Skip to contentSkip to conversation

Usage of "p" constraint in BPF inline asm

DISCUSSION Jose E. Marchesi 3 participants 13 messages Aug 10, 2023 – Aug 11, 2023 BPF

lore
Share & tools
lore
#0 lore raw

Hello.

We found that some of the BPF selftests use the "p" constraint in inline assembly snippets, for input operands for MOV (rN = rM) instructions.

This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

  #define __imm_ptr(name) [name]"p"(&name)

Example:

  int consume_first_item_only(void *ctx)
  {
        struct bpf_iter_num iter;
        asm volatile (
                /* create iterator */
                "r1 = %[iter];"
                [...]
                :
                : __imm_ptr(iter)
                : CLOBBERS);
        [...]
  }

Little equivalent reproducer:

  int bar ()
  {
    int jorl;
    asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
    return jorl;
  }

The "p" constraint is a tricky one. It is documented in the GCC manual section "Simple Constraints":

An operand that is a valid memory address is allowed. This is for ``load address'' and ``push address'' instructions.

p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.

There are two problems:

1. It is questionable whether that constraint was ever intended to be used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:

foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~

I would assume the same happens with aarch64, riscv, and most/all other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference.

To avoid that error, the usage of the "p" constraint in internal GCC instruction templates is supposed to be complemented by the 'a' modifier, like in:

     asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));

Internally documented (in GCC's final.cc) as:

%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference to that address.

That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand".

But...

2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN = rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:

     bar:
        #APP
            r1 = r10-4
        #NO_APP

In other cases we could conceibly also end with a 64-bit label that may overflow the 32-bit immediate operand of `rN = imm32' instructions:

r1 = foo

All of which is clearly wrong.

clang happens to do "the right thing" in the current usage of __imm_ptr in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

bar:

	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.

I would suggest to change that macro (and similar out of macro usages of the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Thoughts?

PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

reviewer
#1 lore raw
Quoted 102 lines · Jose E. Marchesi

Hello.

We found that some of the BPF selftests use the "p" constraint in inline assembly snippets, for input operands for MOV (rN = rM) instructions.

This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

   #define __imm_ptr(name) [name]"p"(&name)

Example:

   int consume_first_item_only(void *ctx)
   {
         struct bpf_iter_num iter;
         asm volatile (
                 /* create iterator */
                 "r1 = %[iter];"
                 [...]
                 :
                 : __imm_ptr(iter)
                 : CLOBBERS);
         [...]
   }

Little equivalent reproducer:

   int bar ()
   {
     int jorl;
     asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
     return jorl;
   }

The "p" constraint is a tricky one. It is documented in the GCC manual section "Simple Constraints":

An operand that is a valid memory address is allowed. This is for ``load address'' and ``push address'' instructions.

p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.

There are two problems:

1. It is questionable whether that constraint was ever intended to be used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:

foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~

I would assume the same happens with aarch64, riscv, and most/all other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference.

To avoid that error, the usage of the "p" constraint in internal GCC instruction templates is supposed to be complemented by the 'a' modifier, like in:

asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));

Internally documented (in GCC's final.cc) as:

%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference to that address.

That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand".

But...

2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN = rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:

bar:
#APP
r1 = r10-4
#NO_APP

In other cases we could conceibly also end with a 64-bit label that may overflow the 32-bit immediate operand of `rN = imm32' instructions:

r1 = foo

All of which is clearly wrong.

clang happens to do "the right thing" in the current usage of __imm_ptr in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

Quoted 6 lines
   bar:
	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Unfortunately, the modifier 'a' won't work for clang.

$ cat t.c 

int bar ()

{

int jorl;

  asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl)); 

return jorl;

}

$ gcc -O2 -g -S t.c 
$ clang --target=bpf -O2 -g -S t.c 

clang: ../lib/Target/BPF/BPFAsmPrinter.cpp:126: virtual bool {anonymous}::BPFAsmPrinter::PrintAsmMemoryOperand(const llvm::MachineInstr*, unsigned int, const char*, llvm::raw_ostream&): Assertion `Offs etMO.isImm() && "Unexpected offset for inline asm memory operand."' failed.
...

I guess BPF backend can try to add support for this 'a' modifier if necessary.

Quoted 8 lines

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.

I would suggest to change that macro (and similar out of macro usages of the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Could you specify what is the syntax ("r" constraint) which will work for both clang and gcc?

