[rv_dm] Return a TL-UL error on writes to ROM region

[msfschaffner]

This fixes multiple TL-UL errors that we are currently seeing at the chip-level: #14200, #14653 and #14921. I am currently running the seeds mentioned on these issues to double check whether they can be closed.

Signed-off-by: Michael Schaffner [email protected]

[msfschaffner]

Looks like I got ahead of myself with this one.

This cannot be fixed in this way, because the RV_DM debug memory has both RO and RW regions. These regions are implemented inside the vendored-in RV_DM code, and hence this is not accurately captured in rv_dm.hjson.

@sriyerg has added support for the custom CSR node manually in the block-level DV environment by adding rv_dm_debug_mem.hjson. The debug memory RAL is included in the top-level simulation as well, but probably the exceptions for read-only and unmapped memory regions are missing (the RV_DM does currently not error back if these regions are accessed). These exceptions are defined in the env config in the block-level testbench, and probably need to be added to the chip-level environment as well:

opentitan/hw/ip/rv_dm/dv/env/rv_dm_env_cfg.sv

Lines 74 to 99 in 97d123b

	protected virtual function void post_build_ral_settings(dv_base_reg_block ral);
	// The backdoor HDL paths are set incorrectly on the debug mem RAL structures by the reggen
	// tool. We just remove all HDL paths and skip backdoor writes entirely.
	// TODO: Enable backdoor writes later.
	if (ral.get_name() == rom_ral_name) begin
	rv_dm_debug_mem_reg_block debug_mem_ral;
	uvm_reg regs[$];
	ral.get_registers(regs);
	foreach (regs[i]) begin
	regs[i].clear_hdl_path("ALL");
	end
	// ROM within the debug mem is RO - it ignores writes instead of throwing an error response.
	`downcast(debug_mem_ral, ral)
	debug_mem_ral.rom.set_write_to_ro_mem_ok(1);
	debug_mem_ral.rom.set_mem_partial_write_support(1);
	// TODO(#10837): Accesses to unmapped regions of debug mem RAL space does not return an error
	// response. Fix this if design is updated.
	debug_mem_ral.set_unmapped_access_ok(1);
	// Debug mem does not error on any type of sub-word writes.
	debug_mem_ral.set_supports_sub_word_csr_writes(1);
	end
	endfunction

I will open another PR once this is working.

[msfschaffner]

Update: it looks like the exceptions are fully defined for the rv_dm_debug_mem RAL model in the chip-level environment:

opentitan/hw/top_earlgrey/dv/env/chip_env_cfg.sv

Lines 224 to 234 in 97d123b

	// ROM within the debug mem is RO - it ignores writes instead of throwing an error response.
	`downcast(debug_mem_ral, ral)
	debug_mem_ral.rom.set_write_to_ro_mem_ok(1);
	debug_mem_ral.rom.set_mem_partial_write_support(1);
	// TODO(#10837): Accesses to unmapped regions of debug mem RAL space does not return an error
	// response. Fix this if design is updated.
	debug_mem_ral.set_unmapped_access_ok(1);
	// Debug mem does not error on any type of sub-word writes.
	debug_mem_ral.set_supports_sub_word_csr_writes(1);

That probably means that we need to add an exclusion tag to the ROM window stub which overlaps with the rv_dm_debug_mem RAL model, since the blanket ro swaccess attribute is not correct:

opentitan/hw/ip/rv_dm/data/rv_dm.hjson

Lines 105 to 114 in 97d123b

	rom: [
	// ROM size (given as `items` below) must be a power of two.
	{ window: {
	name: "ROM"
	items: "1024" // 4 KiB
	swaccess: "ro",
	desc: '''Access window into the debug ROM.'''
	}
	}
	]