|
<b>I am reposting this in an attempt to focus on the issue.
THIS IS NOT A BUMP! ( I know better ) </b>
I am well aware what volatile keyword does.
In my application it prevents accidental access to memory AKA RPi hardware.
The task is to map the hardware to memory. It works just fine resulting in "base" pointer to memory map.
Now I need to add specific offset to the "base" and have the "result" to be of volatile type.
The included snippet of test code does that with ONE exception - no matter what combination of code I use I cannot achieve VALID <b>volatile</b> result.
It only works when no volatile keyword is used , none at all.
I did try casts and dynamic_cast.
I have optimizer turned off and obviously doing something wrong applying pointers.
This code test snippet is located in main(), no other code is involved.
It gives pointer to "base" ( value is irrelevant) and adds "int" offset and produces new "result" pointer.
<pre lang="c++">
int *base;
int offset = 1;
int *result;
result = base + offset;
</pre>
Any help will be greatly appreciated.
Cheers
Vaclav
|
|
|
|
|
You have not declared result as volatile, and it is not clear what you are actually trying to achieve. How do you determine that the variable is not volatile?
|
|
|
|
|
Indeed. 
|
|
|
|
|
It's all cleared up .. ostream has a behaviour with volatile pointers
Quote: There are no overload for pointers to non-static member, pointers to volatile, or function pointers (other than the ones with signatures accepted by the (10-12) overloads). Attempting to output such objects invokes implicit conversion to bool, and, for any non-null pointer value, the value 1 is printed
It prints 1 because the volatile pointer he makes is non zero and that is what the ostream standard says it should do.
I must admit I hadn't run across it because I rarely use the ostream to output because of quirks like this.
In vino veritas
|
|
|
|
|
|
Richard is right; what you expect as a 'valid volatile result'? How do you check it?
|
|
|
|
|
<pre lang="c++">
cout << "base " << base << '\n';
cout << "offset " << offset << '\n';
cout << "result " << result << '\n';
</pre>
Please read the OP again.
Here is a repeat.
Base gets set to a pointer, simple , right?
I add an "offset" to it, still pretty straight.
I expect the result pointer to be base plus offset.
It is <b>AS LONG AS NONE</b> of the variables is typed with "volatile" keyword.
|
|
|
|
|
Your current code as posted will print some random value because you haven't set the variable base to a start value.
Specifically it will be what junk is the memory address it gets assigned, basic rules of C.
Could you please try this code
#include <stdio.h>
int main (void){
int *base = 0;
int offset = 1;
volatile int *result;
result = base + offset;
printf("Volatile result pointer = %p\n", result);
if ((int)result != sizeof(int)) printf("compiler is broken\n");
}
It should return 2, 4 or 8 for the result depending on the size of int on your compiler
As you are doing pointer addition the offset is defined as the sizeof(int) by C standards.
If you see the message "compiler is broken" there are 10 different gcc compilers for the Pi use another one or change version of your current.
GCC being a public domain support compiler does get bugs in versions from time to time.
As a last comment the above possible multiple answers is specifically why you shouldn't use int for hardware registers.
The registers on the Pi have to be accessed as 32bit, so the only safe way to do it is uint32_t
The following code will give an answer of 4 on EVERY C COMPILER, it is safe and portable and even works on the Pi3 64bit compilers
My advice is DO IT THIS WAY.
#include <stdint.h>
#include <stdio.h>
int main (void){
uint32_t *base = 0;
uint32_t offset = 1;
volatile uint32_t *result;
result = base + offset;
printf("Volatile result pointer = %p\n", result);
}
In vino veritas
modified 22-May-18 22:32pm.
|
|
|
|
|
OK, it's late but here is the latest.
This code with result declared as volatile PRINTS the correct value of "result".
However "cout" outputs value of 1 - that is ONE.
After removing "volatile" from declaration of result - cout will output 4.
I'll let you gentlemen decide where is my problem , it the meantime I am initializing the pointer to zero and will be checking for absence of zero and hoping computer is adding the values correctly.
BTW the correct additions of both values was never my concern , hence I did not care about the uninitialized value of "base" and was using "int" to test.