Quoted 9 lines

Thoughts?

PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

#2 lore raw
Quoted 104 lines · Jose E. Marchesi
Quoted 83 lines

Hello.
We found that some of the BPF selftests use the "p" constraint in inline
assembly snippets, for input operands for MOV (rN = rM) instructions. This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

   #define __imm_ptr(name) [name]"p"(&name)
Example:
   int consume_first_item_only(void *ctx)
   {
         struct bpf_iter_num iter;
         asm volatile (
                 /* create iterator */
                 "r1 = %[iter];"
                 [...]
                 :
                 : __imm_ptr(iter)
                 : CLOBBERS);
         [...]
   }
Little equivalent reproducer:
   int bar ()
   {
     int jorl;
     asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
     return jorl;
   }

The "p" constraint is a tricky one. It is documented in the GCC manual
section "Simple Constraints":
An operand that is a valid memory address is allowed. This is
for
``load address'' and ``push address'' instructions.
p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.
There are two problems:
1. It is questionable whether that constraint was ever intended to be
used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:
foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~
I would assume the same happens with aarch64, riscv, and
most/all
other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference. To avoid that error, the usage of the "p" constraint in internal
GCC
instruction templates is supposed to be complemented by the 'a' modifier, like in:
asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
Internally documented (in GCC's final.cc) as:
%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference
to that address.
That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand". But...
2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN =
rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:
bar:
#APP
r1 = r10-4
#NO_APP
In other cases we could conceibly also end with a 64-bit label
that
may overflow the 32-bit immediate operand of `rN = imm32' instructions:
r1 = foo
All of which is clearly wrong.
clang happens to do "the right thing" in the current usage of __imm_ptr
in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

Quoted 6 lines
   bar:
	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Unfortunately, the modifier 'a' won't work for clang.

$ cat t.c  int bar ()  {     int jorl;     asm volatile ("r1 =
%a[jorl]" : : [jorl]"p"(&jorl));     return jorl;  }  $ gcc -O2 -g -S
t.c  $ clang --target=bpf -O2 -g -S t.c  clang:
../lib/Target/BPF/BPFAsmPrinter.cpp:126: virtual bool
{anonymous}::BPFAsmPrinter::PrintAsmMemoryOperand(const
llvm::MachineInstr*, unsigned int, const char*, llvm::raw_ostream&):
Assertion `Offs
etMO.isImm() && "Unexpected offset for inline asm memory operand."' failed.
...
Quoted 2 lines

I guess BPF backend can try to add support for this 'a' modifier if necessary.

I wouldn't advise that: it is an internal GCC detail that just happens to work in inline asm. Also, even if you did that constraint may result in operands that are not single registers. It would be better to use "r" constraint instead.

Quoted 10 lines
Quoted 9 lines

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.
I would suggest to change that macro (and similar out of macro usages of
the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Quoted 2 lines

Could you specify what is the syntax ("r" constraint) which will work for both clang and gcc?

Instead of:

   #define __imm_ptr(name) [name]"p"(&name)

Use this:

   #define __imm_ptr(name) [name]"r"(&name)

That assures that the operand (the pointer value) will be available in the form of a single register.

Quoted 9 lines
Quoted 9 lines

Thoughts?
PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

reviewer
#3 lore raw
Quoted 104 lines · Jose E. Marchesi
Quoted 104 lines · Jose E. Marchesi
Quoted 83 lines

Hello.
We found that some of the BPF selftests use the "p" constraint in inline
assembly snippets, for input operands for MOV (rN = rM) instructions. This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

    #define __imm_ptr(name) [name]"p"(&name)
Example:
    int consume_first_item_only(void *ctx)
    {
          struct bpf_iter_num iter;
          asm volatile (
                  /* create iterator */
                  "r1 = %[iter];"
                  [...]
                  :
                  : __imm_ptr(iter)
                  : CLOBBERS);
          [...]
    }
Little equivalent reproducer:
    int bar ()
    {
      int jorl;
      asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
      return jorl;
    }

The "p" constraint is a tricky one. It is documented in the GCC manual
section "Simple Constraints":
An operand that is a valid memory address is allowed. This is
for
``load address'' and ``push address'' instructions.
p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.
There are two problems:
1. It is questionable whether that constraint was ever intended to be
used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:
foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~
I would assume the same happens with aarch64, riscv, and
most/all
other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference. To avoid that error, the usage of the "p" constraint in internal
GCC
instruction templates is supposed to be complemented by the 'a' modifier, like in:
asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
Internally documented (in GCC's final.cc) as:
%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference
to that address.
That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand". But...
2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN =
rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:
bar:
#APP
r1 = r10-4
#NO_APP
In other cases we could conceibly also end with a 64-bit label
that
may overflow the 32-bit immediate operand of `rN = imm32' instructions:
r1 = foo
All of which is clearly wrong.
clang happens to do "the right thing" in the current usage of __imm_ptr
in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

Quoted 6 lines
    bar:
	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Unfortunately, the modifier 'a' won't work for clang.

