JTAG state machine #379

aystarik · 2023-09-18T11:39:22Z

No description provided.

trabucayre · 2023-09-19T04:33:13Z

Hi,
Do you see any objection I apply some commits manually. There is straigforward ones

ch347 and GD32 are unrelated to the rest and may be applied directly
it's also true for d03ad17 and a45a9e5 seems a good improvement.

It will be more easy to have a closer looker to others a bit more tricky.

Thanks!

aystarik · 2023-09-19T05:24:40Z

Hi, not at all. Please feel free to cherry-pick however you like.

trabucayre · 2023-09-23T06:17:40Z

src/gowin.cpp

-#define spi_gowin_write(_wr, _rd, _len) do { \
-	_jtag->shiftDR(_wr, _rd, _len); \
-	_jtag->toggleClk(6); } while (0)
-


As mentioned previously I prefer a small, inline, function to avoid too many copy of the same code sequence.

yes, at some point I had a patch with this function added back, must have fell through...

src/gowin.cpp

trabucayre · 2023-09-24T06:31:46Z

src/gowin.cpp

+	 */
+	char mess[256];
+	try {
+		string hdr = _fs->getHeaderVal("checkSum");
 		if (hdr.empty()) {


must be !hdr.empty()

if empty -> can't compare so fail

if !empty -> to compare to know ko/ok

trabucayre · 2023-09-24T06:37:06Z

src/gowin.cpp

+			if (ucode != user_code) {
+				snprintf(mess, 256, "Read 0x%08x (checksum: 0x%08x, user_code: 0x%08x)\n",
+					 ucode, checksum, user_code);
+				printError("CRC check : FAIL");
+				printError(mess);
 			}
+			printSuccess("CRC check: Success");
 		}


maybe:

if (ucode == user_code) { printSuccess("CRC check: Success"); return; }

So if no match: message in lines 385-387 is displayed and this avoid to have twice error handling.

Also, in your code if fail, since there is not return the success message is always displayed

right, rewrote it

trabucayre · 2023-09-25T05:23:58Z

src/gowin.cpp

+	printInfo("Write SRAM ", false);
+	if (_verbose)
+		displayReadReg("before write sram", readStatusReg());
+	ProgressBar progress("Writing to SRAM", length, 50, _verbose);
+	send_command(CONFIG_ENABLE); // config enable
+	send_command(INIT_ADDR); // address initialize
+	send_command(XFER_WRITE); // transfer configuration data
+
+	_jtag->shiftDR(data, NULL, length);
+	progress.done();
+	send_command(0x0a);
+	uint32_t checksum = static_cast<FsParser *>(_fs)->checksum();
+	checksum = htole32(checksum);
+	_jtag->shiftDR((uint8_t *)&checksum, NULL, 32);
+	send_command(0x08);
+
+	send_command(CONFIG_DISABLE); // config disable
+	send_command(NOOP); // noop
+
+	if (_verbose)
+		displayReadReg("after write sram", readStatusReg());
+	if (readStatusReg() & STATUS_DONE_FINAL) {
+		printSuccess("DONE");


please keep progressBar (I know it's a slow xfer but it's more visual for user)

trabucayre · 2023-10-13T05:20:01Z

src/gowin.cpp

+	if (readStatusReg() & STATUS_DONE_FINAL) {
 		printError("FAIL");
 		return false;
+	} else {
+		printSuccess("DONE");
+		return true;
 	}


Why fail when flag is set?
Maybe also keeping test for STATUS_MEMORY_ERASE to check if this step is fine.

Take a look at UG290 p.38, figure 6-15. DoneFinal==0 is success for Erase SRAM and Erase FLASH
My understanding is that DoneFinal is FPGA has loaded config in SRAM. There is no flag for success of FLASH erase as such, so they issue load after erase and if FPGA stays clean, FLASH have been erased.

aystarik · 2023-10-19T11:14:01Z

Do you have plans to import these remaining patches or should just close the pull request?

trabucayre · 2023-10-19T11:23:09Z

Sorry for the big long delay.
I'm okay, I wish just to take time to do some test with real hardware. I have experimented, by the past, some issue with gowin devices (mainly littlebee) when something goes wrong.

trabucayre · 2023-10-20T05:27:29Z

I have tried tomorrow with a gw1n1: SRAM is working but FLASH no. I have to compare original code with modified one to see why.

aystarik · 2023-10-20T06:05:54Z

Did erase flash happen? Major difference in erase flash -- I've made "shift-dr 32 bit" only for GW1N-4 as written in documentation, it helped with erase of GW1N-9, but may be GW1N-1 still prefers to have it. In programming main difference with original core is that I do not add blocks before the bitstream, but modify 4 bytes with auto-load pattern in bitstream itself. I've checked with LA, and gowin programmer does exactly that. Actually, may be, you'll get some insight to what gowin programmer intentions are by using programmer_cli to generate svf file -- it can do that for flash programming, and it somewhat easier to read than the LA log.

trabucayre · 2023-10-21T06:33:27Z

Hi,
I have created a branch with a mix between your PR and the master branch. There is a few modifications (mainly to use disableCfg/enableCfg and writeFLASH is original one but I have to compare and merge). This version works with my gw1n1 device (and I have to test with all FPGAs I have).

