rgmii_tx_clk = {0x100016, 0x10001C, 0x100022, 0x100028, 0x10002E},
.rmii_ref_clk = {0x100015, 0x10001B, 0x100021, 0x100027, 0x10002D},
.rmii_ext_tx_clk = {0x100017, 0x10001D, 0x100023, 0x100029, 0x10002F},
.qlevel = {0x604, 0x614, 0x624, 0x634, 0x644},
.ptpegr_ts = {0xC0, 0xC4, 0xC8, 0xCC, 0xD0},
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_control = 0x18,
.ptpclk = 0x19,
.ptpclkrate = 0x1B,
.ptptsclk = 0x1C,
};
struct sja1105_info sja1105e_info = {
.device_id = SJA1105E_DEVICE_ID,
.part_no = SJA1105ET_PART_NO,
.static_ops = sja1105e_table_ops,
.dyn_ops = sja1105et_dyn_ops,
.ptp_ts_bits = 24,
.ptpegr_ts_bytes = 4,
.reset_cmd = sja1105et_reset_cmd,
.fdb_add_cmd = sja1105et_fdb_add,
.fdb_del_cmd = sja1105et_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105et_ptp_cmd,
.regs = &sja1105et_regs,
.name = "SJA1105E",
};
struct sja1105_info sja1105t_info = {
.device_id = SJA1105T_DEVICE_ID,
.part_no = SJA1105ET_PART_NO,
.static_ops = sja1105t_table_ops,
.dyn_ops = sja1105et_dyn_ops,
.ptp_ts_bits = 24,
.ptpegr_ts_bytes = 4,
.reset_cmd = sja1105et_reset_cmd,
.fdb_add_cmd = sja1105et_fdb_add,
.fdb_del_cmd = sja1105et_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105et_ptp_cmd,
.regs = &sja1105et_regs,
.name = "SJA1105T",
};
struct sja1105_info sja1105p_info = {
.device_id = SJA1105PR_DEVICE_ID,
.part_no = SJA1105P_PART_NO,
.static_ops = sja1105p_table_ops,
.dyn_ops = sja1105pqrs_dyn_ops,
.ptp_ts_bits = 32,
.ptpegr_ts_bytes = 8,
.setup_rgmii_delay = sja1105pqrs_setup_rgmii_delay,
.reset_cmd = sja1105pqrs_reset_cmd,
.fdb_add_cmd = sja1105pqrs_fdb_add,
.fdb_del_cmd = sja1105pqrs_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105pqrs_ptp_cmd,
.regs = &sja1105pqrs_regs,
.name = "SJA1105P",
};
struct sja1105_info sja1105q_info = {
.device_id = SJA1105QS_DEVICE_ID,
.part_no = SJA1105Q_PART_NO,
.static_ops = sja1105q_table_ops,
.dyn_ops = sja1105pqrs_dyn_ops,
.ptp_ts_bits = 32,
.ptpegr_ts_bytes = 8,
.setup_rgmii_delay = sja1105pqrs_setup_rgmii_delay,
.reset_cmd = sja1105pqrs_reset_cmd,
.fdb_add_cmd = sja1105pqrs_fdb_add,
.fdb_del_cmd = sja1105pqrs_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105pqrs_ptp_cmd,
.regs = &sja1105pqrs_regs,
.name = "SJA1105Q",
};
struct sja1105_info sja1105r_info = {
.device_id = SJA1105PR_DEVICE_ID,
.part_no = SJA1105R_PART_NO,
.static_ops = sja1105r_table_ops,
.dyn_ops = sja1105pqrs_dyn_ops,
.ptp_ts_bits = 32,
.ptpegr_ts_bytes = 8,
.setup_rgmii_delay = sja1105pqrs_setup_rgmii_delay,
.reset_cmd = sja1105pqrs_reset_cmd,
.fdb_add_cmd = sja1105pqrs_fdb_add,
.fdb_del_cmd = sja1105pqrs_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105pqrs_ptp_cmd,
.regs = &sja1105pqrs_regs,
.name = "SJA1105R",
};
struct sja1105_info sja1105s_info = {
.device_id = SJA1105QS_DEVICE_ID,
.part_no = SJA1105S_PART_NO,
.static_ops = sja1105s_table_ops,
.dyn_ops = sja1105pqrs_dyn_ops,
.regs = &sja1105pqrs_regs,
.ptp_ts_bits = 32,
.ptpegr_ts_bytes = 8,
.setup_rgmii_delay = sja1105pqrs_setup_rgmii_delay,
.reset_cmd = sja1105pqrs_reset_cmd,
.fdb_add_cmd = sja1105pqrs_fdb_add,
.fdb_del_cmd = sja1105pqrs_fdb_del,
net: dsa: sja1105: Add support for the PTP clock The design of this PHC driver is influenced by the switch's behavior w.r.t. timestamping. It exposes two PTP counters, one free-running (PTPTSCLK) and the other offset- and frequency-corrected in hardware through PTPCLKVAL, PTPCLKADD and PTPCLKRATE. The MACs can sample either of these for frame timestamps. However, the user manual warns that taking timestamps based on the corrected clock is less than useful, as the switch can deliver corrupted timestamps in a variety of circumstances. Therefore, this PHC uses the free-running PTPTSCLK together with a timecounter/cyclecounter structure that translates it into a software time domain. Thus, the settime/adjtime and adjfine callbacks are hardware no-ops. The timestamps (introduced in a further patch) will also be translated to the correct time domain before being handed over to the userspace PTP stack. The introduction of a second set of PHC operations that operate on the hardware PTPCLKVAL/PTPCLKADD/PTPCLKRATE in the future is somewhat unavoidable, as the TTEthernet core uses the corrected PTP time domain. However, the free-running counter + timecounter structure combination will suffice for now, as the resulting timestamps yield a sub-50 ns synchronization offset in steady state using linuxptp. For this patch, in absence of frame timestamping, the operations of the switch PHC were tested by syncing it to the system time as a local slave clock with: phc2sys -s CLOCK_REALTIME -c swp2 -O 0 -m -S 0.01 Signed-off-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2019-06-08 15:04:34 +03:00
.ptp_cmd = sja1105pqrs_ptp_cmd,
.name = "SJA1105S",
};