$ cat t.c  int bar ()  {     int jorl;     asm volatile ("r1 =
%a[jorl]" : : [jorl]"p"(&jorl));     return jorl;  }  $ gcc -O2 -g -S
t.c  $ clang --target=bpf -O2 -g -S t.c  clang:
../lib/Target/BPF/BPFAsmPrinter.cpp:126: virtual bool
{anonymous}::BPFAsmPrinter::PrintAsmMemoryOperand(const
llvm::MachineInstr*, unsigned int, const char*, llvm::raw_ostream&):
Assertion `Offs
etMO.isImm() && "Unexpected offset for inline asm memory operand."' failed.
...
Quoted 7 lines
Quoted 2 lines

I guess BPF backend can try to add support for this 'a' modifier if necessary.

I wouldn't advise that: it is an internal GCC detail that just happens to work in inline asm. Also, even if you did that constraint may result in operands that are not single registers. It would be better to use "r" constraint instead.

Sounds good. We also do not want to add support for this 'a' thing if there are alternatives.

Quoted 16 lines
Quoted 11 lines
Quoted 10 lines

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.
I would suggest to change that macro (and similar out of macro usages of
the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Quoted 2 lines

Could you specify what is the syntax ("r" constraint) which will work for both clang and gcc?

Instead of:

Quoted 8 lines
    #define __imm_ptr(name) [name]"p"(&name)

Use this:

    #define __imm_ptr(name) [name]"r"(&name)

That assures that the operand (the pointer value) will be available in the form of a single register.

Okay, this seems work for both gcc and clang. Eduard, what do you think about the above suggested change?

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

Thoughts?
PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

maintainer
#4 lore raw
Quoted 137 lines · Yonghong Song
Quoted 104 lines · Jose E. Marchesi
Quoted 104 lines · Jose E. Marchesi
Quoted 83 lines

Hello.
We found that some of the BPF selftests use the "p" constraint in inline
assembly snippets, for input operands for MOV (rN = rM) instructions. This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

    #define __imm_ptr(name) [name]"p"(&name)
Example:
    int consume_first_item_only(void *ctx)
    {
          struct bpf_iter_num iter;
          asm volatile (
                  /* create iterator */
                  "r1 = %[iter];"
                  [...]
                  :
                  : __imm_ptr(iter)
                  : CLOBBERS);
          [...]
    }
Little equivalent reproducer:
    int bar ()
    {
      int jorl;
      asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
      return jorl;
    }

The "p" constraint is a tricky one. It is documented in the GCC manual
section "Simple Constraints":
An operand that is a valid memory address is allowed. This is
for
``load address'' and ``push address'' instructions.
p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.
There are two problems:
1. It is questionable whether that constraint was ever intended to be
used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:
foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~
I would assume the same happens with aarch64, riscv, and
most/all
other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference. To avoid that error, the usage of the "p" constraint in internal
GCC
instruction templates is supposed to be complemented by the 'a' modifier, like in:
asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
Internally documented (in GCC's final.cc) as:
%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference
to that address.
That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand". But...
2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN =
rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:
bar:
#APP
r1 = r10-4
#NO_APP
In other cases we could conceibly also end with a 64-bit label
that
may overflow the 32-bit immediate operand of `rN = imm32' instructions:
r1 = foo
All of which is clearly wrong.
clang happens to do "the right thing" in the current usage of __imm_ptr
in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

Quoted 6 lines
    bar:
	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Unfortunately, the modifier 'a' won't work for clang.

