GitLab

Add paranoid checks for correct values in essential registers in the SB600/SB700/... SPI driver. If something else changes the values we
wrote, we will see severe read/write corruption.
sb600spi will now abort the access and return an error if it detects
this sort of corruption.

Note: This corruption can be caused by a few different events:
- IPMI/BMC/IMC accesses flash
- Other software accesses flash
The nature of flash access (read/write/ID/...) is irrelevant. Each such
access will cause corruption for all other accesses happening at the
same time.

Thanks to Matthias Kretz for testing this patch.

Corresponding to flashrom svn r1145.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Matthias Kretz <kretz@kde.org>

Programmer specific functions are of absolutely no interest to any file
except those dealing with programmer specific actions (special SPI
commands and the generic core).

The new header structure is as follows (and yes, improvements are
possible):
flashchips.h  flash chip IDs
chipdrivers.h  chip-specific read/write/... functions
flash.h  common header for all stuff that doesn't fit elsewhere
hwaccess.h hardware access functions
programmer.h  programmer specific functions
coreboot_tables.h  header from coreboot, internal programmer only
spi.h SPI command definitions

Corresponding to flashrom svn r1112.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

Corresponding to flashrom svn r1099.
Signed-off-by: Michael Karcher <flashrom@mkarcher.dialup.fu-berlin.de>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>

Actually check if the unlock worked instead of just assuming it worked.

Corresponding to flashrom svn r1082.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Michael Karcher <flashrom@mkarcher.dialup.fu-berlin.de>

However, wrap the write functions in a compat layer to allow converting
the rest of flashrom later. Tested on Intel NM10 by David Hendricks.

Corresponding to flashrom svn r1080.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Michael Karcher <flashrom@mkarcher.dialup.fu-berlin.de>

Unistd.h was only used to get a definition of NULL in all files. Add our
own NULL #define and remove unistd.h from flash.h
stdio.h has no place in flash.h, it should be included only in files
which really need it.
Add #include statements in individual .c files where needed.

Replace a few printf with msg_* to eliminate the need for stdio.h.

Corresponding to flashrom svn r1021.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

- x86/x86_64 (little endian)
- PowerPC (big endian)
- MIPS (big+little endian)

No changes to programmer specific code. This means any drivers with MMIO
access will _not_ suddenly start working on big endian systems, but with
this patch everything is in place to fix them.

Compilation should work on all architectures listed above for all
drivers except nic3com and nicrealtek which require PCI Port IO which is
x86-only for now.

To compile without nic3com and nicrealtek, run
make distclean
make CONFIG_NIC3COM=no CONFIG_NICREALTEK=no

Thanks to Misha Manulis for testing early versions of this patch on
PowerPC (big endian) with the satasii programmer.
Thanks to Segher Boessenkool for design review and for helping out with
compiler tricks and pointing out that we need eieio on PowerPC.
Thanks to Vladimir Serbinenko for compile testing on MIPS (little
endian) and PowerPC (big endian) and for runtime testing on MIPS (little
endian).
Thanks to David Daney for compile testing on MIPS (big endian).
Thanks to Uwe Hermann for compile and runtime testing on x86_64.

DO NOT RUN flashrom ON NON-X86 AFTER APPLYING THIS PATCH!
This patch only provides the infrastructure, but does not convert any
drivers, so flashrom will compile, but it won't do the right thing on
non-x86 platforms.

Corresponding to flashrom svn r1013.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Misha Manulis <misha@manulis.com>
Acked-by: Vladimir 'phcoder/φ-coder' Serbinenko <phcoder@gmail.com>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>
Acked-by: Segher Boessenkool <segher@kernel.crashing.org>

Every SPI programmer driver had its own completely different chip write
implementation, and all of them were insufficiently commented. Create
spi_write_chunked as a copy of spi_read_chunked and convert all SPI
programmers to use it. No functional changes except: - Bus Pirate uses
12 Byte writes instead of 8 Byte writes - SB600 uses 5 Byte writes
instead of 1 Byte writes

Corresponding to flashrom svn r1005.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Michael Karcher <flashrom@mkarcher.dialup.fu-berlin.de>
Acked-by: David Hendricks <dhendrix@google.com>

Some of the spi programmer drivers required chipdrivers.h, needs fixing later:
  it87spi.c
  ichspi.c
  sb600spi.c
  wbsio_spi.c
  buspirate_spi.c
  ft2232spi.c
  bitbang_spi.c
  dediprog.c

Corresponding to flashrom svn r914.
Signed-off-by: Sean Nelson <audiohacked@gmail.com>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>

Corresponding to flashrom svn r857.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Sean Nelson <audiohacked@gmail.com>

Programmer debug messages during programmer init/shutdown are useful
because they print hardware settings and desired configuration.

They help in getting a quick overview of hardware and software state on
startup and shutdown.

Programmer debug messages during flash chip access are mostly a
distraction in logs and should only be enabled if someone is having
problems which are suspected to stem from a programmer hardware or
programmer software bug. Disable those messages by default, they can be
reenabled by #define COMM_DEBUG in the affected programmer file.

