Boot Freezes with a single P8-P9 connection; reproduced on two boards

GoogleGroups
1

I have found a reproducible boot freeze (on two boards); could someone else possibly check; it’s very easy to try.
No connections besides power plug and one (or two) wires from header P8 to P9

Scenario 1: connect P8-43 to P9-30, plug in power, only power light comes on and BBB does not boot up

Scenario 2: connect P8-44 to P9-28, plug in power, only power light comes on and BBB does not boot up
Scenario 3: connect both P8-43 to P9-30 and P8-44 to P9-28, plug in power, power light and all 4 LEDs come on steady and BBB does not boot up.

Discussion:

  1. doing these tests numerous times on 2 BBB’s did not hurt either one of the boards I have, but this behavior is so strange that it might be risky; please don’t be upset with me if it hurts your board.
  2. uname -a gives Linux beaglebone 3.8.13-bone70 #1 SMP Fri Jan 23 02:14:42 UTC 2015 armv7l GNU/Linux. My OS version is Debian wheezy 7.8
  3. without wires connected, after booting and doing a

sudo cat /sys/kernel/debug/pinctrl/44e10800.pinmux/pins shows
…
pin 42 (44e108a8) 0000002f (this is P8-43 in mode 7 which should be gpio2_8)
pin 43 (44e108ac) 0000002f (this is P8-44 in mode 7 which should be gpio2_9)
…
pin 102 (44e10998) 00000027 (this is P9-30 in mode 7 which should be gpio3_17)
pin 103 (44e1099c) 00000027 (this is P9-28 in mode 7 which should be gpio3_16)

Therefore at boot, pins P8-43 and P8-44, without applying device trees, are both configured as inputs with pull-downs;
pins P9-30 and P9-28, without applying device trees, are both configured as inputs with pull_up/pull_down’s disabled.

  1. making the connections after boot does not have any discernible effects

  2. It seems to me that having two gpio’s connected together with both configured as inputs and one having a pull-down should not cause the BBB to freeze at boot.

  3. It is also interesting that having two connections as in 5) causes different behavior; that is the boot process goes further to light all four LED’s and then freezes.

Could someone check to see if they are getting similar behavior; I can’t imagine that I am doing something real stupid with only a single wire connection; but I have previously done many rather stupid things so you never know (without verification).

Would someone in the expert category have some suggestions how to get around this problem; if we can’t find a work around, months and months of work will be thrown away. For example, if we configured the device tree that is loaded at boot up, would it take precedence before the default pin mux that causes the freeze (if this happens to be the problem, currently we have no insight into what is causing this behavior).

Has anyone else seen similar problems (we may also have something funny with connections to P8-45 and P8-46 but have not tracked this down yet to simple scenarios.

I’m open to suggestions as we are currently stuck. Thanks.

2

you are messing with the boot pins. its been stated here 100s of times
you cannot do anything to these pins before board boots.
check the BBB schematic and the using the bbb board guide.,

3

This is not clear as they are connected to other inputs which I would normally think means they are not being driven. Is there a definition you can give a link to defining what driven is? Also, can you supply a link of how to control and what are the mux defaults before boot? Thank you.

4

ANY load can affect how they are read. These pins are already well loaded as it is. If you want to use them, use a buffer to isolate them until after the processor is running.

Gerald

5

If they cause the BBB to not boot, by hooking an input to them, I would argue they should never have been brought out to the header, and that examples of how to handle this sensitivity while still making use of the header pins should be more readily available. The documentation on this sensitivity appears to be a single poorly written paragraph in the SRM which only states they should not be driven (actually 3 lines only in section 7.12.1 following Fig 58; to be specific: “If you plan to use any of these signals, then on power up, these pins should not be driven. If you do, it can affect the boot mode of the processor and could keep the processor from booting or working correctly.” I can’t see this addressed anywhere else in the documentation). It should state that not even inputs can not be connected to these header pins. There is no documentation that I can see on why they were even brought out to the header. Issues like this make other alternatives to using the BBB’s look more attractive (such as the new PI). At a minimum, this will cost us another 4 lost weeks and $2K for new populated boards for another iteration on our cape. Very frustrating, and I would argue both poor design and poor documentation. Looking at the schematic, it appears the existing load on each pin is 2 100K resistors? If this is correct, I would not call this “well loaded” when a typical gpio output current is 4-6mA.

