Re: [PATCH] bpf: Added linker support for R_BPF_64_NODYLD32.

[Cupertino Miranda <cupertino.miranda@oracle.com>]

[-- Attachment #1: Type: text/plain, Size: 288 bytes --]


Hi Jose,

Thanks for your review.
I realized I missed to update the macro for the relocation where the
properties like size, overflow check, (etc.) are defined.
I have added the comment near the relocation macro definition.

Sending the patch for review once again.

Regards,
Cupertino


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-bpf-Added-linker-support-for-R_BPF_64_NODYLD32.patch --]
[-- Type: text/x-diff, Size: 2137 bytes --]

From bb8afde9a8f17d42393067c3a667bec1f08b252a Mon Sep 17 00:00:00 2001
From: Cupertino Miranda <cupertino.miranda@oracle.com>
Date: Wed, 20 Dec 2023 12:24:03 +0000
Subject: [PATCH] bpf: Added linker support for R_BPF_64_NODYLD32.

This patch adds linker support to patch R_BPF_64_NODYLD32 relocations.
The implementation was based on comments and code in LLVM, as the GNU
toolchain does not uses this relocation type.
---
 bfd/bpf-reloc.def | 17 ++++++++++-------
 bfd/elf64-bpf.c   |  1 +
 2 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/bfd/bpf-reloc.def b/bfd/bpf-reloc.def
index 7e7497892fa..42ba1a169ea 100644
--- a/bfd/bpf-reloc.def
+++ b/bfd/bpf-reloc.def
@@ -90,17 +90,20 @@
 
   /* R_BPF_64_NODYLD32 is not used by GNU tools - but it is generated by LLVM.
      We provide an entry here so that tools like strip can safely handle BPF
-     binaries generated by other tools.  */
+     binaries generated by other tools.
+     R_BPF_64_NODYLD32 should be fixed at linker like a R_BPF_64_ABS32.
+     The difference to ABS32 is that LLVM execution engine does not resolve
+     R_BPF_64_NODYLD32 relocations.  */
   BPF_HOWTO (R_BPF_64_NODYLD32,	/* type */
 	 0,			/* rightshift */
-	 0,			/* size */
-	 0,			/* bitsize */
+	 4,			/* size */
+	 32,			/* bitsize */
 	 false,			/* pc_relative */
 	 0,			/* bitpos */
-	 complain_overflow_dont, /* complain_on_overflow */
+	 complain_overflow_bitfield, /* complain_on_overflow */
 	 bpf_elf_generic_reloc, /* special_function */
 	 "R_BPF_64_NODYLD32",	/* name */
 	 false,			/* partial_inplace */
-	 0,			/* src_mask */
-	 0,			/* dst_mask */
-	 false)			/* pcrel_offset */
+	 0xffffffff,		/* src_mask */
+	 0xffffffff,		/* dst_mask */
+	 true)			/* pcrel_offset */
diff --git a/bfd/elf64-bpf.c b/bfd/elf64-bpf.c
index c932a4024ba..0bffe2c5717 100644
--- a/bfd/elf64-bpf.c
+++ b/bfd/elf64-bpf.c
@@ -276,6 +276,7 @@ bpf_elf_relocate_section (bfd *output_bfd ATTRIBUTE_UNUSED,
           }
 	case R_BPF_64_ABS64:
 	case R_BPF_64_ABS32:
+	case R_BPF_64_NODYLD32:
 	  {
 	    addend = bfd_get (howto->bitsize, input_bfd, where);
 	    relocation += addend;
-- 
2.39.2


[-- Attachment #3: Type: text/plain, Size: 1559 bytes --]



Jose E. Marchesi writes:

> Hi Cuper.
>
>> Hi everyone,
>>
>> This patch is in context of Nick Cliftons request in thread:
>>   https://sourceware.org/pipermail/binutils/2023-October/130194.html
>> due to bug reported in:
>>   https://bugzilla.redhat.com/show_bug.cgi?id=2245296
>>
>> For the time being the linker is not used in BPF infrastructure.
>> Considering that and without a proper way to validate the code, the
>> following patch is what I think is missing in order to add linker
>> support for the particular relocation.
>>
>> Looking forward to your review.
>>
>> Best regards,
>> Cupertino
>>
>> This patch adds linker support to patch R_BPF_64_NODYLD32 relocations.
>> The implementation was based on comments and code in LLVM, as the GNU
>> toolchain does not uses this relocation type.
>> ---
>>  bfd/elf64-bpf.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/bfd/elf64-bpf.c b/bfd/elf64-bpf.c
>> index aefad7da5ac..5820dd3d7d4 100644
>> --- a/bfd/elf64-bpf.c
>> +++ b/bfd/elf64-bpf.c
>> @@ -276,6 +276,7 @@ bpf_elf_relocate_section (bfd *output_bfd ATTRIBUTE_UNUSED,
>>            }
>>  	case R_BPF_64_ABS64:
>>  	case R_BPF_64_ABS32:
>> +	case R_BPF_64_NODYLD32:
>
> I would add here a comment explaining that R_BPF_64_NODYLD32 shall be
> handled/resolved like R_BPF_64_ABS32 and that it differs from ABS32 only
> in that the LLVM execution engine is not supposed to resolve it.
>
> Given that, the patch is OK.
> Thanks!
>
>>  	  {
>>  	    addend = bfd_get (howto->bitsize, input_bfd, where);
>>  	    relocation += addend;