An added benefit is a tremendous size reduction in verbose
probe/read/write/erase logs because only flash chip driver messages
remain. In some cases, logs will shrink from 65 MB to 10 kB or less.

The right(tm) fix would be two different debug levels (DEBUG and SPEW)
and the ability to differentiate between programmer debug messages and
flash chip debug messages. Until the design for the message printing
infrastructure is finished, this is the best stop-gap measure we can
get.

Corresponding to flashrom svn r834.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Sean Nelson <audioahcked@gmail.com>

Corresponding to flashrom svn r739.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

Corresponding to flashrom svn r723.
Signed-off-by: Stefan Reinauer <stepan@coresystems.de>
Acked-by: Ronald G. Minnich <rminnich@gmail.com>

I decided to fill in the info for a
few chips to illustrate how this works both for uniform and non-uniform
sector sizes.

struct eraseblock{
int size; /* Eraseblock size */
int count; /* Number of contiguous blocks with that size */
};

struct eraseblock doesn't correspond with a single erase block, but with
a group of contiguous erase blocks having the same size.
Given a (top boot block) flash chip with the following weird, but
real-life structure:

top
16384
8192
8192
32768
65536
65536
65536
65536
65536
65536
65536
bottom

we get the following encoding:
{65536,7},{32768,1},{8192,2},{16384,1}

Although the number of blocks is bigger than 4, the number of block
groups is only 4. If you ever add some flash chips with more than 4
contiguous block groups, the definition will not fit into the 4-member
array anymore and gcc will recognize that and error out. No undetected
overflow possible. In that case, you simply increase array size a bit.
For modern flash chips with uniform erase block size, you only need one
array member anyway.

Of course data types will need to be changed if you ever get flash chips
with more than 2^30 erase blocks, but even with the lowest known erase
granularity of 256 bytes, these flash chips will have to have a size of
a quarter Terabyte. I'm pretty confident we won't see such big EEPROMs
in the near future (or at least not attached in a way that makes
flashrom usable). For SPI chips, we even have a guaranteed safety factor
of 4096 over the maximum SPI chip size (which is 2^24). And if such a
big flash chip has uniform erase block size, you could even split it
among the 4 array members. If you change int count to unsigned int
count, the storable size doubles. So with a split and a slight change of
data type, the maximum ROM chip size is 2 Terabytes.

Since many chips have multiple block erase functions where the
eraseblock layout depends on the block erase function, this patch
couples the block erase functions with their eraseblock layouts.
struct block_eraser {
  struct eraseblock{
    unsigned int size; /* Eraseblock size */
    unsigned int count; /* Number of contiguous blocks with that size */
  } eraseblocks[NUM_ERASEREGIONS];
  int (*block_erase) (struct flashchip *flash, unsigned int blockaddr, unsigned int blocklen);
} block_erasers[NUM_ERASEFUNCTIONS];

Corresponding to flashrom svn r719.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

Some SPI chip drivers and the generic 1-byte SPI chip write functions
didn't include the automatic erase present in other chip drivers.

Since the majority is definitely auto-erase, change the remaining
explicit-erase cases to be auto-erase as well.

Corresponding to flashrom svn r673.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Carlos Arnau Perez <cemede@gmail.com>

Serprog compilation is now controlled by a Makefile variable.
Replace munmap with physunmap where appropriate.

Corresponding to flashrom svn r671.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

If we only send an opcode and no additional data/address, the SPI
controller will read one byte too few from the chip. Basically, the
last byte of the chip response is discarded and will not end up in the
FIFO. It is unclear if the CS# line is set high too early as well. That
hardware bug is undocumented as of now, but I'm working with AMD to add
a detailed description of it to the errata.

Add loads of additional debugging to SB600/SB700 init.

Add explanatory comments for unintuitive code flow.

Thanks go to Uwe for testing quite a few iterations of the patch.

Kill the SB600 flash chip status register special case, which was a
somewhat misguided workaround for that hardware erratum.

Note for future added features in the SB600 SPI driver: It may be
possible to read up to 15 bytes of command response with overlapping
reads due to the ring buffer design of the FIFO if the command can be
repeated without ill effects. Same for skipping up to 7 bytes between
command and response.

Corresponding to flashrom svn r661.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

Some drivers support only a few combinations of read/write length and
return error otherwise. Having a distinct return code for this error
means we can handle it in upper layers.

Corresponding to flashrom svn r653.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

Tested-by: Jakob Bornecrantz <wallbraker@gmail.com>

Corresponding to flashrom svn r651.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Tested it on Epia-m700 worked okay.
Acked-by: Jakob Bornecrantz <wallbraker@gmail.com>

Some SPI opcodes need to be sent in direct succession after each other
without any chip deselect happening in between. A prominent example is
WREN (Write Enable) directly before PP (Page Program). Intel calls the
first opcode in such a row "preopcode".

Right now, we ignore the direct succession requirement completely and it
works pretty well because most onboard SPI masters have a timing or
heuristics which make the problem disappear.
The FT2232 SPI flasher is different. Since it is an external flasher,
timing is very different to what we can expect from onboard flashers and
this leads to failure at slow speeds.

