Add support for the clocks used by the SpacemiT K3 QSPI
controller by retrieving and enabling the qspi_en and qspi
clocks during probe.
Signed-off-by: Jichuan Feng<jichuan.or@isrc.iscas.ac.cn>
This patch ports the Cadence xSPI controller driver from
the Linux kernel. The controller supports three operating modes:
1. ACMD (Auto Command) mode
- Includes PIO and CDMA submodes.
- CDMA mode uses linked descriptors for high-performance,
low-overhead operation.
- PIO mode is suitable for simple, single-command transactions.
2. STIG (Software Triggered Instruction Generator) mode
- Issues low-level 128-bit instructions to memory.
- Uses the Slave DMA interface for data transfers.
3. Direct mode
- Enables direct data access through the slave interface
without commands.
Currently, only the STIG work mode is enabled. Additional modes will be
supported in future updates.
At the same time, also enabling the kconfig option for xSPI driver.
This driver has been ported and functionally verified on the Intel Simics
platform. It is intended for evaluation and experimental use at this stage.
Link: https://lore.kernel.org/all/1632038734-23999-1-git-send-email-pthombar@cadence.com/
Signed-off-by: Boon Khai Ng <boon.khai.ng@altera.com>
Reviewed-by: Tien Fong Chee <tien.fong.chee@altera.com>
We found that some serial console may pull TX line to GROUND during board
boot time. Airoha uses TX line as one of it's BOOT pins. This will lead
to booting in RESERVED boot mode.
It was found that some flashes operates incorrectly in RESERVED mode.
Micron and Skyhigh flashes are definitely affected by the issue,
Winbond flashes are NOT affected.
Details:
--------
DMA reading of odd pages on affected flashes operates incorrectly. Page
reading offset (start of the page) on hardware level is replaced by 0x10.
Thus results in incorrect data reading. Usage of UBI make things even
worse. Any attempt to access UBI leads to ubi damaging. As result OS loading
becomes impossible.
Non-DMA reading is OK.
This patch detects booting in reserved mode, turn off DMA and print big
fat warning.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
The spinand driver do 3 type of dirmap requests:
* read/write whole flash page without oob
(offs = 0, len = page_size)
* read/write whole flash page including oob
(offs = 0, len = page_size + oob_size)
* read/write oob area only
(offs = page_size, len = oob_size)
The trick is:
* read/write a single "sector"
* set a custom sector size equal to offs + len. It's a bit safer to
round up "sector size" value 64.
* set the transfer length equal to custom sector size
And it works!
Thus we can find all data directly from dirmap request, so flash specific
parameters is not needed anymore. Also
* airoha_snand_nfi_config(),
* airoha_snand_nfi_setup()
functions becomes unnecessary.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Set custom sector size equal to flash page size including oob. Thus we
will always read a single sector. The maximum custom sector size is
8187, so all possible flash sector sizes are supported.
This patch is a necessary step to avoid usage of flash specific
parameters.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
This just reduce the number of modification of REG_SPI_NFI_CNFG and
REG_SPI_NFI_SECCUS_SIZE registers during dirmap operation.
This patch is a necessary step to avoid usage of flash specific
parameters.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
spi-airoha-snfi uses custom sector size in REG_SPI_NFI_SECCUS_SIZE
register, so setting of page/oob sizes in REG_SPI_NFI_PAGEFMT is not
required.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
During writing, the entire flash page (including OOB) will be updated
with the values from the temporary buffer, so we need to fill the
untouched areas of the buffer with 0xff value to prevent accidental
data overwriting.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Airoha snfi spi controller supports acceleration of DUAL/QUAD
operations, but does not supports DUAL_IO/QUAD_IO operations.
Luckily DUAL/QUAD operations do the same as DUAL_IO/QUAD_IO ones,
so we can issue corresponding DUAL/QUAD operation instead of
DUAL_IO/QUAD_IO one.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
This driver can accelerate single page operations only, thus
continuous reading mode should not be used.
Continuous reading will use sizes up to the size of one erase block.
This size is much larger than the size of single flash page. Use this
difference to identify continuous reading and return an error.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
This patch speed up cache reading/writing/updating opearions.