.
<pre lang="c++">/*
uint32_t *base = 0;
uint32_t offset = 1;
volatile uint32_t *result;
result = base + offset;
printf("Volatile result pointer = %p\n", result);
*/
uint32_t *base = 0;
uint32_t offset = 1;
volatile uint32_t *result;
result = base + offset;
cout << "base " << base << '\n';
cout << "offset " << offset << '\n';
THIS LINE IS AN UNRESOLVED ISSUE
OUTPUT IS CORRECT WHEN NO VOLATILE KEYWORD IS USED
result = 1 WHEN VOLATILE IS USED
cout << "result " << result << '\n';
this always prints correct value
printf("Volatile result pointer = %p\n", result);
</pre>
PS
ill try "bad compiler" in the morning.
|
|
|
|
|
That ostream object is a pile of overloaded functions on the operator <<
std::basic_ostream::operator<< - cppreference.com[^]
NOTE FROM ABOVE .. I underlined the statement parts:
Quote: There are no overload for pointers to non-static member, pointers to volatile, or function pointers (other than the ones with signatures accepted by the (10-12) overloads). Attempting to output such objects invokes implicit conversion to bool, and, for any non-null pointer value, the value 1 is printed
I avoid the thing like the plague because you have to know it's behaviour .. so yes a non zero volatile pointer will always print as 1 .. it tells you that.
So the good news is there is nothing wrong with your compiler it is following the c++ standard.
If you don't like it take it up with the c++ standard comittee.
Ostream has other strange (but documented) filter behaviour.
For example with char* pointers it has a special overload expecting it to be a pointer to a null terminated string
cout << (char*)0x41 << endl;
cout << (void*)0x41 << endl;
You could try putting a void* typecast so it goes thru the void* pointer overload ... like so should work
cout << "result " << (void*)result << '\n';
If you are intending to use ostream you will need to learn it's ins and outs and quirky behaviour.
Anyhow it's not a big issue and rolls around ostream behaviour .. so don't make it more than what it is.
Your pointers are clearly adding properly you were just having a display issue with ostream 
In vino veritas
modified 23-May-18 1:08am.
|
|
|
|
|
Leon and everybody, thanks for all your help.
Now I can sleep better.
One more "thing" - if I use uint32_t adding 1 results in pointer being incremented by 4. If I use uint8_t same process increments the pointer by 1.
So - is the BCM2835 memory "only" 8 bits wide?
Cheers and thanks again.
Vaclav
Somebody is trying to tell me something - I just got very close hit by lightning and lost power so I have to start over.
This "byte" reference make no sense to me.
If uint32 is 4 bytes wide - why did they name it "32"?
uint32 IS 32 bits variable and if adding 1 uint32 variable to the original pointer comes up with ADDRESS offset of 4 that means there are 4 8 bits wide memory locations AKA memory is 8 bits wide.
I know I am not the best communicator - but I never did question pointer arithmetic.
I do not get why some people always volunteer to "teach" and then complain about it.
Ever since I join this forum I appreciated straight talk, now it looks very similar to forum ,which should remain nameless, - everybody is a teacher and fails to read the posts.
Of course when I express my feelings I get flamed.
I better quit - the storm is getting too close.
-- modified 23-May-18 12:54pm.
|
|
|
|
|
Quote: So - is the BCM2835 memory "only" 8 bits wide?
What do you mean with that?
|
|
|
|
|
The uint32_t type (the clue is in the name) is 32 bits wide, or 4 bytes. The uint8_t is 1 byte. When you add 1 to a pointer it will increment it by the size of the object(s) pointed at. So a pointer to a uint32_t will increment by 4 bytes when adding 1, in order to point to the next variable of that type. Pointer arithmetic in C and C++ has always done this.
|
|
|
|
|
Vaclav_ wrote: if I use uint32_t adding 1 results in pointer being incremented by 4. If I use uint8_t same process increments the pointer by 1.
Pointer arithmetic in C.
Software rusts. Simon Stephenson, ca 1994. So does this signature. me, 2012
|
|
|
|
|
Quote: If uint32 is 4 bytes wide - why did they name it "32"?