This patch allows any function to submit multiple SPI commands in a
stream to any flasher. Support in the individual flashers isn't
implemented yet, so there is one generic function which passes the each
command in the stream one-by-one to the command functions of the
selected SPI flash driver.
Tested-by: Jakob Bornecrantz <wallbraker@gmail.com>

Corresponding to flashrom svn r645.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Jakob Bornecrantz <wallbraker@gmail.com>

Make a few local functions in sb600spi.c static.

Corresponding to flashrom svn r623.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Luc Verhaegen <libv@skynet.be>

That means you can tell flashrom to read exactly bytes 12345-56789
(start 12345, length 44445) and it will not fetch a single byte more.

Uwe tested this on one LPC, one SPI, and one parallel flash board.

Corresponding to flashrom svn r596.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

And even when it checks if the erase worked, the result of that check is
often ignored.

Convert all erase functions and actually check return codes
almost everywhere.
Check inside all erase_* routines if erase worked, not outside.
erase_sector_jedec and erase_block_jedec have changed prototypes to
enable erase checking.

Uwe successfully tested LPC on an CK804 box and SPI on some SB600 box.

Corresponding to flashrom svn r595.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Signed-off-by: Urja Rannikko <urjaman@gmail.com>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

This was partly due to a design problem in the abstraction layer.

There should be exactly two different functions for reading SPI chips:
- memory mapped reads
- SPI command reads.

Each of them should be contained in a separate function, optionally
taking parameters where needed.

This patch solves the problems mentioned above, shortens the code and
makes the code logic a lot more obvious.

Since open-coding the min() function leads to errors, include it in this
patch as well.

Corresponding to flashrom svn r589.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Ronald G. Minnich <rminnich@gmail.com>

Right now, the annotation only differentiates between SPI and non-SPI.
Anyone who knows more about a specific flash chip should feel free to
update it.

The existing flashbus variable was abused to denote the SPI controller
type. Use an aptly named variable for that purpose.

Once this patch is merged, the chipset/programmer init functions can set
supported flash chip types and flashrom can automatically select only
matching probe/read/erase/write functions. A side benefit of that will
be the elimination of the Winbond W29EE011 vs. AMIC A49LF040A conflict.

Corresponding to flashrom svn r556.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

Some MMIO accesses used volatile, others didn't (and risked
non-execution of side effects) and even with volatile, some accesses
looked dubious.

Since the MMIO accessor functions and the onboard flash accessor
functions are functionally identical (but have different signatures),
make the flash accessors wrappers for the MMIO accessors.

For some of the conversions, I used Coccinelle. Semantic patch follows:

@@ typedef uint8_t; expression a; volatile uint8_t *b; @@ - b[a] + *(b
+ a) @@ expression a; volatile uint8_t *b; @@ - *(b) |= (a); + *(b) =
*(b) | (a); @@ expression a; volatile uint8_t *b; @@ - *(b) = (a); +
mmio_writeb(a, b); @@ volatile uint8_t *b; @@ - *(b) + mmio_readb(b) @@
type T; T b; @@ ( mmio_readb | mmio_writeb ) (..., - (T) - (b) + b )

Corresponding to flashrom svn r524.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>

Uwe tested read, write, erase with this patch on a random board to make
sure nothing breaks.
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

Build-tested on 32bit x86.

Corresponding to flashrom svn r521.
Signed-off-by: Uwe Hermann <uwe@hermann-uwe.de>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

Change SPI architecture to handle 1-byte chunk chip writing differently
from 256-byte chunk chip writing.

Annotate SPI chip write functions with _256 or _1 suffix denoting the
number of bytes they write at maximum.

The 1-byte chunk writing is cut-n-pasted to different SPI drivers right
now. A later patch can move them to the generic spi_chip_write_1.

Corresponding to flashrom svn r485.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>

Until the ICH SPI driver can handle preopcodes as standalone opcodes,
we should handle such special opcode failure gracefully on ICH and
compatible chipsets.

This fixes chip erase on almost all ICH+VIA SPI masters.

Thanks to Ali Nadalizadeh for helping track down this bug!

Corresponding to flashrom svn r484.
Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>

Add missing copyright year.

Corresponding to flashrom svn r428 and coreboot v2 svn r4107.
Signed-off-by: Uwe Hermann <uwe@hermann-uwe.de>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>

Corresponding to flashrom svn r354 and coreboot v2 svn r3782.
Signed-off-by: Jason Wang <Qingpei.wang@amd.com>
Reviewed-by: Joe, Bao <Zheng.Bao@amd.com>
Acked-by: Stefan Reinauer <stepan@coresystems.de>

This has been tested by Uwe Hermann on an RS690/SB600 board.

Corresponding to flashrom svn r351 and coreboot v2 svn r3779.
Signed-off-by: Jason Wang <Qingpei.Wang@amd.com>
Reviewed-by: Joe Bao <zheng.bao@amd.com>
Acked-by: Uwe Hermann <uwe@hermann-uwe.de>