In preparation to enable ARCH_MMP on ARM64, a couple of fixes are needed
to build the pxa3xx_nand driver:

Legacy DMA will only used on ARM, so also make it condtional on
CONFIG_ARM.
__raw_{read,write}sl are not available on ARM64 or generically, so use
the readsl/writesl variants instead. Somewhat inconsistently,
{read,write}sl are inherently non-swapping with the generic version
using __raw_{read,write}l.
Signed-off-by: NRob Herring <robh@kernel.org>
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: linux-mtd@lists.infradead.org
[Brian: added one more __raw_readsl -> readsl]
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Now that the driver handles the FIFO draining in a threaded interrupt, we can
base our timeout on jiffies and sleeping, instead of using mdelay.
Signed-off-by: NMaxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

return type of wait_for_completion_timeout is unsigned long not int, this
patch uses the return value of wait_for_completion_timeout in the condition
directly rather than assigning it to an incorrect type variable.

The variable used for handling the return of wait_for_cmpletion_timeout
was int but should be unsigned long, where it was not in use for
anything else and the return value in case of completion (>0) is not
used it was removed and wait_for_completion_timeout() used directly in
the if condition.

To make the timeout values a bit simpler to read and also handle all of
the corner cases correctly the declarations are moved to
msecs_to_jiffies().

The timeout declaration cleanup is just for readability
Signed-off-by: NNicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

pxa3xx_flash_ids wasn't initialized to 0, which in certain cases could
end up containing corrupted values in its members. Fix this to avoid
possible issues.
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

As the devicetree binding doesn't require num_cs to exist or be strictly
positive, and neither does the platform data case, a bug appear when
num_cs is set to 0 and panics the kernel.

The issue is that in alloc_nand_resource(), chip is dereferenced without
having a value assigned when num_cs == 0.

Fix this by returning ENODEV is num_cs == 0.

The panic seen is :
Unable to handle kernel NULL pointer dereference at virtual address 000002b8
pgd = c0004000
[000002b8] *pgd=00000000
Internal error: Oops: 5 [#1] PREEMPT ARM
Modules linked in:
Hardware name: Marvell PXA3xx (Device Tree Support)
task: c3822aa0 ti: c3826000 task.ti: c3826000
PC is at alloc_nand_resource+0x180/0x4a8
LR is at alloc_nand_resource+0xa0/0x4a8
pc : [<c0275b90>]    lr : [<c0275ab0>]    psr: 68000013
sp : c3827d90  ip : 00000000  fp : 00000000
r10: c3862200  r9 : 0000005e  r8 : 00000000
r7 : c3865610  r6 : c3862210  r5 : c3924210  r4 : c3862200
r3 : 00000000  r2 : 00000000  r1 : 00000000  r0 : 00000000
Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment kernel
Control: 0000397f  Table: 80004018  DAC: 00000035
Process swapper (pid: 1, stack limit = 0xc3826198)
Stack: (0xc3827d90 to 0xc3828000)
...zip...
[<c0275b90>] (alloc_nand_resource) from [<c0275ff8>] (pxa3xx_nand_probe+0x140/0x978)
[<c0275ff8>] (pxa3xx_nand_probe) from [<c0258c40>] (platform_drv_probe+0x48/0xa4)
[<c0258c40>] (platform_drv_probe) from [<c0257650>] (driver_probe_device+0x80/0x21c)
[<c0257650>] (driver_probe_device) from [<c0257878>] (__driver_attach+0x8c/0x90)
[<c0257878>] (__driver_attach) from [<c0255ec4>] (bus_for_each_dev+0x58/0x88)
[<c0255ec4>] (bus_for_each_dev) from [<c0256ec8>] (bus_add_driver+0xd8/0x1d4)
[<c0256ec8>] (bus_add_driver) from [<c0257f14>] (driver_register+0x78/0xf4)
[<c0257f14>] (driver_register) from [<c00088a8>] (do_one_initcall+0x80/0x1e4)
[<c00088a8>] (do_one_initcall) from [<c048ed08>] (kernel_init_freeable+0xec/0x1b4)
[<c048ed08>] (kernel_init_freeable) from [<c0377d8c>] (kernel_init+0x8/0xe4)
[<c0377d8c>] (kernel_init) from [<c00095f8>] (ret_from_fork+0x14/0x3c)
Code: e503b234 e5953008 e1530001 caffffd1 (e59002b8)
---[ end trace a5770060c8441895 ]---
Signed-off-by: NRobert Jarzmik <robert.jarzmik@free.fr>
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Change the handling of the data stage in the driver : don't pump data in
the top-half interrupt, but rather schedule a thread for non dma cases.