6

Those pins are the LCD pins also. Please read the tables for the pin outs and you will see signals labeled LCD. That is why they wee brought out.

Sorry for the poorly written sentence. Sorry for your bad experience. I would have been happy to review your design before you went to that expense.

I will not take credit for designing the processor. I leave that to TI.

If you think the design is bad, I respect that. But I did not design the board for your application or your circuit.

Gerald

7

dont you prototype your very first one and test it before you order
production boards ?

If you would have searched this forum for boot mode or you may have
found your issue.

test a proto first. the information is there for the reading.

8

Kenneth:

There are fifteen lines, and a field of 100 K Ohm resistors that tell the Sitara how to boot.
There are locations for a pull up and a pull down on each of the boot instruction lines.
Only one of those resistors is populated for each line. Read the SRM.

Any load that would cause a logic state established by a 100K resistor, pull up or pull down, to change logic state, will modify, and likely kill the boot process.

If your “input” was a CMOS gate, it would probably have worked.
If your “input” was one of those Panasonic driver transistors with the built-in resistors to ground, you are changing the boot instructions by attaching it.

— Graham

9

inputs are not inputs are not inputs as far as loading goes.

if I connect an instrument with a 50 ohm impedance to a TTL/CMOS gate,
then it's going to see low at that gate input, even though they are
all inputs. Different logic series gates look like different things
at the inputs. Those inputs affect the other inputs.

The GPIO current out is not the issue, it's what the GPIO input sees
when the circuit is booted that determines what happens.

Harvey

10

That’s what the 4 weeks and $2K is for; another iteration on a prototype. We are long past where we can put out a prototype using either a white proto board, or hand soldering. We are now at a double sided, 4 layer board that is highly populated with on the top, and fairly populated on the bottom. With modern surface mount chips, we can not reliably place the components ourselves (we did for versions 1 and 2, the next version we have to put out is 5). The cape supports 4 I2C’s, 1 SPI, 1 UART, 1 CAN, 1 8-pin GPIO, 1 RTC, a programmable power supply for outgoing signals, 5V translation on all signals, an 8-input logic analyzer connected to PRU inputs similar to that of Kumar Abhishek (but with significant differences - only 8 bits, only 50MHz max, triggers on changes with an optional added delay using IEP timer) and numerous headers to make connecting to the outside world or to the built-in logic analyzer easy; 4 weeks and $2K is our costs not counting labour of putting out another iteration of our prototype. We intend to use the board primarily for our own use as it would cost too much for us to ever to make any profit selling it (in the volumes we expect we could achieve) so we doubt it will ever see production.

11

I down loaded the most recent SRM to make sure we are discussing the same one. In the most recent one, in Fig 56, I see 16 100K pull-up resistors and 16 42.2K pull-down resistors; although the schematic shows the pull-down resistors as being 100K different than the SRM. I can not see anywhere in the SRM that only some of these resistors are populated on the actual boards or which ones are actually populated; could you please tell me which page and paragraph. This is the first I have heard of this. I then went to page 6 of 11 of the schematic that shows all the resistors and I now see that some of them have DNI typed next to them. I can now guess (and this is only a guess) that DNI stands for Do Not Insert? but I can not see anywhere else about which resistors are populated and why? Could you please direct me to this documentation as it is necessary to understand the BBB?

On a related note, now that I understand only external pull-down or pull-up resistors are included, why use 100K resistors, why not use, for example 4.7K resistors as there is no dc current through them; this would mean connecting these pins to other GPIO inputs that have default pull-ups or pull-downs in the opposing direction would not cause errors?

Finally, shouldn’t the SRM show which resistors are present and which are not; it may be there but I can’t find it despite having looked a number of times; could you please direct me to where it is, and also documentation on how this effects the boot sequence?

Kenneth:

There are fifteen lines, and a field of 100 K Ohm resistors that tell the Sitara how to boot.
There are locations for a pull up and a pull down on each of the boot instruction lines.
Only one of those resistors is populated for each line. Read the SRM.

I have many times; could you please help me here and tell me which page describes which resistors are present and why? I can’t find it.

