Hi Will, hi Chris,
thanks for your hints. I thought it would be less risky only to reduce the
counts and not the size of the buffers.
I tweaked a bit with the values of MSYS_1 and found this is working for my
current app:
MSYS_1_BLOCK_COUNT: 10 # was 12
MSYS_1_BLOCK_SIZE: 180 # was 292
If I make BLOCK_SIZE smaller, taskstat and some other stats will still work,
but mpstat not:
It’s no longer Error 2, but:
Error: NMP timeout; op=0 group=0 id=3 seq=66 peer=bledefs.BleDev{AddrType:0,
Addr:bledefs.BleAddr{Bytes:[6]uint8{0xc, 0xa, 0xa, 0xa, 0xa, 0xa}}}
If MSYS_1_BLOCK_COUNT < 10, mpstat also no longer works.
All stat requests work with this 10x180 setting, except "stat att“.
But that doesn’t matter for me.
@Will regarding the stats, logging, …
I’m a freelancer since 30 years and had a look at mbed, riot, zephyr and mynewt
- Mynewt impressed me most.
I hope it will follow the footsteps of other great apache projects, like http
webserver, tomcat, Felix, ant, maven, wicket, cordova...
BTW:
every now and then I get this timeout error:
newtmgr -c ble0c stat ble_gattc
Error: Timeout waiting for host <-> controller sync
I’m using a microbit running the blehci, connected via adafruit FT232H with
rts/cts.
As an alternative, I tried to use the blehostd on a Raspberry Pi zero W with
these settings:
syscfg.vals:
OS_MAIN_TASK_PRIO: 1
# Disable logging. Writing lots of log output to the UART seems to cause
# the uart poller task to starve the system in some environemnts.
LOG_LEVEL: 255
# We're running in sim; it's OK to over-allocate.
MSYS_1_BLOCK_COUNT: 1000
BLE_ACL_BUF_COUNT: 1000
BLE_HCI_EVT_HI_BUF_COUNT: 1000
BLE_HCI_EVT_LO_BUF_COUNT: 1000
BLE_HCI_ACL_OUT_COUNT: 1000
BLE_MAX_CONNECTIONS: 64
# More stability; less correctness. This is a long-running process.
MCU_NATIVE_USE_SIGNALS: 0
BLE_SOCK_USE_LINUX_BLUE: 1
BLE_SOCK_USE_TCP: 0
BLE_SOCK_LINUX_DEV: 0
But it didn’t work:
pi@pizero:~/dev/mynewt/core-fork/apps/blehostd $ newtmgr -c ble0c echo ok
Error: blehostd did not connect to socket; controller not attached?
newtmgr conn show ble0c
Connection profiles:
ble0c: type=ble,
connstring='own_addr_type=random,peer_addr_type=public,peer_addr=0C:0A:0A:0A:0A:0A,bhd_path=blehostd.elf,ctlr_path=notused'
I tried with hci0 up or down, bluez stopped or running
Another bluetooth-tool is working fine on this raspi:
pi@pizero:~/dev/go/src/github.com/paypal/gatt $ ./discoverer
2017/06/16 19:46:29 dev: hci0 up
2017/06/16 19:46:29 dev: hci0 reset
2017/06/16 19:46:29 dev: hci0 down
2017/06/16 19:46:29 dev: hci0 opened
State: PoweredOn
scanning...
Peripheral ID:0C:0A:0A:0A:0A:0A, NAME:(master-bleprph)
Local Name = master-bleprph
…
Any hints, what typically goes wrong with this configuration?
Regards
Alfred
> Am 16.06.2017 um 21:08 schrieb will sanfilippo <wi...@runtime.io>:
>
> Alfred:
>
> Another thing with the nimble stack. The mbuf block size of 292, which Chris
> shows reduced to 200, is much larger than it needs to be. I am not sure what
> the smallest size actually is, but you can set it to 100 bytes and it should
> work. This can get you more buffers and/or reduce your memory usage quite a
> bit. The amount of buffers you actually need is quite application dependent
> and generally requires some experimentation unfortunately.
>
> And thanks for the kind words regarding the stats, logging and image update
> functionality. We appreciate it!
>
>> On Jun 16, 2017, at 10:39 AM, Christopher Collins <ch...@runtime.io> wrote:
>>
>> Hi Alfred,
>>
>> On Fri, Jun 16, 2017 at 07:02:47PM +0200, Alfred Schilken wrote:
>>> Hello all, I’m experimenting with mynewt for several weeks now and I’m
>>> especially impressed by the statistics, logging and image update
>>> functionality.
>>> I’m having problems with getting all the statistics using newtmgr via BLE.
>>>
>>> I took bleprph as base, added ADC and enabled several stats, logs and so
>>> on.
>>> To reduce flash size I had to switch off debug and reduce the logging.
>>>
>>> When I started the program I got an Assert-reboot-loop.
>>> The target board is a bbc microbit.
>>>
>>> This seems to be cause of the error:
>>> ble_att_svr_entry_mem = malloc(
>>> OS_MEMPOOL_BYTES(ble_hs_max_attrs, sizeof (struct ble_att_svr_entry)));
>>> if (ble_att_svr_entry_mem == NULL) {
>>> rc = BLE_HS_ENOMEM;
>>> goto err;
>>> }
>>
>> Fitting BLE and newtmgr into 16kB requires some creativity :). There
>> are certainly some memory and mbuf optimizations that could be made, we
>> just need to go through the exercise of scouring the code.
>>
>>> I tweaked these four config values to fix the crash-loop.
>>> BLE_ACL_BUF_COUNT: 4 # was 4
>>> BLE_ACL_BUF_SIZE: 128 # was 255
>>> MSYS_1_BLOCK_COUNT: 4 # was 12
>>> MSYS_1_BLOCK_SIZE: 292 # was 292
>>>
>>> The program boots now, I can see and connect to the BLE services and all
>>> this is fine.
>>>
>>> If I use newtmgr via ble transport to read the statistics, some of them are
>>> responding, others return just Error: 2.
>>> mpstat and taskstat also don’t work.
>>> But „image list“ is working, I could even upload a new image.
>>>
>>>
>>> My question is:
>>> What are the best config settings to reduce ram usage, so that an app runs
>>> in 16 kb with least impact on functionality?
>>
>> I was able to get mpstats working on a 16kB device with the following
>> mbuf settings:
>>
>> MSYS_1_BLOCK_COUNT: 11
>> MSYS_1_BLOCK_SIZE: 200
>>
>>> Some hints to reduce flash size would also be appreciated :-)
>>
>> You might want to look into the split image functionality:
>> https://mynewt.apache.org/latest/os/modules/split/split. This allows
>> you dedicate more flash space to your application code. I don't think
>> this will help with your immediates issues, however, since BLE and
>> newtmgr would both need to go into the loader image.
>>
>> Chris
>