4 bytes each of 8 bits = 4 x 8 = 32 bits .. its unsigned and a type .... so uint32_t
8 bytes of 8 bits = 8 x 8 = 64 bits .... so uint64_t also means 8 bytes.
Unlike int, long etc the sizes in bits and bytes are fixed in stone on every C compiler that meets the standard.
So we can basically say uint32_t is always 4 bytes, uint64_t is always 8 bytes on any C/C++ compiler
Yes correct it adds by 4 and as EVERYTHING on the PI 1,2 or 3 is aligned to 4 bytes they are an ARM RISC processor
and unaligned access is less than desirable. It is actually illegal and not predictable on many earlier ARM CPU's.
On your Pi3 which is 4 x Cortexa53 cores unaligned access is generally configured to raise an error exception but
can handle it if required. Now here is the big but, there is a MASSIVE penalty for unaligned access not just from the
CPU but the caches don't work properly.
The bottom line however is every hardware register is aligned to 4 bytes (that is it's divisible by 4) and can only
be written with 4 bytes at once (even in 64 bit mode). So the behaviour of uint32_t pointer addition is desirable and
ideal.
There is one further thing we generally do to make it compatible to for 64bit code for the Pi3 which is we declare
the pointers and/or structs representing hardware as align 4 with a GCC attribute. A typical sample will look like
this
struct __attribute__((__packed__, aligned(4))) SystemTimerRegisters {
uint32_t ControlStatus;
uint32_t TimerLo;
uint32_t TimerHi;
uint32_t Compare0;
uint32_t Compare1;
uint32_t Compare2;
uint32_t Compare3;
};
The reason is that the cortexa53 in AARCH64 mode natively runs everything in 8 byte alignment which it calls a word. It does however
have the ability to do half word (4 byte writes) and the aligned(4) is the way to tell it to use half word access. The packed
attribute is because without that the struct will have uint32_t spaced every 8 bytes (so 4 bytes uint32_t then 4 byte gap)
A volatile pointer to a hardware register is similarly written with an aligned 4
#define GPIO ((volatile __attribute__((aligned(4))) uint32_t*)(uintptr_t)(RPi_IO_Base_Addr + 0x200000))
So things written in this way work with 32bit and 64bit compilers for the Pi3.
It's not important on 32bit compilers but 64 bit compilers produce invalid code without it.
I program mainly in baremetal on the Pi .. so no O/S at all we just put our files on the blank SD card
GitHub - LdB-ECM/Raspberry-Pi: My public Baremetal Raspberry Pi code[^]
In vino veritas
modified 23-May-18 13:36pm.
|
|
|
|
|
Thanks Leon
it always makes more sense after your posts.
And as far as my project goes I got too confident and ended up with a mess , as always.
Luckily I had a backup so no lost.
If you recall I had some issues with memory mapping.
Ill have to prove this to myself , but it seems that when the memory map uses /dev/gpiomem I cannot map other devices such as SPI. Makes sense when you pointed out that each "/dev" has its own file.
I think it is just the way the BCM library tries to be "universal" and does everything in one file.
I actually only need SPI,I2C and GPIO.
And I have a class which handless the GPIO nicely.
Cheers
Vaclav
|
|
|
|
|
leon de boer wrote: 4 bytes each of 8 bits = 4 x 8 = 32 bits .. its unsigned and a type .... so uint32_t 
8 bytes of 8 bits = 8 x 8 = 64 bits .... so uint64_t also means 8 bytes.
Unlike int, long etc the sizes in bits and bytes are fixed in stone on every C compiler that meets the standard.
The size of "uint32_t" is fixed at four bytes on compilers which implement uint8_t. A conforming compiler where "char" is 16 bits (e.g. because it targets a platform where memory can only be written in 16-bit chunks) and either "int" or "long" is a 32-bit two's-complement integer without padding bits must, however, define type `uint16_t` with a size of 1 and `uint32_t` with a size of two.
|
|
|
|
|
What you have said is misleading you are just talking about a compiler that use the clause that says
Quote: Various implementations of C and C++ reserve 8, 9, 16, 32, or 36 bits for the storage of a byte
So a byte in your case is defined as 16 bits and that is why the uint8_t type doesn't exist for you.