12

I totally understand their are many compromises in designs; I just wish there was documentation; in this case, I find it sadly inadequate. Finally, the input is another GPIO header pin that has a reset defaultof input with a pull down which I believe is about 23K (it took me 2 hours to find documentation on reset defaults - one statement in a very large TRM table describing registers stating to look at data sheets, about 1 hour to find and understand in data sheet where the mux reset defaults are given) . If the pull-up and pull-down resistors on the board had been chose to be something like 4.7K this error would not have happened, and I don’t believe this would have any impact on power except maybe 0.7mA additional current just during the time the boot button is pushed at power up, which is insignificant.

13

BeagleBone Black SRM Version C.1

Section 6.7 Boot Configuration Design, Page 67.

See Page 68, Figure 38.

Yes, DNI means Do Not Insert.

Or some time the term NP , meaning No Pop or “do not populate” is used.

Section 6.8 Default Boot Options, Page 68.

It is covered again on page 106, Section 8.3 Pin Usage Considerations.

— Graham

14

The discussion are about the affect of holding the boot pin down, not how the various SYS_BOOT signals affects the boot sequence. The DNI’s on the schematics were not in the previous SRM’s; I unfortunately, did not notice this change, and even if I had noticed it, I would not have known the DNI’s indicate the resistors were not included without this being explained somewhere in the text. I believe also that Fig. 39 on Page 68 may not have been included in earlier SRM’s (version C came out after most our design was done). Irrespective, I can only guess at what Fig. 39 means, as there is no written explanation. My guess would be: SYS_BOOT6-13 have no effect and are do not cares. SYS_BOOT14-15 change some clock frequency, but which clock is unknown. SYS_BOOT5 enables some clock, and SYS_BOOT04 should be either 11100 or 11000. The former, which is when the boot button is not pushed, boots MMC1 and then MMC0, whereas the latter boots SPI0 and then MMC0. I thought MMC1 was the microSD card? Is this not correct? Also, for the latter case, where SPI0 is booted first, I do not know what SPI0 is? Could you help here?

If the SRM included a couple of paragraphs here describing in words what is going on, it would make designing Capes much easier. Just stating in the text what DNI designated would really have helped.

15

Feel free to provide any wording that makes this understandable to all and I will consider adding it in.

Gerald

16

Simply DO NOT use GPIO2-6 to GPIO2-15 for inputs until AFTER the system has booted up. How do I know this?? …
(GPIO2-26 to GPIO2-31 are not exposed to P8 or P9.)

Fortunately for me it was just a matter of removing all 6 8 way resnets (48 inputs) and just use the internal pull ups making sure that the sensors attached to those pins do not trigger until after the boot, reasonably easy with my project.

I had used the same pins for outputs in another project but had no idea of the consequences of using them as input, easy to say Read The Fabulous Manual but there is so much that one cannot possibly foresee all. But I’m sure I never forget the lesson learned. :slight_smile:

17

Gerald, I’ve looked into it, and I don’t think I could do a good job as I can’t understand TI’s documentation. The boot process is described in Section 26.1 starting at pg 4898 (of SPRUH73K) and summarized in table 26-7 which takes 5 pages; one of the difficulties I have is I don’t know what all the acronyms are: for example, one of the boot options is SPI0; I think this is an EEPROM with an SPI interface, but I don’t think the BBB has such a device? If this is the case, why do we have it as one of the first boot devices? Do we need to consider booting from an EEPROM with an SPI interface? I believe the BBB has two SPI interfaces with the first one being used for the HDMI interface, and the second going to the header, so my guess is we can not boot from an EEPROM with an SPI interface. There are other options such as booting from a UART, or booting from an Ethernet interface, or booting from the USB interface (not sure which one), which also seem like very difficult choices to implement if we wanted to use them; are these boot options we want to consider for the BBB? Also, I’m guessing MMC0 in the tables refers to the eMMC, whereas MMC1 stands for booting the micro SD card, but these are only guesses; do you know if these guesses are correct. The TI reference manual in Chapt. 26 is difficult to understand, and it may not be possible to understand it without help from TI which is not available for the BBB. Given my failure to be able to understand Chapter 26, I could not do a reasonable job of documenting the use of the SYSBOOT pins; I’m guessing this aspect of the AM335X SOCs is perhaps convoluted and maybe not well done, or at least not well explained in TI’s documentation.

18

See inline below.

Gerald