$ cat t.c  int bar ()  {     int jorl;     asm volatile ("r1 =
%a[jorl]" : : [jorl]"p"(&jorl));     return jorl;  }  $ gcc -O2 -g -S
t.c  $ clang --target=bpf -O2 -g -S t.c  clang:
../lib/Target/BPF/BPFAsmPrinter.cpp:126: virtual bool
{anonymous}::BPFAsmPrinter::PrintAsmMemoryOperand(const
llvm::MachineInstr*, unsigned int, const char*, llvm::raw_ostream&):
Assertion `Offs
etMO.isImm() && "Unexpected offset for inline asm memory operand."' failed.
...
Quoted 7 lines
Quoted 2 lines

I guess BPF backend can try to add support for this 'a' modifier if necessary.

I wouldn't advise that: it is an internal GCC detail that just happens to work in inline asm. Also, even if you did that constraint may result in operands that are not single registers. It would be better to use "r" constraint instead.

Sounds good. We also do not want to add support for this 'a' thing if there are alternatives.

Quoted 20 lines
Quoted 11 lines
Quoted 10 lines

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.
I would suggest to change that macro (and similar out of macro usages of
the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Quoted 2 lines

Could you specify what is the syntax ("r" constraint) which will work for both clang and gcc?

Instead of:

    #define __imm_ptr(name) [name]"p"(&name)

Use this:

Quoted 7 lines
Quoted 4 lines
    #define __imm_ptr(name) [name]"r"(&name)

That assures that the operand (the pointer value) will be available in the form of a single register.

Okay, this seems work for both gcc and clang. Eduard, what do you think about the above suggested change?

BPF selftests are passing with this change.
The macro in question is used in 3 files:
- verifier_subprog_precision.c
- iters_state_safety.c
- iters_looping.c

I don't see any difference in the generated object files (at-least for cpuv4).

So, I guess we should be fine.

Quoted 12 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

Thoughts?
PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

#5 lore raw
Quoted 153 lines · Yonghong Song
Quoted 137 lines · Jose E. Marchesi
Quoted 104 lines
Quoted 104 lines · Jose E. Marchesi
Quoted 83 lines

Hello.
We found that some of the BPF selftests use the "p" constraint in inline
assembly snippets, for input operands for MOV (rN = rM) instructions. This is mainly done via the __imm_ptr macro defined in tools/testing/selftests/bpf/progs/bpf_misc.h:

    #define __imm_ptr(name) [name]"p"(&name)
Example:
    int consume_first_item_only(void *ctx)
    {
          struct bpf_iter_num iter;
          asm volatile (
                  /* create iterator */
                  "r1 = %[iter];"
                  [...]
                  :
                  : __imm_ptr(iter)
                  : CLOBBERS);
          [...]
    }
Little equivalent reproducer:
    int bar ()
    {
      int jorl;
      asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
      return jorl;
    }

The "p" constraint is a tricky one. It is documented in the GCC manual
section "Simple Constraints":
An operand that is a valid memory address is allowed. This is
for
``load address'' and ``push address'' instructions.
p in the constraint must be accompanied by address_operand as the predicate in the match_operand. This predicate interprets the mode specified in the match_operand as the mode of the memory reference for which the address would be valid.
There are two problems:
1. It is questionable whether that constraint was ever intended to be
used in inline assembly templates, because its behavior really depends on compiler internals. A "memory address" is not the same than a "memory operand" or a "memory reference" (constraint "m"), and in fact its usage in the template above results in an error in both x86_64-linux-gnu and bpf-unkonwn-none:
foo.c: In function ‘bar’:
foo.c:6:3: error: invalid 'asm': invalid expression as operand
6 | asm volatile ("r1 = %[jorl]" : : [jorl]"p"(&jorl));
| ^~~
I would assume the same happens with aarch64, riscv, and
most/all
other targets in GCC, that do not accept operands of the form A + B that are not wrapped either in a const or in a memory reference. To avoid that error, the usage of the "p" constraint in internal
GCC
instruction templates is supposed to be complemented by the 'a' modifier, like in:
asm volatile ("r1 = %a[jorl]" : : [jorl]"p"(&jorl));
Internally documented (in GCC's final.cc) as:
%aN means expect operand N to be a memory address
(not a memory reference!) and print a reference
to that address.
That works because when the modifier 'a' is found, GCC prints an "operand address", which is not the same than an "operand". But...
2. Even if we used the internal 'a' modifier (we shouldn't) the 'rN =
rM' instruction really requires a register argument. In cases involving automatics, like in the examples above, we easily end with:
bar:
#APP
r1 = r10-4
#NO_APP
In other cases we could conceibly also end with a 64-bit label
that
may overflow the 32-bit immediate operand of `rN = imm32' instructions:
r1 = foo
All of which is clearly wrong.
clang happens to do "the right thing" in the current usage of __imm_ptr
in the BPF tests, because even with -O2 it seems to "reload" the fp-relative address of the automatic to a register like in:

Quoted 6 lines
    bar:
	r1 = r10
	r1 += -4
	#APP
	r1 = r1
	#NO_APP

Unfortunately, the modifier 'a' won't work for clang.