It was tested on en7523/an7581 and some other Airoha chips.
It will speed up
* page reading/writing without oob
* page reading/writing with oob
* oob reading/writing (significant for UBI scanning)
The only know issue appears in a very specific conditions for en7523 family
chips only.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Booting without this patch and disabled dirmap support results in
[ 2.980719] spi-nand spi0.0: Micron SPI NAND was found.
[ 2.986040] spi-nand spi0.0: 256 MiB, block size: 128 KiB, page size: 2048, OOB size: 128
[ 2.994709] 2 fixed-partitions partitions found on MTD device spi0.0
[ 3.001075] Creating 2 MTD partitions on "spi0.0":
[ 3.005862] 0x000000000000-0x000000020000 : "bl2"
[ 3.011272] 0x000000020000-0x000010000000 : "ubi"
...
[ 6.195594] ubi0: attaching mtd1
[ 13.338398] ubi0: scanning is finished
[ 13.342188] ubi0 error: ubi_read_volume_table: the layout volume was not found
[ 13.349784] ubi0 error: ubi_attach_mtd_dev: failed to attach mtd1, error -22
[ 13.356897] UBI error: cannot attach mtd1
If dirmap is disabled or not supported in the spi driver, the dirmap requests
will be executed via exec_op() handler. Thus, if the hardware supports
dual/quad spi modes, then corresponding requests will be sent to exec_op()
handler. Current driver does not support such requests, so error is arrised.
As result the flash can't be read/write.
This patch adds support of dual and quad wires spi modes to exec_op() handler.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
This operation is not needed because airoha_snand_write_data() and
airoha_snand_read_data() will properly handle data transfers above
SPI_MAX_TRANSFER_SIZE.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Coverity finds a potential integer overflow in the following code:
ncycles += ((op->data.nbytes * 8) / op->data.buswidth) / (op->data.dtr ? 2 : 1);
A quick analysis shows that the only caller of the suspicious code is the
spinand_select_op_variant() function from the drivers/mtd/nand/spi/core.c
file.
According to the code the value of op->data.nbytes is equal to
nanddev_per_page_oobsize(nand) + nanddev_page_size(nand)
Therefore it's maximum value a bit larger than 4Kb (I never seen flashes
with page size large than 4Kb). So op->data.nbytes always fits within
13 bits. As result an overflow will never happen.
Anyway it's better fix an issue to eliminate the error message.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
This driver references the get_timer macro while relying on an
indirection inclusion of <time.h>. Add the missing include directly.
Signed-off-by: Tom Rini <trini@konsulko.com>
A few spi drivers cannot build without access to some platform
specific header files. Express those requirements in Kconfig as well.
Signed-off-by: Tom Rini <trini@konsulko.com>
In _spi_get_bus_and_cs the check for stacked parallel support needing
multiple chip select support does a direct return on error. Instead it
should set the error code in ret and then use the unwind goto.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In erratum_err050568 the test for apllicability uses logical or to check
multiple chip IDs but this means the test will always evaluate to true
as at least 1 term will always be true. Logical and should have been
used so that the expression evaluates to true if all terms are true
which would mean that no chip ID of interest was in use.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In npcm_fiu_uma_operation to enter a code block nbytes must be non-zero.
So testing for nbytes inside the code block is redundant and can be
removed.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In ich_spi_exec_op_swseq the variable with_address is only assigned a
value in the case of op->addr.nbytes being non-zero.
Initialise with_address to zero. so that it is always valid.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In spi_calibration if the low range fails to calibrate then the code
attempted to return the variable err but this has not been set in this
case. Instead just return -EIO.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
In cadence_qspi_apb_exec_flash_cmd the test for a timeout uses a
post-decrement on the variable retry which will result in a value of -1
after the loop exit, or it would if the variable were signed.
To fix this make retry a signed variable and test its value for being
equal to -1.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
This reverts commit a040578d82.
Based on feedback other SOCs (for example Star64) are using driver in non
DMA mode which is causing issues that's why revert this patch.
cdns,is-dma should be removed differently.
