Theodor-Adrian Stana authored 11 years ago and Grzegorz Daniluk committed 10 years ago

The flash-write tool is operational. The call to read status register
from flash-host is not working yet, and that is to be solved. In the
meantime, there are sleep functions inserted where we're supposed
to poll the status register.

Signed-off-by: Theodor Stana <t.stana@cern.ch>

Theodor-Adrian Stana authored 11 years ago and Grzegorz Daniluk committed 10 years ago

wrc_main still in test-mode, currently doing a read of the first two
sectors (in the purpose of testing flash-write in tools -- see next
commit)

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

eeprom_present() returned a value nobody used. Rename to eeprom_init()
instead. Also, save the two parameters so that they are not needed
in other functions (see later commits).

Also, turn int8_t to int as function arguments (int is better for the
CPU).

This adds 20 bytes to the legacy case, and removes 4 to the sdb case.
The legacy size is reduced by later commits.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This allows stuff like

  ./MAKEALL spec_legacy_defconfig spec_sdb_defconfig

to choose your relevant configuration files alone.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is the simpler thing: have vendor ID in the data structure.

We have 100 bytes more of data, but we save 50 bytes of code.  Total:
50 more but with simpler code (not bound to cern and gsi, thus
reusable).

Please maintainer choose and squash.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

[Note: this patch can be preserved or squashed into commit
  replace sdb.c with new library-based code
at the maintainer's best choice]

My initial code assumed the vendor is always VID_CERN, but this is
not true: some cells we use are VID_GSI.  This takes care of the thing.

To save a little space, instead of adding a 64-bit field to the data
structure describing a device, I add an integer field stating whether
the device is CERN or GSI.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This adds almost 800 bytes to all configurations, but removes
200 bytes from the gsi_defconfig.  This is because sdb-eeprom
is already managed with this library, and thus the commit is removing
a duplication.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

It is not mandatory to actually register the tree, but it's
a good move for consistency.

Later we'll need to scan several devices, for example to look for a
mac address in carrier flash or eeprom. To do that we'll need to have
devices properly registered.

This costs 150 bytes of binary size in gsi_defconfig.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This makes the binary size 500 bytes bigger for gsi_defconfig
(the configuration already using the library) and strangely a little
smaller for etherbone_defconfig.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We agreed, long ago, to have 128k on the spec, like GSI is
already doing. This changes the default in Kconfig and removes
specific settings in configs/*

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

If the host wants a consistent view (*if* it wants it), it
can wait for the sequence number to be even, read the lot and
then check it did not change.

If it is odd, it means the structure is being modified. If it changed
after reading it means a new modification started.

Not my invention: it's old stuff, though smart. See sequential locks
in the kernel for example.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Conflicts:

	arch/lm32/ram-wrs.ld

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

"char valid" in a structure of integer makes alignment problems
(because we are accessing it from a different CPU we'd better not have
holes).

Also, there is no inherent consistency needed here, and this valid bit
is racy anyways. We could solve it with a sequence number, if any, but
let's avoid it altogether by now.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Alessandro Rubini authored 10 years ago and Grzegorz Daniluk committed 10 years ago

The global pointer is used by some special assembly instructions
that we are not using (i.e., the compiler is not generating them
for us, because we don't use PIC or other "strange" stuff.

So, we'd better remove an assignement in the linker script that makes
little sense (it must be a typo of some kind, in the dark ages of this
code base).

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Conflicts:
	softpll/softpll_ng.c

We finally agreed to not use ptp-noposix any more. This allows simplifying
the main loop and library functions, in the future.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Pietro Fezzardi authored 10 years ago and Alessandro Rubini committed 10 years ago

The previous version was compiled into a 64bit division
by gcc. in this way it uses no division, but a bitshift.

This reduces the compiled file by 4 bytes.

Pietro Fezzardi authored 10 years ago and Alessandro Rubini committed 10 years ago

Some declaration was int8_t, some int32_t and some didn't match
the function definition. Now everything returns just int.
Anyway the functions return only 0 or -1, so there are no problems
returning int. Moreover in this way the resulting assembler is
smaller for every target architecture (even if here we always
build for lm32)

This reduces the binary by 30 bytes.

Pietro Fezzardi authored 10 years ago and Alessandro Rubini committed 10 years ago

This adds 2.3kB to the GSI configuration, beause it enable ptpdump.
Earlier it was always forces off, due to wrong comparison of ram size.