$ cat t.c  int bar ()  {     int jorl;     asm volatile ("r1 =
%a[jorl]" : : [jorl]"p"(&jorl));     return jorl;  }  $ gcc -O2 -g -S
t.c  $ clang --target=bpf -O2 -g -S t.c  clang:
../lib/Target/BPF/BPFAsmPrinter.cpp:126: virtual bool
{anonymous}::BPFAsmPrinter::PrintAsmMemoryOperand(const
llvm::MachineInstr*, unsigned int, const char*, llvm::raw_ostream&):
Assertion `Offs
etMO.isImm() && "Unexpected offset for inline asm memory operand."' failed.
...
Quoted 7 lines
Quoted 2 lines

I guess BPF backend can try to add support for this 'a' modifier if necessary.

I wouldn't advise that: it is an internal GCC detail that just happens to work in inline asm. Also, even if you did that constraint may result in operands that are not single registers. It would be better to use "r" constraint instead.

Sounds good. We also do not want to add support for this 'a' thing if there are alternatives.

Quoted 20 lines
Quoted 11 lines
Quoted 10 lines

Which is what GCC would generate with -O0. Whether this is by chance or by design (Nick, do you know?) I don't think the compiler should be expected to do that reload driven by the "p" constraint.
I would suggest to change that macro (and similar out of macro usages of
the "p" constraint in selftests/bpf/progs/iters.c) to use the "r" constraint instead. If a register is what is required, we should let the compiler know.

Quoted 2 lines

Could you specify what is the syntax ("r" constraint) which will work for both clang and gcc?

Instead of:

    #define __imm_ptr(name) [name]"p"(&name)

Use this:

Quoted 7 lines
Quoted 4 lines
    #define __imm_ptr(name) [name]"r"(&name)

That assures that the operand (the pointer value) will be available in the form of a single register.

Okay, this seems work for both gcc and clang. Eduard, what do you think about the above suggested change?

BPF selftests are passing with this change.
The macro in question is used in 3 files:
- verifier_subprog_precision.c
- iters_state_safety.c
- iters_looping.c

Quoted 4 lines

I don't see any difference in the generated object files (at-least for cpuv4).

So, I guess we should be fine.

Note the same fix would be needed in the inline asm in selftests/bpf/progs/iters.c:iter_err_unsafe_asm_loop.

Quoted 13 lines
Quoted 13 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

Thoughts?
PS: I am aware that the x86 port of the kernel uses the "p" constraint
in the percpu macros (arch/x86/include/asm/percpu.h) but that usage is in a different context (I would assume it is used in x86 instructions that get constant addresses or global addresses loaded in registers and not automatics) where it seems to work well.

maintainer
#6 lore raw

On Thu, 2023-08-10 at 21:10 +0200, Jose E. Marchesi wrote:
[...]

Quoted 2 lines

Note the same fix would be needed in the inline asm in selftests/bpf/progs/iters.c:iter_err_unsafe_asm_loop.

Right, sorry. Tested with that change as well, no changes in the generated object files for clang. Can't grep "p" anywhere else in selftests.

[...]

#7 lore raw
Quoted 11 lines · Jose E. Marchesi

[...]

Quoted 2 lines

Note the same fix would be needed in the inline asm in selftests/bpf/progs/iters.c:iter_err_unsafe_asm_loop.

Right, sorry. Tested with that change as well, no changes in the generated object files for clang. Can't grep "p" anywhere else in selftests.

[...]

Thank you.

Ok, so since people seems to agree to the proposed fix, I will prepare a patch for this. First I will have to set up a kernel BPF development environment... but I guess it is about time for that ;)

Other than running the kernel selftests, is there any other testing you would recommend to a n00b like me, for changes like this (code generation changes)?

maintainer
#8 lore raw
Quoted 13 lines · Jose E. Marchesi
Quoted 10 lines · Jose E. Marchesi

[...]