Signed-off-by: Michal Simek <michal.simek@amd.com>
Reported-by: E Shattow <e@freeshell.de>
This series adds significant and valuable work by Mikhail Kshevetskiy to
align spi-mem with Linux 6.16. It also includes contributions to the mtd
performance patches, a work started by Miquel Raynal and improved by
Mikhail Kshevetskiy. Additionally, two patches tighten dependencies on
the Atmel driver.
The patches pass the pipeline CI:
https://source.denx.de/u-boot/custodians/u-boot-nand-flash/-/pipelines/27873
Currently the best variant picked in the first one in the list provided
in the manufacturer driver. This worked well while all operations where
performed at the same speed, but with the introduction of DTR transfers
this no longer works correctly.
Let's continue iterating over all the alternatives, even if we find a
match, keeping a reference over the theoretically fastest
operation. Only at the end we can tell which variant is the best.
This logic happening only once at boot.
The patch is based on linux commit
666c299be696 (mtd: spinand: Enhance the logic when picking a variant)
created by Miquel Raynal <miquel.raynal@bootlin.com>
The code was a bit restricted in the functionality since not all
required functionality is supported in the u-boot.
Signed-off-by: Mikhail Kshevetskiy <mikhail.kshevetskiy@iopsys.eu>
Reviewed-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Signed-off-by: Michael Trimarchi <michael@amarulasolutions.com>
Add i.MX8ULP flexspi compatible string and driver data.
The flexspi on i.MX8ULP only has 16 LUT sequences and uses 1KB RX FIFO.
Signed-off-by: Ye Li <ye.li@nxp.com>
According to i.MX8DXL A1 errata ERR050601, concurrent read accesses
from the A35 cores to the peripherals within the LSIO subsystem
(region 0_5DXX_XXXX) and address spaces in the regions
[0_0000_0000 – 0_1BFF_FFFF] and [4_0000_0000 – 4_3FFF_FFFF] can collide
and cause data corruption in the returned data, with no failure report.
Even a single A35 core accessing both these regions can trigger the issue
because an A35 core can have more than one parallel read operation in
progress.
The flexspi0 AHB memory is in LSIO region mentioned in above errata.
So we can't use AHB read, only can read data from FIFO.
Add the compatible string for 8DXL and use a flag for the IPS read.
Signed-off-by: Ye Li <ye.li@nxp.com>
Use a dedicated LUT (second last) for AHB read command, so we can
directly read from the AHB memory-mapped address and booting M core
for XIP on Flexspi NOR.
Signed-off-by: Ye Li <ye.li@nxp.com>
In spi_rx_tx there comes a test for execution of a code block that
allows execution if rxp is not NULL or stopping is true. However all the
code in this block relies on rxp being valid so allowing entry just if
stopping is true does not make sense. So remove this from the test
expression leaving just a NULL check for rxp.
This issue was found by Smatch.
Signed-off-by: Andrew Goodbody <andrew.goodbody@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
These values were calculated but not set.
They are required for the calculation of the delays, as stated in the
"QorIQ LS1043A Reference Manual, Rev. 6, 07/2020" page 2172.
The delays are calculated as (1/freq)*PCSSCK*CSSCK and
(1/freq)*PASC*ASC.
Signed-off-by: Tomas Alvarez Vanoli <tomas.alvarez-vanoli@hitachienergy.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
In atmel_qspi_transfer(), the status register is polled with:
imr = QSPI_SR_INSTRE | QSPI_SR_CSR;
return readl_poll_timeout(aq->regs + QSPI_SR, sr,
(sr & imr) == imr,
ATMEL_QSPI_TIMEOUT);
However, this is racy: QSPI_SR_INSTRE can be set before QSPI_SR_CSR,
and will then be cleared by the read. If that happens, the condition
"(sr & imr) == imr" can never be true, and the function times out.
This race condition is avoided in at91bootstrap by accumulating the
status bits across reads until both bits have been observed:
/* Poll INSTruction End and Chip Select Rise flags. */
imr = (QSPI_SR_INSTRE | QSPI_SR_CSR);
sr = 0;
do {
udelay(1);
sr |= qspi_readl(qspi, QSPI_SR) & imr;
} while ((--timeout) && (sr != imr));
Update U-Boot's atmel_qspi_transfer() to use the same pattern,
ensuring that both flags are observed even if they are not set
simultaneously.