This will enable latencies in the data pumping, especially if delays are
required. Moreover platform shall be more reactive as other interrupts
can be served while pumping data.

No throughput degradation was observed, at least on the zylonite
platform, while a slight degradation was being expected.
Signed-off-by: NRobert Jarzmik <robert.jarzmik@free.fr>
Tested-by: NMaxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The NDDB register holds the data that are needed by the read and write
commands.

However, during a read PIO access, the datasheet specifies that after each 32
bytes read in that register, when BCH is enabled, we have to make sure that the
RDDREQ bit is set in the NDSR register.

This fixes an issue that was seen on the Armada 385, and presumably other mvebu
SoCs, when a read on a newly erased page would end up in the driver reporting a
timeout from the NAND.

Cc: <stable@vger.kernel.org> # v3.14
Signed-off-by: NMaxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: NBoris Brezillon <boris.brezillon@free-electrons.com>
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The pxa3xx_nand driver currently uses __raw_writel() and __raw_readl()
to access I/O registers. However, those functions do not do any
endianness swapping, which means that they won't work when the CPU
runs in big-endian but the I/O registers are little endian, which is
the common situation for ARM systems running big endian.

Since __raw_writel() and __raw_readl() do not include any memory
barriers and the pxa3xx_nand driver can only be compiled for ARM
platforms, the closest I/o accessors functions that do endianess
swapping are writel_relaxed() and readl_relaxed().

This patch has been verified to work on Armada XP GP: without the
patch, the NAND is not detected when the kernel runs big endian while
it is properly detected when the kernel runs little endian. With the
patch applied, the NAND is properly detected in both situations
(little and big endian).
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: <stable@vger.kernel.org> # v3.13+
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit adds support for the user to specify the ECC strength
and step size through the devicetree. We keep the previous behavior,
when there is no DT parameter provided.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Let's make pxa_ecc_init() return a negative errno on error or zero
if succesful, which is standard kernel practice. Also, report the
selected ECC strength and step size, which is important information.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Make of_device_id array const, because all OF functions
handle it as const.
Signed-off-by: NJingoo Han <jg1.han@samsung.com>
Acked-by: NJason Cooper <jason@lakedaemon.net>
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The actual ECC strength used to select the ECC scheme is 'ecc_strength'.
Use it in the error message.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This macro is not used so it's safe to remove it.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit adds support for devices with 2048B page sizes and
4-bit ECC strength requirements. This is achieved by enabling the BCH
ECC engine, which provides a higher strength: 16-bit over 2048 bytes.

Additionally, add a proper ECC layout to model the controller's view
of the device (where 'U' means unused and 'B' is the bad block marker):

 ----------------------------------------------------
 | 2048B data | B | B | 30B spare | 30B ECC | U | U |
 ----------------------------------------------------
Signed-off-by: NRodolfo Giometti <giometti@linux.it>
[Brian: updated with Ezequiel's patch description]
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Now that the driver can support the Armada 370/XP SoC NAND controller,
add the devicetree compatible string, enabling its use.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Acked-by: NJason Cooper <jason@lakedaemon.net>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

In order to avoid code duplication, let's consolidate the ECC setting
for all SoC variants. Such decision is based on page size and ECC
strength requirements.

