5 ms·

From Zero to main(): Bare metal C (2019)

255 points by olalonde 4y ago

kaycebasques 4y ago
Courtesy heads up for any hardware n00bs like myself who are buying the parts to work through this series hands-on: some of the required parts are hard to find. You should double-check that your replacement parts will work. https://news.ycombinator.com/item?id=33360830 https://news.ycombinator.com/item?id=33360830
- 082349872349872 4y ago
  Even as a pure software weenie it may be worth playing with gcc -nostdlib sometime, just to get a feel for what is done for you in crt0 and libc.
  - galangalalgol 4y ago
    Had a class back in college where the first project was to write a tiny stdlib in 68000 assembly, then we got to call it from c to write a simple os. It was very illuminating.
  - pitaj 4y ago
    On that topic, god I wish it was possible to get some of the crt without the stdlib. I wish I didn't have to manually implement 64-bit division in those cases.
  - sargstuff 4y ago
    Much more concise answer than from [1], but [1] is free advice less internet connection cost(s): 1) 'Is main() a required startup function?' & 'Is standard library initialization required for main() startup?' 2) But sdiff -s of answer from [1] to 'Is a dynamic linker required for startup page alignement?' with 'Is a dynamic linker required for startup page alignment?' is kinda bit verbose. 3) 'Is standard library initialization required for main() startup? How much resources should be allocated to startup libraries?' [1] answer sounds like to overly broad an approach. [1] : https://ask-paulg.up.railway.app/ https://ask-paulg.up.railway.app/
_8j50 4y ago
> Thou shalt initialize thy static variables, else The Machine shall set them to zero. Isn't that last part a bad assumption since it could be arbitrary values as well? Because people expecting a default zero value might be surprised otherwise.
- shadowofneptune 4y ago
  > If an object that has automatic storage duration is not initialized explicitly, its value is indeterminate. If an object that has static storage duration is not initialized explicitly, then: • if it has pointer type, it is initialized to a null pointer; • if it has arithmetic type, it is initialized to (positive or unsigned) zero; [end quote] —C99 Standard, §6.7.8 EDIT: That said, it can only relied upon to be zero at the first call of the function.
  - InfiniteRand 4y ago
    I’ve seen a couple boards that purely ignore this, at least for some TI targets their position is if they supply some obscure linker option that makes it possible to initialize static memory then they’ve satisfied the standard
    - Kamq 4y ago
      Surely this would be the compiler's job to emit the right instructions. I mean, they can do things that make it easier or harder, but satisfying a language spec isn't what they do (unless you want to count an ISA as a language spec, which it kind of is).
- monocasa 4y ago
  In addition to what's been stated in the c standard below, the way this works is these variables are included in the .bss section, which means in a bare metal env you memset it to zero at init about the same time you copy a fresh version of .data over from flash.
travisgriggs 4y ago
Ahh the SAMD21G18. Of all the chips to show up on HN today. This chip I know not half as much as I'd like to and half as much I wish I didn't have to. We ported a rarely heard of RTOS to this chip (TNEO), and one of the most difficult parts was running tickless with sleepable states. The difficulty (discovered only by blood, sweat, tears, wtf oscope moments, dumb luck, and a random comment found on some obscure forum, definitely not in the data sheets) was that accessing certain registers (in our case RTC.CNT) would cause the chip to go in to a different clock domain temporarily and suddenly take as much as 300 usecs to read. Just this line: uint currentClock = RTC.CNT; That's all it took. Whereas most store some register into some local variable took nanoseconds, this one took a long long time. My experience in this world of embedded chips, is that many of the tutorials you read do not deal with power consumption really. They assume that you're playing with a dev board. Or that your application has copious amounts of power. But if you live in a battery powered world, your entire game lies around navigating between sleep and wake states on the chip.
- not_the_fda 4y ago
  I hate that chip. I had a client pick it for a project that they wanted low power. You can't put the chip fully to sleep. There is errata on the NVM controller that it doesn't always wake from sleep so when the chip wakes it executes garbage. You have to keep the NVM controller powered during sleep to recover from sleep.
  - 4gotunameagain 4y ago
    I was able to get this microcontroller into deep sleep of the range of a few microamps, and would always reliably wake up from the RTC interrupt. I wonder what caused your issues ?
    - not_the_fda 4y ago
      You may have had newer silicon that fixed the issue, or you didn't have NVMCTRL fully asleep. Its in the errata "1.14.2 Power-down Modes and Wake-up From Sleep"
keepquestioning 4y ago
Bad news.
BrianHenryIE 4y ago
Similarly interesting: CppCon 2018: Matt Godbolt “The Bits Between the Bits: How We Get to main()” https://www.youtube.com/watch?v=dOfucXtyEsU https://www.youtube.com/watch?v=dOfucXtyEsU
teleforce 4y ago
New and soon to be published book on bare metal C by Steve Oualline (authors of several C programming books): https://nostarch.com/bare-metal-c https://nostarch.com/bare-metal-c
sedivy94 4y ago
This article took me back to my ECE courses. I haven’t read C since 2017 and I’m a glorified Powershell script kiddie these days. Thanks for the nostalgia.