Using SVF is a good way but has some drawback: there is no loop so a wait for a ''timeout'' duration is applied before simply reading status with a delay more or less dependent to the device and not always true for all. When using ILA or sniffing USB traffic you have sometime access to a more rich information. So a mix between these two approach is sometime good.

aystarik · 2023-10-23T10:15:34Z

Hi, I still can't find how you came up with

/* we have to send
* bootcode a X=0, Y=0 (4Bytes)
* 5 x 32 dummy bits
* full bitstream
*/

TN653 (TN653-1.07E, 11/18/2019) does not mention it, SVF and LA log from programmer do not show it...

trabucayre · 2023-10-23T10:27:14Z

In fact I have observed the 5x32 dummy bits by reading USB trafic, maybe this part is no more true and was a gowin programmer's bug.
I have to test by simply adding pattern at the beginning of the flash

aystarik · 2023-10-23T10:53:12Z

in my devel branch i have made a small exec which only reads the fs file -- it may be of use to compare the data from LA and from the file

aystarik · 2023-10-23T11:18:19Z

BTW, I can't find TN653 listed in gowin docs site, it might have been superseded by UG290

trabucayre · 2023-10-24T04:35:46Z

You can find this AN here

I have tried by removing dummy byte and it works.
When I have (2019) started to integrate gowin FPGAs I have compared fs file with USB packets content and seen this sequence (pattern + dummy) before configuration data. Maybe it's a bug somewhere, or a solution to obfuscate trafic...
The question here is to know if this pattern must be BEFORE fs or if the firt 4B must be overriden with this sequence...

aystarik · 2023-10-24T06:06:56Z

I've compared whole file data with output from LA (whole session) and patterns replace first 4 bytes in the data stream. Everything else is the same between file and LA stream

aystarik · 2023-10-24T06:09:35Z

No, I have no problem downloading the 2019 version of this file by link -- programmer manual still has one after the ug290 link. I've pointed out that it is not possible to find it in the list of documentation.

trabucayre · 2023-10-24T06:17:08Z

Ah ok. sorry for the mistake. I keep this TN because some informations was lost on UG290 (in fact I play with both app notes to find required informations).
Maybe I have to add a comment with the direct link to this file.

aystarik · 2023-10-24T06:38:00Z

Yes, documentation is worst chinese style -- text does not explain anything, figures contradict text, and there are several documents which contradict each other and the output of their own program... BTW, have you noticed that it is possible to write auto-boot pattern over read-back pattern without flash erase -- it only needs to clear some more bits.
'''
read_pattern = 0xf0f00f0f|ab_pattern
'''

trabucayre · 2023-10-29T04:58:32Z

Hi.
I have tested code with more or less all Gowin I have and updated my branch.
If you're agree with modifications I apply this patch.

For readback vs autoboot: it's may interesting to add this feature to verify code after write (or to force autoboot pattern after write).

Thanks.

aystarik · 2023-10-29T06:50:28Z

Hi, looks good, thanks!

trabucayre · 2023-10-29T07:08:41Z

I have applied gowin commit. Now it remains jtag rework part.

Thanks

aystarik · 2023-10-29T07:17:47Z

Do you need any help with jtag parts?

trabucayre · 2023-10-29T08:35:39Z

Not now. I have to review and to find the best compromise between speed and pedagogy.

aystarik · 2023-11-27T09:58:28Z

... to find the best compromise between speed and pedagogy.
Hi, could you please describe "compromise between speed and pedagogy" in more detail? I struggle to even put these words in the same sentence...

trabucayre · 2023-11-28T07:28:42Z

Exactly.
In one hand there is orignal code with detail on state transitions. But this code is slow since same transition are computed all the time.
In another hand there is your approach where all transition are precomputed (at compile time ?) and automatically faster. But it's less visual/pedagogic.

aystarik · 2023-11-28T08:14:55Z

But it's less visual/pedagogic.