Also, provide a default value for the case where such ECC information
is not provided (non-ONFI devices).
Tested-by: NArnaud Ebalard <arno@natisbad.org>
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Currently, we have two different cmdfunc's implementations:
one for PXA3xx SoC variant and one for Armada 370/XP SoC variant.

The former is the legacy one, typically constrained to devices
with page sizes smaller or equal to the controller's FIFO buffer.
On the other side, the latter _only_ supports the so-called extended
command semantics, which allow to handle devices with larger
page sizes (4 KiB, 8 KiB, ...).

This means we currently don't support devices with smaller pages on the
A370/XP SoC. Fix it by first renaming the cmdfuncs variants, and then
make the choice based on device page size (and SoC variant), rather than
SoC variant alone.

While at it, add a check for page size, to make sure we don't allow larger
pages sizes on the PXA3xx variant.
Tested-by: NArnaud Ebalard <arno@natisbad.org>
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Currently the driver assumes all commands will eventually trigger a RnB
transition, and thus a "device is ready" IRQ.

This assumption means that on every issued command, the dev_ready completion
handler is init'ed and the need_wait flag is set.

However this is incorrect: some commands (such as NAND_CMD_STATUS) don't
make the device 'busy' and thus a RnB transition never occurs.
Given, the NAND core never calls waitfunc() after such commands, this
is not a problem.

Therefore, it's possible to only clear the need_wait flag on every command
that is started.

This fixes a current bug that can be reproduced on PXA boards by writing
blank (all 0xff'ed) to a page:

  1. The kernel issues NAND_CMD_STATUS and sets need_wait=1. The flag
     won't be cleared for this command since no RnB transition is
     involved.

  2. NAND_CMD_PAGEPROG is issued but since the data is blank, the driver
     decides not to execute the command (and no IRQ activity is
     involved).

  3. The NAND core calls waitfunc() and waits for the dev_ready
     completion, which will never end since the device _is_ already ready.
Tested-by: NArnaud Ebalard <arno@natisbad.org>
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The Armada BCH configuration in this driver uses one of the two
following ECC schemes:

 16-bit correction per 2048 bytes
 16-bit correction per 1024 bytes

These are sufficient for mapping to the 4-bit per 512-bytes and 8-bit
per 512-bytes (respectively) minimum correctability requirements of many
common NAND.

The current code only checks for the required strength (4-bit or 8-bit)
without checking the ECC step size that is associated with that strength
(and simply assumes it is 512). While that is often a safe assumption to
make, let's make it explicit, since we have that information.
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>
Acked-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>

This commit extends the ECC correctable error detection to include
ECC BCH errors. The number of BCH correctable errors can be any up to 16,
and the actual value is exposed in the NDSR register.

Therefore, we change some symbol names to refer to correctable or
uncorrectable (instead of single-bit or double-bit as it was in the
Hamming case) and while at it, cleanup the detection code slightly.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit adds write support for large pages (4 KiB, 8 KiB).
Such support is implemented by issuing a multiple command sequence,
transfering a set of 2 KiB chunks per transaction.

The splitted command sequence requires to send the SEQIN command
independently of the PAGEPROG command and therefore it's set as
an execution command.

Since PAGEPROG enables ECC, each 2 KiB chunk of data is written
together with ECC code at a controller-fixed location within
the flash page.

Currently, only devices with a 4 KiB page size has been tested.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

As preparation work to fully support large pages, this commit adds
the initial infrastructure to support splitted (aka chunked) I/O
operation. This commit adds support for read, and follow-up patches
will add write support.

When a read (aka READ0) command is issued, the driver loops issuing
the same command until all the requested data is transfered, changing
the 'extended' command field as needed.

For instance, if the driver is required to read a 4 KiB page, using a
chunk size of 2 KiB, the transaction is splitted in:
1. Monolithic read, first 2 KiB page chunk is read
2. Last naked read, second and last 2KiB page chunk is read

If ECC is enabled it is calculated on each chunk transfered and added
at a controller-fixed location after the data chunk that must be
spare area.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

In preparation to support multiple (aka chunked, aka splitted)
page I/O, this commit adds 'data_buff_pos' and 'oob_buff_pos' fields
to keep track of where the next read (or write) should be done.