255 points by olalonde 4y ago

Courtesy heads up for any hardware n00bs like myself who are buying the parts to work through this series hands-on: some of the required parts are hard to find. You should double-check that your replacement parts will work. https://news.ycombinator.com/item?id=33360830 https://news.ycombinator.com/item?id=33360830

082349872349872 4y ago
Even as a pure software weenie it may be worth playing with gcc -nostdlib sometime, just to get a feel for what is done for you in crt0 and libc.
- galangalalgol 4y ago
  Had a class back in college where the first project was to write a tiny stdlib in 68000 assembly, then we got to call it from c to write a simple os. It was very illuminating.
- pitaj 4y ago
  On that topic, god I wish it was possible to get some of the crt without the stdlib. I wish I didn't have to manually implement 64-bit division in those cases.
- sargstuff 4y ago
  Much more concise answer than from [1], but [1] is free advice less internet connection cost(s): 1) 'Is main() a required startup function?' & 'Is standard library initialization required for main() startup?' 2) But sdiff -s of answer from [1] to 'Is a dynamic linker required for startup page alignement?' with 'Is a dynamic linker required for startup page alignment?' is kinda bit verbose. 3) 'Is standard library initialization required for main() startup? How much resources should be allocated to startup libraries?' [1] answer sounds like to overly broad an approach. [1] : https://ask-paulg.up.railway.app/ https://ask-paulg.up.railway.app/

_8j50 4y ago

> Thou shalt initialize thy static variables, else The Machine shall set them to zero. Isn't that last part a bad assumption since it could be arbitrary values as well? Because people expecting a default zero value might be surprised otherwise.

shadowofneptune 4y ago
> If an object that has automatic storage duration is not initialized explicitly, its value is indeterminate. If an object that has static storage duration is not initialized explicitly, then: • if it has pointer type, it is initialized to a null pointer; • if it has arithmetic type, it is initialized to (positive or unsigned) zero; [end quote] —C99 Standard, §6.7.8 EDIT: That said, it can only relied upon to be zero at the first call of the function.
- InfiniteRand 4y ago
  I’ve seen a couple boards that purely ignore this, at least for some TI targets their position is if they supply some obscure linker option that makes it possible to initialize static memory then they’ve satisfied the standard
  - Kamq 4y ago
    Surely this would be the compiler's job to emit the right instructions. I mean, they can do things that make it easier or harder, but satisfying a language spec isn't what they do (unless you want to count an ISA as a language spec, which it kind of is).
monocasa 4y ago
In addition to what's been stated in the c standard below, the way this works is these variables are included in the .bss section, which means in a bare metal env you memset it to zero at init about the same time you copy a fresh version of .data over from flash.

travisgriggs 4y ago

Ahh the SAMD21G18. Of all the chips to show up on HN today. This chip I know not half as much as I'd like to and half as much I wish I didn't have to. We ported a rarely heard of RTOS to this chip (TNEO), and one of the most difficult parts was running tickless with sleepable states. The difficulty (discovered only by blood, sweat, tears, wtf oscope moments, dumb luck, and a random comment found on some obscure forum, definitely not in the data sheets) was that accessing certain registers (in our case RTC.CNT) would cause the chip to go in to a different clock domain temporarily and suddenly take as much as 300 usecs to read. Just this line: uint currentClock = RTC.CNT; That's all it took. Whereas most store some register into some local variable took nanoseconds, this one took a long long time. My experience in this world of embedded chips, is that many of the tutorials you read do not deal with power consumption really. They assume that you're playing with a dev board. Or that your application has copious amounts of power. But if you live in a battery powered world, your entire game lies around navigating between sleep and wake states on the chip.

not_the_fda 4y ago
I hate that chip. I had a client pick it for a project that they wanted low power. You can't put the chip fully to sleep. There is errata on the NVM controller that it doesn't always wake from sleep so when the chip wakes it executes garbage. You have to keep the NVM controller powered during sleep to recover from sleep.
- 4gotunameagain 4y ago
  I was able to get this microcontroller into deep sleep of the range of a few microamps, and would always reliably wake up from the RTC interrupt. I wonder what caused your issues ?
  - not_the_fda 4y ago
    You may have had newer silicon that fixed the issue, or you didn't have NVMCTRL fully asleep. Its in the errata "1.14.2 Power-down Modes and Wake-up From Sleep"

keepquestioning 4y ago

Bad news.

BrianHenryIE 4y ago

Similarly interesting: CppCon 2018: Matt Godbolt “The Bits Between the Bits: How We Get to main()” https://www.youtube.com/watch?v=dOfucXtyEsU https://www.youtube.com/watch?v=dOfucXtyEsU

teleforce 4y ago

New and soon to be published book on bare metal C by Steve Oualline (authors of several C programming books): https://nostarch.com/bare-metal-c https://nostarch.com/bare-metal-c

sedivy94 4y ago

This article took me back to my ECE courses. I haven’t read C since 2017 and I’m a glorified Powershell script kiddie these days. Thanks for the nostalgia.