Quoted 2 lines

Note the same fix would be needed in the inline asm in selftests/bpf/progs/iters.c:iter_err_unsafe_asm_loop.

Right, sorry. Tested with that change as well, no changes in the generated object files for clang. Can't grep "p" anywhere else in selftests.

[...]

Thank you.

Quoted 7 lines

Ok, so since people seems to agree to the proposed fix, I will prepare a patch for this. First I will have to set up a kernel BPF development environment... but I guess it is about time for that ;)

Other than running the kernel selftests, is there any other testing you would recommend to a n00b like me, for changes like this (code generation changes)?

What I do is run same kernel selftests as CI:
- test_verifier
- test_progs
- test_progs-no_alu32
- test_progs-cpuv4

Do you need any help with the environment itself?
(I can describe my setup if you need that).

maintainer
#9 lore raw

On Fri, 2023-08-11 at 15:18 +0300, Eduard Zingerman wrote:
[...]

Quoted 8 lines
Quoted 7 lines

Ok, so since people seems to agree to the proposed fix, I will prepare a patch for this. First I will have to set up a kernel BPF development environment... but I guess it is about time for that ;)

Other than running the kernel selftests, is there any other testing you would recommend to a n00b like me, for changes like this (code generation changes)?

Quoted 5 lines

What I do is run same kernel selftests as CI:
- test_verifier
- test_progs
- test_progs-no_alu32
- test_progs-cpuv4

Forgot "test_maps", sorry.

[...]

#10 lore raw
In reply to Eduard Zingerman
Quoted 23 lines · Jose E. Marchesi
Quoted 13 lines
Quoted 10 lines · Jose E. Marchesi

[...]

Quoted 2 lines

Note the same fix would be needed in the inline asm in selftests/bpf/progs/iters.c:iter_err_unsafe_asm_loop.

Right, sorry. Tested with that change as well, no changes in the generated object files for clang. Can't grep "p" anywhere else in selftests.

[...]

Thank you.

Quoted 7 lines

Ok, so since people seems to agree to the proposed fix, I will prepare a patch for this. First I will have to set up a kernel BPF development environment... but I guess it is about time for that ;)

Other than running the kernel selftests, is there any other testing you would recommend to a n00b like me, for changes like this (code generation changes)?

Quoted 8 lines

What I do is run same kernel selftests as CI:
- test_verifier
- test_progs
- test_progs-no_alu32
- test_progs-cpuv4

Do you need any help with the environment itself?
(I can describe my setup if you need that).

That would be useful yes, thank you.

maintainer
#11 lore raw
Quoted 4 lines · Jose E. Marchesi
Quoted 2 lines

Do you need any help with the environment itself?
(I can describe my setup if you need that).

That would be useful yes, thank you.

There are several things needed:
- bpf-next source code:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git
- Specific kernel configuration
- QEMU to run selftests in
- Root file system for QEMU to boot
- Means to kickstart tests execution inside the VM.

There is a script vmtest.sh located in the kernel repository:

tools/testing/selftests/bpf/vmtest.sh

Which takes care of kernel configuration, compilation, selftests compilation, qemu rootfs image download and test execution.

This is not exactly what I use but I tested in right now and it works with a few caveats. Explaining my setup would take longer so I'll start with this one. I will submit patches with fixes for caveats.

## Caveat #1: libc version

The script downloads rootfs image from predefined location on github (aws?) and that image is based on debian bullseye. libc version on my system is newer, so there is an error when test binaries built on my system are executed inside VM. So, I have to prepare my own rootfs image and point vmtest.sh to it. It might not be a problem in your case, if so -- skip the rest of the section.

Unfortunately, there is no option to override rootfs via command line of that script, so the following patch is needed:

```diff
diff --git a/tools/testing/selftests/bpf/vmtest.sh b/tools/testing/selftests/bpf/vmtest.sh
index 685034528018..3d0c7e7c0135 100755
--- a/tools/testing/selftests/bpf/vmtest.sh
+++ b/tools/testing/selftests/bpf/vmtest.sh
@@ -124,6 +124,15 @@ download_rootfs()
                exit 1
        fi
 
+        echo "download_rootfs: $ROOTFS_OVERRIDE"
+        if [[ "$ROOTFS_OVERRIDE" != "" ]]; then
+            if [[ ! -e $ROOTFS_OVERRIDE ]]; then
+               echo "Can't find rootfs image referred to by ROOTFS_OVERRIDE: $ROOTFS_OVERRIDE"
+               exit 1
+            fi
+            cat $ROOTFS_OVERRIDE | zstd -d | sudo tar -C "$dir" -x
+            exit
+        fi
        download "${ARCH}/libbpf-vmtest-rootfs-$rootfsversion.tar.zst" |
                zstd -d | sudo tar -C "$dir" -x
 }
```