So yes a uint32_t in your case would be 2 bytes (because your byte is 16 bits) and all you have proven is what we have long argued to the standards committee that C has butchered the word byte it should have been called register. I mean all you have is a 16 bit register that can't break to 8 bits that is creating the problem.
I have always wanted to see a C compiler with 9 bit bytes because I think that would make for some real fun
So lets reword it uint32_t is always 4 x uint8_t assuming both are implemented because we have to avoid using the word byte because it's been butchered in C.
In vino veritas
modified 5-Jun-18 3:12am.
|
|
|
|
|
I don't think "register" is a good term to describe a system's smallest addressable storage unit. If you don't like "byte", I'd suggest "smallest addressable storage unit" would be a precise term.
In any case, I was responding to the claim that the size of a uint32_t is "set in stone" at four bytes for all conforming implementations, which a reasonable person might interpret as saying that sizeof (uint32_t) is required to be 4.
I think it's rather silly that the Standard defines no category of programs between Strictly Conforming programs, a category so narrow as to be almost useless, and Conforming programs, a category so broad as to be essentially meaningless, and only defines two categories of implementations, either of which would be allowed (because of the "One Program" loophole) to behave in arbitrary fashion when given almost any program. If the Standard sought to actually categorize things usefully, separating out common-integer-size implementations from weird-integer-size implementations would make a lot of sense. Unfortunately, the authors of the Standard seem to go out of their way to avoid suggesting that some implementations should be considered inferior to others.
|
|
|
|
|
In the following code, I tried to return a 2D local array from the getArray function and display it in the main function.
I use the following line to display the matrix
cout << " " << *(*(ptr+i*COL)+j) but it is not displayed.
Does anyone know how to display the array?
#include <iostream>
#define ROW 3
#define COL 4
using namespace std;
int main()
{
int **ptr;
int **getArray();
ptr = getArray();
cout << "\n Array tab[][] in main(): " << endl << endl;
for (int i=0; i<ROW; i++)
{
for (int j=0; j<COL; j++)
cout << " " << *(*(ptr+i*COL)+j);
cout << endl;
}
return 0;
}
int **getArray()
{
static int tab[ROW][COL] = {
11, 12, 13, 14,
21, 22, 23, 24,
31, 32, 33, 34,
};
return (int**)tab;
}
|
|
|
|
|
Try this:
for (int j=0; j<COL; j++)
cout << " " << (int)*(ptr+i+j) << endl;
cout << endl;
|
|
|
|
|
Why this line?
cout << " " << (int)*(ptr+i+j) << endl;
Can you give me more details ?
Why do you need a type-casting conversion ?
|
|
|
|
|
Multidimensional arrays declared the way you have are just an abstraction for programmers,
since the same results can be achieved with a simple array, by multiplying its indices:
int jimmy [3][5]; // is equivalent to
int jimmy [15]; // (3 * 5 = 15)
In effect GetArray returns a 1D pointer array and you could have simply returned it as int* as it's really a big 1 dimensional array
You returned int** so the typecast is to deal with it but it would have been simpler like this
int *getArray()
{
static int tab[ROW][COL] = {
11, 12, 13, 14,
21, 22, 23, 24,
31, 32, 33, 34,
};
return (int*)tab;
}
Then the access becomes more obvious
int *ptr = getArray();
cout << "\n Array tab[][] in main(): " << endl << endl;
for (int i = 0; i<ROW; i++)
{
for (int j = 0; j<COL; j++)
cout << " " << ptr[i*COL + j];
cout << endl;
}
In vino veritas
|
|
|
|
|
For some reason, cout was printing the numbers in hex rather than decimal.
|
|
|
|
|
The return type for getArray needs to be int (*getArray())[COL];. An expression of type T [M][N] decays to type T (*)[N], not T **.
So your code needs to be
int main()
{
int (*ptr)[N];
int (*getArray())[N];
...
}
int (*getArray())[N]
{
...
return tab;
}
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
