Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Make all MUXs explicitly log deep Add a Kogge-Stone OR network for arbitration (makes arbitration scale log(n) with n masters, not O(n))

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Tomasz Wlostowski authored 13 years ago and Tomasz Wlostowski committed 12 years ago

wishbone/wb_slave_adapter: prevents a pipelined master from chaging data/addr lines when talking to classic slave

Tomasz Wlostowski authored 13 years ago and Tomasz Wlostowski committed 12 years ago

wishbone/xwb_bus_fanout: now determines the number of peripheral select bits from number of connected peripherals

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

wb_crossbar: Remove all the special-case code dealing with errors. It was hard to reason about, timing-wise.

Make the error device a completely standard wishbone device;
we don't need high performance segfaults. Clearer is better.

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

This bug allowed a high priority master to seize control of a lower priority master if:
1. The low priority master had control for exactly one cycle
2. Combinatorial crossbar mode was off

... this made it possible to "steal the ACK" of another device and cause the bus to stall.

This commit will be followed by a few others which also tidy up the code a
bit in a way that would have made this bug more obvious to a reviewer.

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

lm32_testsys: It is much easier to trace the workings of the crossbar if you can see the granted pins!

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Wesley W. Terpstra authored 13 years ago and Tomasz Wlostowski committed 12 years ago

When GSI replaced the lattice specific ram block with inferred memory, it became impossible to clear the register file as described by the two deleted lines. It is also unnecessary as the ram block starts zeroed.

Signed-off-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

Tomasz Wlostowski authored 13 years ago and Tomasz Wlostowski committed 12 years ago

wishbone: wb_lm32: defaulted undefined/Hi-Z states on LM32 data busses to 0 to avoid simulation hangs