We should use at least 27e6b00, to avoid the readl/writel
naming conflicts addressed by that commit

kernel: fix edge emulation timeout & check for subsequent VIC IRQs before returning from the handler.

The latter keeps handling VIC irqs in a loop until none of the IRQ lines is active anymore. This prevents
situations when two or more IRQs come almost simultaneously (e.g. reading sychronized sample buffers) and the one with
lower priority gets starved.

The VIC is the standard BE-CO-HT module for multiplexing FPGA-internal IRQ sources. Support for
it in the carrier driver is provided for convenience of FMC driver developers. More info in the
svec-sw manual.

This way, no external dependency on /acc/... directories exists

Signed-off-by: Juan David Gonzalez Cobas <dcobas@cern.ch>

When registering several mezzanines, the second device_register
causes a reprogram that should not do anything (the bitstream
should not change). The release of fmc->sdb information
must happen before, so that fmc_scan_sdb_tree refills
the new SDB information properly.

This is a kludge over a kludge: actually, fmc_scan_sdb_tree
should be called once in the carrier, so that we do not fall into
this.

This commit reverts 95a811a4, as the correct use of FMC_NO_MEZZANINE
prevents from faking an EEPROM and registering the (physically inexistent)
device

We need to track the commit this binary comes from, as a general
policy

golden has the same bitbanged i2c interface as the fine delay

It is useless and should be no more