Here is how to prepare the disk image for bookworm:

    $ git clone https://github.com/libbpf/ci libbpf-ci
    $ cd libbpf-ci
    $ sudo ./rootfs/mkrootfs_debian.sh -d bookworm
      # !! eddy -- is my user name locally, update accordingly
    $ sudo chown eddy libbpf-vmtest-rootfs-2023.08.11-bookworm-amd64.tar.zst
    $ export ROOTFS_OVERRIDE=$(realpath libbpf-vmtest-rootfs-2023.08.11-bookworm-amd64.tar.zst)

Script stores Qemu disk image in ~/.bpf_selftests/root.img . We need to prepare/update that image using the following command:

# !! make sure ROOTFS_OVERRIDE is set
$ cd <kernel-sources>
$ cd tools/testing/selftests/bpf
$ ./vmtest.sh -i

(Note: script uses sudo internally, so it might ask for password).

## Caveat #2: make headers

Kernel compilation command requires the following patch:

```diff
diff --git a/tools/testing/selftests/bpf/vmtest.sh b/tools/testing/selftests/bpf/vmtest.sh
index 685034528018..3d0c7e7c0135 100755
--- a/tools/testing/selftests/bpf/vmtest.sh
+++ b/tools/testing/selftests/bpf/vmtest.sh
@@ -137,6 +146,7 @@ recompile_kernel()
 
        ${make_command} olddefconfig
        ${make_command}
+        ${make_command} headers
 }
 
 mount_image()
```

## Running tests

Running tests is simple:

    $ cd <kernel-sources>
    $ cd tools/testing/selftests/bpf
    $ ./vmtest.sh -- ./test_verifier

The script will rebuild both kernel and selftests if necessary. The log should look as follows:

    $ ./vmtest.sh -- ./test_verifier
    Output directory: /home/eddy/.bpf_selftests
    ... build log ....
    [    0.000000] Linux version 6.5.0-rc4-g2adbb7637fd1-dirty ...
    ... boot log ...
    + /etc/rcS.d/S50-startup
    ./test_verifier
    #0/u BPF_ATOMIC_AND without fetch OK
    #0/p BPF_ATOMIC_AND without fetch OK
    #1/u BPF_ATOMIC_AND with fetch OK
    ... test_verifier log ...
    #524/p wide load from bpf_sock_addr.msg_src_ip6[3] OK
    Summary: 790 PASSED, 0 SKIPPED, 0 FAILED
    [    3.724015] ACPI: PM: Preparing to enter system sleep state S5
    [    3.725169] reboot: Power down
    Logs saved in /home/eddy/.bpf_selftests/bpf_selftests.2023-08-11_18-53-05.log

## Selecting individual tests

For test_verifier individual tests could be selected using command:

    $ ./vmtest.sh -- ./test_verifier -vv 42

(-vv forces detailed logging).

For test_progs/test_progs-no_alu32/test_progs-cpuv4 using the following command:

    $ ./vmtest.sh -- ./test_progs-cpuv4 -vvv -a verifier_ldsx

(-a stands for allow and filters tests by names).

test_maps do not take any options AFAIK.

---

Hope this helps.
Feel free to ask about any issues, or we can have a call in zoom.

#12 lore raw

Thanks. I will give it a try.

Quoted 148 lines · Jose E. Marchesi
Quoted 4 lines
Quoted 2 lines

Do you need any help with the environment itself?
(I can describe my setup if you need that).

That would be useful yes, thank you.

There are several things needed:
- bpf-next source code:
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git
- Specific kernel configuration
- QEMU to run selftests in
- Root file system for QEMU to boot
- Means to kickstart tests execution inside the VM.

There is a script vmtest.sh located in the kernel repository:

tools/testing/selftests/bpf/vmtest.sh

Which takes care of kernel configuration, compilation, selftests compilation, qemu rootfs image download and test execution.

This is not exactly what I use but I tested in right now and it works with a few caveats. Explaining my setup would take longer so I'll start with this one. I will submit patches with fixes for caveats.

