Re: [PATCH dwarves v3 1/9] dwarf_loader: Reduce parameter checking with clang DW_AT_calling_convention attr

[Yonghong Song <yonghong.song@linux.dev>]

On 3/23/26 5:56 AM, Alan Maguire wrote:
> On 22/03/2026 17:36, Yonghong Song wrote:
>>
>> On 3/21/26 4:10 PM, Jiri Olsa wrote:
>>> On Fri, Mar 20, 2026 at 12:09:22PM -0700, Yonghong Song wrote:
>>>> Currently every function is checked for its parameters to identify whether
>>>> the signature changed or not. If signature indeed changed, pahole may do
>>>> some adjustment for parameters for true signatures.
>>>>
>>>> In clang, any function with the following attribute
>>>>         DW_AT_calling_convention        (DW_CC_nocall)
>>>> indicates this function having signature changed.
>>>> pahole can take advantage of this to avoid parameter checking if
>>>> DW_AT_calling_convention is not DW_CC_nocall.
>>>>
>>>> But more importantly, DW_CC_nocall can identify signature-changed functions
>>>> and parameters can be checked one-after-another to create the true
>>>> signatures. Otherwise, it takes more effort to identify whether a
>>>> function has signature changed or not. For example, for funciton
>>>>     __bpf_kfunc static void bbr_main(struct sock *sk, u32 ack, int flag,
>>>>        const struct rate_sample *rs) { ... }
>>>> and bbr_main() is a callback function in
>>>>     .cong_control   = bbr_main
>>>> in 'struct tcp_congestion_ops tcp_bbr_cong_ops'.
>>>> In the above bbr_main(...), parameter 'ack' and 'flag' are not used.
>>>> The following are some details:
>>>>
>>>> 0x0a713b8d:     DW_TAG_formal_parameter
>>>>                     DW_AT_location        (indexed (0x28) loclist = 0x0166d452:
>>>>                        [0xffffffff83e77fd9, 0xffffffff83e78016): DW_OP_reg5 RDI
>>>>                        ...
>>>>                     DW_AT_name    ("sk")
>>>>                     DW_AT_type    (0x0a6f5b2b "sock *")
>>>>                     ...
>>>>
>>>> 0x0a713b98:     DW_TAG_formal_parameter
>>>>                     DW_AT_name    ("ack")
>>>>                     DW_AT_type    (0x0a6f58fd "u32")
>>>>                     ...
>>>>
>>>> 0x0a713ba2:     DW_TAG_formal_parameter
>>>>                     DW_AT_name    ("flag")
>>>>                     DW_AT_type    (0x0a6f57d1 "int")
>>>>                     ...
>>>>
>>>> 0x0a713bac:     DW_TAG_formal_parameter
>>>>                     DW_AT_location        (indexed (0x29) loclist = 0x0166d4a8:
>>>>                        [0xffffffff83e77fd9, 0xffffffff83e78016): DW_OP_reg2 RCX
>>>>                        ...
>>>>                     DW_AT_name    ("rs")
>>>>                     DW_AT_type    (0x0a710da5 "const rate_sample *")
>>>>                     DW_AT_decl_line       (1027)
>>>>
>>>> Some analysis for the above dwarf can conclude that the 'ark' and 'flag'
>>>> may be related to RSI and RDX, considering the last one is RCX. Basically this
>>>> requires all parameters are available to collectively decide whether the
>>>> true signature can be found or not. In such case, DW_CC_nocall can make things
>>>> easier as parameter can be checked one after another.
>>>>
>>>> For a clang built bpf-next kernel, in non-LTO setup, the number of kernel functions
>>>> is 69103 and the number of signature changed functions is 875, based on
>>>>         DW_AT_calling_convention        (DW_CC_nocall)
>>>> indication.
>>> we don't display these stats, right? would be great and probably easy
>>> to count different flags and aggregate them
>> The actual stats will be different depending on kernel code and compiler optimization.
>> But I think I can still count them for each commit so people will know
>> which change caused how much true signature.
>>
> I'll send a patch shortly that does kernel build/compare functions for LLVM/clang [1];
> currently our pahole CI only covers gcc kernel builds. That will show differences
> between pahole next and the current changes in github; that workflow has been handy for
> seeing the effects of pahole changes on the set of vmlinux functions.
>
> We should also look at extending pfunct to do more stats; I've got a few patches which
> get it to support split BTF and add a manual page. It has some basic stats if I
> remember correctly, but it would be nice to extend those at some stage.
>
> [1] https://github.com/alan-maguire/dwarves/actions/runs/23435796365/job/68173414313

Thanks for adding llvm test support in github. This definitely helpful.
I will try to have a few more coverages from llvm side.

>   
>>>> Among 875 signature changed functions, after this patch, 495 functions
>>>> can have proper true signatures, mostly due to simple dead argument
>>>> elimination. The number of remaining functions, which cannot get the
>>>> true signature, is 379. They will be addressed in the subsequent commits.
>>> after this change I have more functions added (below) and I checked
>>> few and they seem ok (w/o DW_CC_nocall)
>> Most global functions should be okay. But some global functions
>> esp. if argument is more than 8 byte (struct type), they may be
>> considered as signature change with the current implementation.
>>
>>> jirka
>>>
>>>
>>> acpi_ex_do_debug_object
>>> add_bits
>>> amd_pstate_get_mode_string
>>> dma_alloc_from_contiguous
>>> do_set_pmd
>>> dst_destroy
>>> find_cpio_data
>>> find_microcode_in_initrd
>>> ima_write_policy
>>> insn_init
>>> io_tx_ubuf_complete
>>> ip6_protocol_deliver_rcu
>>> kvm_tdp_mmu_unmap_gfn_range
>>> net_dim_get_def_rx_moderation
>>> net_dim_get_def_tx_moderation
>>> pinctrl_commit_state
>>> string_unescape
>>> syscall_copy_user_array
>>> tdp_mmu_zap_leafs
>>> usb_speed_string
>>> __vlan_find_dev_deep_rcu
>>> __vxlan_fdb_delete
>>> xfs_bmbt_maxrecs
>>> xfs_inobt_maxrecs
>>> ZSTD_buildSequencesStatistics
>>> ZSTD_copyCCtx
>>> ZSTD_findFrameSizeInfo
>>> ZSTD_get1BlockSummary
>>> zstd_get_cparams
>>> ZSTD_getCParams
>>> ZSTD_getCParamsFromCCtxParams
>>> ZSTD_getCParamsFromCDict
>>> ZSTD_getCParams_internal
>>> zstd_get_params
>>> ZSTD_getParams