This will allow multiple calls to handle_data_pio() to continue
the read (or write) operation.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit adds support page programming with a non-zero "column"
address setting. This is important to support OOB writing, through
command sequences such as:

  cmdfunc(mtd, NAND_CMD_SEQIN, mtd->writesize, ofs);
  write_buf(mtd, oob_buf, 6);
  cmdfunc(mtd, NAND_CMD_PAGEPROG, -1, -1);
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

To allow future support of multiple page reading/writing, move the data
buffer clean out of prepare_set_command().

This is done to prevent the data buffer from being cleaned on every command
preparation, when a multiple command sequence is implemented to read/write
pages larger than the FIFO size (2 KiB).
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit splits the prepare_command_pool() function into two
stages: prepare_start_command() / prepare_set_command().

This is a preparation patch without any functionality changes,
and is meant to allow support for multiple page reading/writing
operations.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

READ0 and READOOB command preparation has a falltrough to SEQIN
case, where the command address is specified.
This is certainly confusing and makes the code less readable with
no added value. Let's remove it.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Let's simplify the code by first introducing a helper function
to set the page address, as done by the READ0, READOOB and SEQIN
commands.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Command buffer #3 is not properly cleared and it keeps the last
set value. Fix this by clearing when a command is setup.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

This commit adds the BCH ECC support available in NFCv2 controller.
Depending on the detected required strength the respective ECC layout
is selected.

This commit adds an empty ECC layout, since support to access large
pages is first required. Once that support is added, a proper ECC
layout will be added as well.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Add support for flash-based bad block table using Marvell's
custom in-flash bad block table layout. The support is enabled
a 'flash_bbt' platform data or device tree parameter.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

In pxa3xx_nand_sensing() instead of simply using info->is_ready
after issuing a command, the correct way of checking is to wait
for the device to be ready through the chip's waitfunc().
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The expected behavior of the waitfunc() NAND chip call is to wait
for the device to be READY (this is a standard chip line).
However, the current implementation does almost nothing, which opens
the possibility of issuing a command to a non-ready device.

Fix this by adding a new completion to wait for the ready event to arrive.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Add a comment clarifying the use of pxa3xx_set_datasize() which is only
applicable on data read/write commands (i.e. commands with a data cycle,
such as READID, READ0, STATUS, etc.)
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

There's no need to privately store the device page size as it's
available in mtd structure field mtd->writesize.
Also, this removes the hardcoded page size value, leaving the
auto-detected value only.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Introduce a fifo_size field to represent the size of the controller's
FIFO buffer, and use it to distinguish that size from the amount
of data bytes to be read from the FIFO.

This is important to support devices with pages larger than the
controller's internal FIFO, that need to read the pages in FIFO-sized
chunks.

In particular, the current code is at least confusing, for it mixes
all the different sizes involved: FIFO size, page size and data size.

This commit starts the cleaning by removing the info->page_size field
that is not currently used. The host->page_size field should also
be removed and use always mtd->writesize instead. Follow up commits
will clean this up.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

Whenever possible, it's always better to use the generic chip->cmdfunc
instead of the internal pxa3xx_nand_cmdfunc().
In this particular case, this will allow to have multiple cmdfunc()
implementations for different SoC variants.
Reviewed-by: NHuang Shijie <shijie8@gmail.com>
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

In order to customize early settings depending on the detected SoC variant,
move the detection to be before the nand_chip struct filling.

In a follow-up patch, this change is needed to detect the variant *before*
the call to alloc_nand_resource(), which allows to set a different cmdfunc()
for each variant.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

As per the ecc.read_page() prototype, we must return the maximum number
of bitflips that were corrected on any one region covering an ecc step.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>

The current driver doesn't support sub-page writing, so report
that to the NAND core.
Signed-off-by: NEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tested-by: NDaniel Mack <zonque@gmail.com>
Signed-off-by: NBrian Norris <computersforpeace@gmail.com>