target/riscv: select DMI IR on batch access

[en-sc]

Without the selection the TAP can be left in bypass.

Change-Id: I79c6bf74802dc9c9475947d1787a3d0b797f3952

[en-sc]

@JanMatCodasip, can you please prioritize looking at this one. It's an uncaught bug in #1073.

[JanMatCodasip]

@en-sc I have checked this fix visually and it looks good. Thank you.

Few additional notes (perhaps for future pull requests):

• The calls to select_dmi() in assert_reset(), deassert_reset() and read_memory_progbuf() appear to be redundant. The underlying batch API or dm_read/dm_write() API should always take care of that.
• I would be in favor of having assertions that the DMI has been selected, however the BSCAN tunnel support complicates that - the assertions would not be simple.
• Furthermore, I am afraid we don't have any way how to test the BSCAN tunnel support. (Or do we?)
• I would eventually like to move all the JTAG DTM related items to their own module (riscv013-jtag-dtm.c/h). Then the select_dmi() could be called directly from batch.c, which would put the select_dmi() call closer to the actual execution of the JTAG scans.

[en-sc]

@JanMatCodasip, please, take a look again.
I would like to separate the check in select_dmi() into #1091 and test it some more.

[en-sc]

• The calls to select_dmi() in assert_reset(), deassert_reset() and read_memory_progbuf() appear to be redundant. The underlying batch API or dm_read/dm_write() API should always take care of that.

I think you are right. Filed #1092.

• I would be in favor of having assertions that the DMI has been selected, however the BSCAN tunnel support complicates that - the assertions would not be simple.

I'm not sure it's worthwhile. We do have assertions already JTAG common interface that cover this.

• Furthermore, I am afraid we don't have any way how to test the BSCAN tunnel support. (Or do we?)

AFAIK, we don't.

• I would eventually like to move all the JTAG DTM related items to their own module (riscv013-jtag-dtm.c/h). Then the select_dmi() could be called directly from batch.c, which would put the select_dmi() call closer to the actual execution of the JTAG scans.

I'd like that to. This is essentially the goal of all these commits -- to tidy up the work with DM and move it to a separate file after this.

[JanMatCodasip]

LGTM

[JanMatCodasip]

en-sc: I would like to separate the check in select_dmi() into #1091 and test it some more.

Sure, no problem!

JanMatCodsip: I would be in favor of having assertions that the DMI has been selected (...)

en-sc: I'm not sure it's worthwhile. We do have assertions already JTAG common interface that cover this.

Ok, let's leave it as is, without additional asserts.

JanMatCodasip: I would eventually like to move all the JTAG DTM related items to their own module (...)

en-sc: I'd like that to. This is essentially the goal of all these commits (...)

Great :)