## Caveat #1: libc version

The script downloads rootfs image from predefined location on github (aws?) and that image is based on debian bullseye. libc version on my system is newer, so there is an error when test binaries built on my system are executed inside VM. So, I have to prepare my own rootfs image and point vmtest.sh to it. It might not be a problem in your case, if so -- skip the rest of the section.

Unfortunately, there is no option to override rootfs via command line of that script, so the following patch is needed:

```diff

tools/testing/selftests/bpf/vmtest.sh +9 -0
index 685034528018..3d0c7e7c0135 100755
@@ -124,6 +124,15 @@ download_rootfs()
                exit 1
        fi
 
+        echo "download_rootfs: $ROOTFS_OVERRIDE"
+        if [[ "$ROOTFS_OVERRIDE" != "" ]]; then
+            if [[ ! -e $ROOTFS_OVERRIDE ]]; then
+               echo "Can't find rootfs image referred to by ROOTFS_OVERRIDE: $ROOTFS_OVERRIDE"
+               exit 1
+            fi
+            cat $ROOTFS_OVERRIDE | zstd -d | sudo tar -C "$dir" -x
+            exit
+        fi
        download "${ARCH}/libbpf-vmtest-rootfs-$rootfsversion.tar.zst" |
                zstd -d | sudo tar -C "$dir" -x
 }
```


Here is how to prepare the disk image for bookworm:

    $ git clone https://github.com/libbpf/ci libbpf-ci
    $ cd libbpf-ci
    $ sudo ./rootfs/mkrootfs_debian.sh -d bookworm
      # !! eddy -- is my user name locally, update accordingly
    $ sudo chown eddy libbpf-vmtest-rootfs-2023.08.11-bookworm-amd64.tar.zst
    $ export ROOTFS_OVERRIDE=$(realpath libbpf-vmtest-rootfs-2023.08.11-bookworm-amd64.tar.zst)

Script stores Qemu disk image in ~/.bpf_selftests/root.img .
We need to prepare/update that image using the following command:

      # !! make sure ROOTFS_OVERRIDE is set
    $ cd <kernel-sources>
    $ cd tools/testing/selftests/bpf
    $ ./vmtest.sh -i

(Note: script uses sudo internally, so it might ask for password).

## Caveat #2: make headers

Kernel compilation command requires the following patch:

```diff
tools/testing/selftests/bpf/vmtest.sh +1 -0
index 685034528018..3d0c7e7c0135 100755
@@ -137,6 +146,7 @@ recompile_kernel()
 
        ${make_command} olddefconfig
        ${make_command}
+        ${make_command} headers
 }
 
 mount_image()

```

## Running tests

Running tests is simple:

    $ cd <kernel-sources>
    $ cd tools/testing/selftests/bpf
    $ ./vmtest.sh -- ./test_verifier

The script will rebuild both kernel and selftests if necessary. The log should look as follows:

    $ ./vmtest.sh -- ./test_verifier
    Output directory: /home/eddy/.bpf_selftests
    ... build log ....
    [    0.000000] Linux version 6.5.0-rc4-g2adbb7637fd1-dirty ...
    ... boot log ...
    + /etc/rcS.d/S50-startup
    ./test_verifier
    #0/u BPF_ATOMIC_AND without fetch OK
    #0/p BPF_ATOMIC_AND without fetch OK
    #1/u BPF_ATOMIC_AND with fetch OK
    ... test_verifier log ...
    #524/p wide load from bpf_sock_addr.msg_src_ip6[3] OK
    Summary: 790 PASSED, 0 SKIPPED, 0 FAILED
    [    3.724015] ACPI: PM: Preparing to enter system sleep state S5
    [    3.725169] reboot: Power down
    Logs saved in /home/eddy/.bpf_selftests/bpf_selftests.2023-08-11_18-53-05.log

## Selecting individual tests

For test_verifier individual tests could be selected using command:

    $ ./vmtest.sh -- ./test_verifier -vv 42

(-vv forces detailed logging).

For test_progs/test_progs-no_alu32/test_progs-cpuv4 using the following command:

    $ ./vmtest.sh -- ./test_progs-cpuv4 -vvv -a verifier_ldsx

(-a stands for allow and filters tests by names).

test_maps do not take any options AFAIK.

---

Hope this helps.
Feel free to ask about any issues, or we can have a call in zoom.