My thinking was that printing of transitions from tms bits is enough? It is the commented out code in checking the correctness of tables in the beginning of the file? May be to use it if verbose was asked for?

aystarik · 2023-11-28T10:11:54Z

dive() -- this is recursive depth first search of the shortest path on the directed graph of the JTAG state machine, recursive is safe here as the number of nodes is fixed and quite small. create_paths() calls it for every combination of states thus creating square array of paths. create_paths is called from Jtag constructor once. print_path() can then pretty-print computed TMS sequence.

aystarik force-pushed the to-upstream branch from 4affc9a to 157d2a3 Compare September 18, 2023 14:11

LearnShareAlways mentioned this pull request Sep 18, 2023

CH347 stuck on load for Xilinx and chip id mismatch on Lattice FPGA #380

Open

aystarik force-pushed the to-upstream branch 4 times, most recently from 704f05b to 748094f Compare September 22, 2023 15:38

aystarik changed the title ~~GOWIN, CH347, JTAG state machine~~ GOWIN, JTAG state machine Sep 22, 2023

trabucayre reviewed Sep 23, 2023

View reviewed changes

src/gowin.cpp Outdated Show resolved Hide resolved

aystarik force-pushed the to-upstream branch from 748094f to 7d58177 Compare September 23, 2023 06:55

trabucayre reviewed Sep 24, 2023

View reviewed changes

src/gowin.cpp Outdated Show resolved Hide resolved

aystarik force-pushed the to-upstream branch from 7d58177 to b4e8de5 Compare September 24, 2023 06:04

trabucayre reviewed Sep 24, 2023

View reviewed changes

aystarik force-pushed the to-upstream branch 2 times, most recently from 453f5c9 to 8f27630 Compare September 24, 2023 06:58

trabucayre reviewed Sep 25, 2023

View reviewed changes

aystarik force-pushed the to-upstream branch 3 times, most recently from 0da3510 to ff320ed Compare September 25, 2023 09:10

aystarik force-pushed the to-upstream branch from ff320ed to 57e73c8 Compare October 10, 2023 16:56

trabucayre reviewed Oct 13, 2023

View reviewed changes

aystarik force-pushed the to-upstream branch from 57e73c8 to 1728f89 Compare October 17, 2023 15:36

aystarik force-pushed the to-upstream branch from 1728f89 to 5c00bd4 Compare November 1, 2023 09:53

aystarik changed the title ~~GOWIN, JTAG state machine~~ JTAG state machine Nov 1, 2023

aystarik force-pushed the to-upstream branch from 5c00bd4 to 1728f89 Compare November 28, 2023 14:39

precompute TMS sequencies

123b4ad

aystarik force-pushed the to-upstream branch from 1728f89 to 123b4ad Compare November 28, 2023 14:42

aystarik closed this Dec 8, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

JTAG state machine #379

JTAG state machine #379

aystarik commented Sep 18, 2023

trabucayre commented Sep 19, 2023

aystarik commented Sep 19, 2023 •

edited

Loading

trabucayre Sep 23, 2023

aystarik Sep 23, 2023

trabucayre Sep 24, 2023

trabucayre Sep 24, 2023

aystarik Sep 24, 2023

trabucayre Sep 25, 2023

trabucayre Oct 13, 2023

aystarik Oct 13, 2023

aystarik commented Oct 19, 2023

trabucayre commented Oct 19, 2023

trabucayre commented Oct 20, 2023

aystarik commented Oct 20, 2023

trabucayre commented Oct 21, 2023

aystarik commented Oct 23, 2023 •

edited

Loading

trabucayre commented Oct 23, 2023

aystarik commented Oct 23, 2023

aystarik commented Oct 23, 2023

trabucayre commented Oct 24, 2023

aystarik commented Oct 24, 2023

aystarik commented Oct 24, 2023

trabucayre commented Oct 24, 2023

aystarik commented Oct 24, 2023

trabucayre commented Oct 29, 2023

aystarik commented Oct 29, 2023

trabucayre commented Oct 29, 2023

aystarik commented Oct 29, 2023

trabucayre commented Oct 29, 2023

aystarik commented Nov 27, 2023

trabucayre commented Nov 28, 2023

aystarik commented Nov 28, 2023

aystarik commented Nov 28, 2023 •

edited

Loading

JTAG state machine #379

JTAG state machine #379

Conversation

aystarik commented Sep 18, 2023

trabucayre commented Sep 19, 2023

aystarik commented Sep 19, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

aystarik commented Oct 19, 2023

trabucayre commented Oct 19, 2023

trabucayre commented Oct 20, 2023

aystarik commented Oct 20, 2023

trabucayre commented Oct 21, 2023

aystarik commented Oct 23, 2023 • edited Loading

trabucayre commented Oct 23, 2023

aystarik commented Oct 23, 2023

aystarik commented Oct 23, 2023

trabucayre commented Oct 24, 2023

aystarik commented Oct 24, 2023

aystarik commented Oct 24, 2023

trabucayre commented Oct 24, 2023

aystarik commented Oct 24, 2023

trabucayre commented Oct 29, 2023

aystarik commented Oct 29, 2023

trabucayre commented Oct 29, 2023

aystarik commented Oct 29, 2023

trabucayre commented Oct 29, 2023

aystarik commented Nov 27, 2023

trabucayre commented Nov 28, 2023

aystarik commented Nov 28, 2023

aystarik commented Nov 28, 2023 • edited Loading

aystarik commented Sep 19, 2023 •

edited

Loading

aystarik commented Oct 23, 2023 •

edited

Loading

aystarik commented Nov 28, 2023 •

edited

Loading