3
votes

currently I am working with a RISC-V processor implementation. I need to run partially hand-crafted assembly code. (Finally there will be dynamic code injection.) For this purpose I have to understand the basics of function calls within RISC-V assembly.

I found this topic very helpful: confusion about function call stack

But I am still struggling with the stack layout for a function call. Please consider the following c-code:

void some_func(int a, int b, int* c){
   int cnt = a;
   for(;cnt > 0;cnt--){
      *c += b;
   }
}

void main(){
   int a = 5;
   int b = 6;
   int c = 0;

   some_func(a,b,&c);
}

This program implements a basic multiplication by a sequence of additions. The derived assembly code (riscv64-unknown-elf-gcc -nostartfiles mul.c -o mul && riscv64-unknown-elf-objdump -D mul) looks like this:

0000000000010000 <some_func>:
   10000:   fd010113            addi    sp,sp,-48
   10004:   02813423            sd  s0,40(sp)
   10008:   03010413            addi    s0,sp,48
   1000c:   fca42e23            sw  a0,-36(s0)
   10010:   fcb42c23            sw  a1,-40(s0)
   10014:   fcc43823            sd  a2,-48(s0)
   10018:   fdc42783            lw  a5,-36(s0)
   1001c:   fef42623            sw  a5,-20(s0)
   10020:   0280006f            j   10048 <some_func+0x48>
   10024:   fd043783            ld  a5,-48(s0)
   10028:   0007a703            lw  a4,0(a5)
   1002c:   fd842783            lw  a5,-40(s0)
   10030:   00f7073b            addw    a4,a4,a5
   10034:   fd043783            ld  a5,-48(s0)
   10038:   00e7a023            sw  a4,0(a5)
   1003c:   fec42783            lw  a5,-20(s0)
   10040:   fff7879b            addiw   a5,a5,-1
   10044:   fef42623            sw  a5,-20(s0)
   10048:   fec42783            lw  a5,-20(s0)
   1004c:   fcf04ce3            bgtz    a5,10024 <some_func+0x24>
   10050:   00000013            nop
   10054:   02813403            ld  s0,40(sp)
   10058:   03010113            addi    sp,sp,48
   1005c:   00008067            ret

0000000000010060 <main>:
   10060:   fe010113            addi    sp,sp,-32
   10064:   00113c23            sd  ra,24(sp)
   10068:   00813823            sd  s0,16(sp)
   1006c:   02010413            addi    s0,sp,32
   10070:   00500793            li  a5,5
   10074:   fef42623            sw  a5,-20(s0)
   10078:   00600793            li  a5,6
   1007c:   fef42423            sw  a5,-24(s0)
   10080:   fe042223            sw  zero,-28(s0)
   10084:   fe440793            addi    a5,s0,-28
   10088:   00078613            mv  a2,a5
   1008c:   fe842583            lw  a1,-24(s0)
   10090:   fec42503            lw  a0,-20(s0)
   10094:   f6dff0ef            jal 10000 <some_func>
   10098:   00000013            nop
   1009c:   01813083            ld  ra,24(sp)
   100a0:   01013403            ld  s0,16(sp)
   100a4:   02010113            addi    sp,sp,32
   100a8:   00008067            ret

The important steps, that need clarification are: (some_func(int,int,int))

   10060:   fe010113            addi    sp,sp,-32
   10064:   00113c23            sd  ra,24(sp)
   10068:   00813823            sd  s0,16(sp)
   1006c:   02010413            addi    s0,sp,32

and: (main())

   10000:   fd010113            addi    sp,sp,-48
   10004:   02813423            sd  s0,40(sp)
   10008:   03010413            addi    s0,sp,48

From my understanding: The stackpointer is moved to make space for return-address and parameters. (Main might be a special case here.) How are the passed arguments treated when located on the stack? How are they obtained back? In general, the methodology is clear to me, but how would I hand-code this segment in order to work.

Regarding the related topic, the stack should look somewhat like

| ???                            |
| params for some_func() <???>  |
| ra of some_func()              |
| locals of main()       <int c> |
| locals of main()       <int b> |
| locals of main()       <int a> |
| params for main()      <None>  |

But that is pretty much it. Can anybody point out, how this is arranged, and how these two listings (function call) co-related?

2

2 Answers

1
votes

The first few parameters, type permitting, are passed in registers so they don't even appear on the stack. Other than that, it's unclear what you really want to know. If you do get some arguments that are on the stack, they stay there even after you adjust the stack pointer so you can still address them relative to the adjusted stack pointer or a frame pointer (here $s0 apparently).

The important steps, that need clarification are:

   10060:   fe010113            addi    sp,sp,-32  # allocate space
   10064:   00113c23            sd  ra,24(sp)      # save $ra
   10068:   00813823            sd  s0,16(sp)      # save $s0
   1006c:   02010413            addi    s0,sp,32   # set up $s0 as frame pointer
2
votes

What you want to know is specified by the RISC-V calling conventions.

Main points:

Function arguments are usually passed in the a0 to a7 registers, not on the stack. An argument is only passed via the stack if there is no room in the a* registers left.

Some registers are caller saved while others are callee saved (cf. Table 26.1, Chapter 26 RISC-V Assembly Programmer’s Handbook, in the RISC-V base specification, 2019-06-08 ratified). That means before calling a function the caller has to save all caller saved registers to the stack if it wants to retain their content. Similarly, the called function has to save all callee saved registers to the stack if it wants to use them for its own purposes.