Signed-off-by: Ramin Moussavi <lordrasmus@gmail.com>
[eugen.hristev@linaro.org: remove 'sr' and fix commit msg]
Signed-off-by: Eugen Hristev <eugen.hristev@linaro.org>
A dependency exposed by "make allyesconfig" is that the logic around
this symbol was not quite correct. It needs to depend on ARCH_MVEBU and
ARM64 and then select CLK_MVEBU.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Stefan Roese <sr@denx.de>
Add xfer function to PolarFire SoC coreqspi driver. The read and write
operations are limited to one byte at a time instead of four as CMD18
(multiple block read) reads garbage when four byte ops are enabled.
Signed-off-by: Eoin Dickson <eoin.dickson@microchip.com>
Acked-by: Leo Yu-Chi Liang <ycliang@andestech.com>
In DDR mode, even bytes are read using DMA, while the remaining odd
bytes are read using STIG mode. However, the data is not correctly
transferred into the flash read data lower register because the
supplementary byte of the STIG opcode is not being written to the
opcode extension register, resulting in incorrect data being read.
To resolve this issue, when using STIG transactions, the corresponding
supplementary byte of any STIG opcode must be defined in the Opcode
Extension Register (Lower). Issue has been observed on the Macronix
MX66UM2G45G flashes.
Signed-off-by: Prasad Kummari <prasad.kummari@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250702053953.640046-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
tshsl_ns is the clock delay for chip select deassert. This is the delay in
master reference clocks for the length that the master mode chip select
outputs are de-asserted between transactions.
The minimum delay is always SCLK period to ensure the chip select is never
re-asserted within one SCLK period.
That is why tshsl_ns delay should be at least one sclk_ns value. If it is
less than sclk_ns, set it equal to sclk_ns.
Signed-off-by: Ashok Reddy Soma <ashok.reddy.soma@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250702065717.3871435-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
Starting from 'commit <8077d296adff> ("spi: cadence-quadspi: Use STIG
mode for all ops with small payload") the utilization of STIG mode
has been implemented for read and write operations involving less
than 8 bytes of data.
However, following this commit, encountering timeout issues occurs when
writing odd bytes of data in DDR mode, as indicated below:
"jedec_spi_nor flash@0: flash operation timed out
SF: 3 bytes @ 0x0 Written: ERROR -110"
To resolve this issue, the number of bytes to write has been updated
specifically for DDR mode.
Signed-off-by: Tejas Bhumkar <tejas.arvind.bhumkar@amd.com>
Signed-off-by: Venkatesh Yadav Abbarapu <venkatesh.abbarapu@amd.com>
Link: https://lore.kernel.org/r/20250704040444.671604-1-venkatesh.abbarapu@amd.com
Signed-off-by: Michal Simek <michal.simek@amd.com>
A following patch is replacing our IO accessors with
do { ... } while(0) ones in order to make them usable with KVM.
That leads to an error eventually looking like this:
arch/arm/include/asm/io.h:62:9: error: expected expression before 'do'
62 | do { \
| ^~
arch/arm/include/asm/io.h:211:41: note: in expansion of macro '__raw_writel'
211 | #define out_arch(type,endian,a,v) __raw_write##type(cpu_to_##endian(v),a)
| ^~~~~~~~~~~
arch/arm/include/asm/io.h:223:25: note: in expansion of macro 'out_arch'
223 | #define out_be32(a,v) out_arch(l,be32,a,v)
| ^~~~~~~~
drivers/spi/fsl_dspi.c:127:17: note: in expansion of macro 'out_be32'
127 | out_be32(addr, val) : out_le32(addr, val);
| ^~~~~~~~
So adjust the current macros and code to be compatible with the upcoming
change.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
On 32-bit systems, e.g. i.mxrt-1170 fdt_addr_t may only be 32-bit.
Cast to a "long long" for garbage avoidance.
Signed-off-by: Jonathan Currier <dullfire@yahoo.com>
Chuan