The flag must be checked with "&", not "==".  When I tested
it was the only flag, so I didn't notice.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The call to kzalloc() wasn't checked.
The dev_info() message dereferenced freed memory on error.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

If we don't allocate "arr" then the cleanup path will dereference it and
oops.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The use of the 'readl' and 'writel' identifiers here causes build errors on
architectures where those are macros. This renames the fields to read32/write32
to avoid the problem.
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Alessandro Rubini <rubini@gnudd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

For traceability purposes, we include commit descriptions
as MODULE_VERSION information in drivers that are deployed
in CERN control system.

Ideally, the commit description should be added to all the
binary blobs installed, but the kernel module is critical
in this connection, as it is crucial to trace back quickly
the whereabouts of installed .ko sources.

Change suggested by Emilio Cota, whose coccicheck is newer than mine.
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>

If a client registered with a null fmc-hwdev, there was a small memory
leak.

For multi-mezzanine environments, the hwdev name (the carrier)
is not very meaningful, so use fmc->dev for messages as soon as the
device is registered.

Change suggested by David Cobas.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Multi-slot boards like the svec may have one ore more empty slots.
This patch makes the core ignore empty slots, but otherwise succeed.
Carriers are not expected to register completely-empty mezzanine sets
(but if that happens registration succeeds with no effect).

Change suggested by David Cobas.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

After correctly allocating the the manufacturer or device name,
and correctly copying the not-null-terminated string from ipmi-fru,
I put the termination '\0' at  string[len + 1] .

If you build a recent kernel under slob (which is the default), it
may fail miserably at unexpected places.  For example, if
your string is 15 bytes, I allocated 16 but wrote to the 17th.
slob has 16-byte alloc areas, and so I was corrupting the next area.

This never happened with slab (which I prefer and run) because the smalles
allocation chunk there is 32 bytes -- I would experience the same
with a 31-long manufacturer or device name.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Conflicts:

	doc/fmc-bus.in

This is only exposed by building for older kernels, as
currently err.h is pulled in by other headers I use, it seems
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

The example files and configuration matches what is current in
the fdelay driver, to be released soon after this very package.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

This allows easy access to the eeprom of each mezzanine. For example
I see one at /sys/bus/fmc/devices/FmcDelay1ns4cha-f001/eeprom (fakedev
with a fine-delay eeprom image).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The device is freed while its same work queue is running. I don't know
if this is safe, but definitely using a static work queue is.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

My previous code for device management was wrong, because I didn't
understand the allocation. This fixes the thing, by properly using the
release function.

This change however requires that all fmc devices are allocated
separately (not as an array) and are not freed by the registering
entity.  So this introduces an incompatible change in the prototype
for register_n and unregister_n -- which is noted at build time, but
also an incompatible change in how a device must be allocated/freed
by the carrier that registers a single device. This change is
not introducing a build error, but I hope the only external users by now
are the SPEC and the SVEC.

fmc-fakedev is fixed by this same commit.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The fmc devices are now children of their own device.

This is what we had with fmc-fakedev and 1 mezzanine using a custom eeprom:
tha lead to custom names):

spusa.root# ls -l /sys/bus/fmc/devices/
 FmcDelay1ns4cha-f001 -> ../../../devices/fmc/FmcDelay1ns4cha-f001/

And this is what we have now:

spusa.root# ls -l /sys/bus/fmc/devices/
 FmcDelay1ns4cha-f001 -> ../../../devices/fake-fmc-carrier/FmcDelay1ns4cha-f001/
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

This reshaping of messages make a much better output for multi-mezzanine
carriers. For example we now have this of a 4-slot fmc-fakedev:

  fake-fmc-carrier: mezzanine 0
       Manufacturer: CERN
       Product name: FmcDelay1ns4cha
  fake-fmc-carrier: mezzanine 1
       Manufacturer: fake-vendor
       Product name: fake-design-for-testing
  fake-fmc-carrier: mezzanine 2
       Manufacturer: fake-vendor
       Product name: fake-design-for-testing
  fake-fmc-carrier: mezzanine 3
       EEPROM has no FRU information
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

If there is no mezzanine_name, the core now silently use "fmc-<devid>"
as a name. This warning was confusing, as every mezzanine with invalid
eeprom won't have a predefined name.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

fmc-fakedev fakes removal and re-plugging when the eepromo is overwritten.
Thus, fmc-write-eeprom would match again and again, rewriting every two
seconds, forever.

By offsering ops->validate, the carrier prevents further matches.
To only write the first time, you can use the following, since
identifiers in fakedev start from 0xf001 ("fool").

   insmod fmc-write-eeprom busid=0xf001

The Identifier is increased avery time, so a 4-slot instance will
have identifiers 0xf001-0xf004 the first time and 0xf005-0xf008 the
next one.

This merhod allows fmc-write-eeprom to not match the busid
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

fru-dump includes local headers, where recent commit changed fru_alloc
to correctly receive size_t.  fru-dump has it's own local implementation,
which was still receiving an "int" argument. Thus build failed in 64-bit
(and I wonder why my compiler doesn't even warn in 32 bits.

But reported and fix suggested by Michel Vincent, in private email.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Building for a 64-bit kernel, I found a few warnings, related
to printk format strings.  I'm thus using %zi and %ti for size_t
and ptrdiff_t, resp.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This needless zeroing was used to fix the misbehaving fakedev, but
that is now fixed.  (again, thanks Emilio for noting).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This changes a while how stuff happens in fmc-fakedev, according to
comments by Emilio Cota who noted how I got it badly hacked and not
really stable. This now mallocs and uses dev->release as it should.

It also fills the new fields of fmc_device that are mandated by the
current version of the core (carrier_data, nr_slots, eeprom_addr).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Fixed a few subtle bugs, as yet unexposed by real use, according to
review by Emilio Cota.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Change suggested by David Cobas
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Various changes to documentation in and outside the code. Most of them
are the result of a review by Emilio Cota and Samuel Iglesias (thanks!)
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>