[PATCH 1/5 v1] LoongArch: bfd: Add new reloc types.

liuzhensong liuzhensong@loongson.cn
Wed Jul 20 12:54:05 GMT 2022


在 2022/7/20 下午12:03, WANG Xuerui 写道:
> On 2022/7/20 10:07, liuzhensong wrote:
>>
>> Some relocation name changes make sense. But, how to name a name is a 
>> matter of personal aesthetics, and it is difficult to satisfy 
>> everyone. Does the name of the relocation matter? should we pay more 
>> attention to the specific implementation?  Remove useless and add 
>> needed?
>>
> Names do matter. You can't expect to lower maintenance costs and 
> barrier to entry for newcomers with "bad" names. I showed the original 
> names in the patch to some of my colleagues (with no low-level 
> programming experience; they are all application developers), and they 
> invariably can't get the distinction of "LO12/HI20/LO20/HI12". I then 
> showed them my proposal, granted they needed a little explanation as 
> to what things like "SK12" mean, but after that they can immediately 
> get the feel of the relocs ("ah, it's like for this purpose, put this 
> part of imm into this slot").
>
> LoongArch is expected to remain niche, at least for the coming 3~5 
> years, so you can't expect to have many people working on low-level 
> nitty-gritty details like this. I personally don't think your team's 
> composition is going to remain stable over the years, either; in 
> either case, people come and go, so IMO there is indeed value in this 
> bikeshedding. The community as a whole can work over concrete 
> proposals of better naming, and I think my proposal serves as such a 
> constructive starting point.
>
> Also, if you think the current names are more appropriate, with 
> objective reasons, then *please communicate* your justification and 
> thought behind them. We "outsiders" actually intend to cooperate, and 
> have voiced our thoughts, but we don't know what you and your team 
> think. For now all we have is this "LO12/HI20/LO20/HI12" and outsiders 
> can draw no conclusion other than "hmm this is going to cause 
> confusion". We really need to communicate more in the open for pushing 
> things forward, constructively.

Your relocation is fine, but it's a bit long and complicated and 
informative(personaly opinion). It seems unlikely to modify code just by 
relocation name, there is necessary to understand the instructions 
represented by relocation.

I made some adjustments to some relocations, then, LO12 and HI20 
implement 32-bit relocation, xxx64_LO20 and xxx64_HI12 implement high 
32bits of 64-bit relocation while low 32bits use LO12 and HI20. In my 
opinion, it is enough to know what to do by name.

LO12 [11:0] HI20 [31:12] 64_LO20 [51:32] 64_HI12 [63:52] Changes,

1. B_R_L_GOT64_LO12          B_R_L_GOT_LO12
2. B_R_L_GOT64_HI20          B_R_L_GOT_HI20
3. B_R_L_TLS64_IE_LO12       B_R_L_TLS_IE_LO12
4. B_R_L_TLS64_IE_HI20       B_R_L_TLS_IE_HI20
5. B_R_L_TLS64_LD_HI20       B_R_L_TLS_LD_HI20
6. B_R_L_TLS64_GD_HI20       B_R_L_TLS_GD_HI20

  

    Original name           Suggested name
    -------------           --------------
    B_R_L_B16               B_R_L_PCREL_SK16
    B_R_L_B21               B_R_L_PCREL_SD5K16
    B_R_L_B26               B_R_L_PCREL_SD10K16
    B_R_L_ABS_LO12          B_R_L_ABS_0_SK12
    B_R_L_ABS_HI20          B_R_L_ABS_12_SJ20
    B_R_L_ABS64_LO20        B_R_L_ABS_32_SJ20
    B_R_L_ABS64_HI12        B_R_L_ABS_52_SK12
    B_R_L_PCALA_LO12        B_R_L_PCALA_0_SK12
    B_R_L_PCALA_HI20        B_R_L_PCALA_12_SJ20
    B_R_L_PCALA64_LO20      B_R_L_PCALA_32_SJ20
    B_R_L_PCALA64_HI12      B_R_L_PCALA_52_SK12
    B_R_L_GOT_PC_LO12       B_R_L_GOT_PCALA_0_SK12
    B_R_L_GOT_PC_HI20       B_R_L_GOT_PCALA_12_SJ20
    B_R_L_GOT64_PC_LO20     B_R_L_GOT_PCALA_32_SJ20
    B_R_L_GOT64_PC_HI12     B_R_L_GOT_PCALA_52_SK12
1. B_R_L_GOT_LO12          B_R_L_GOT_ABS_0_SK12
2. B_R_L_GOT_HI20          B_R_L_GOT_ABS_12_SJ20
    B_R_L_GOT64_LO20        B_R_L_GOT_ABS_32_SJ20
    B_R_L_GOT64_HI12        B_R_L_GOT_ABS_52_SK12
    B_R_L_TLS_LE_LO12       B_R_L_TLS_LE_ABS_0_SK12
    B_R_L_TLS_LE_HI20       B_R_L_TLS_LE_ABS_12_SJ20
    B_R_L_TLS_LE64_LO20     B_R_L_TLS_LE_ABS_32_SJ20
    B_R_L_TLS_LE64_HI12     B_R_L_TLS_LE_ABS_52_SK12
    B_R_L_TLS_IE_PC_LO12    B_R_L_TLS_IE_PCALA_0_SK12
    B_R_L_TLS_IE_PC_HI20    B_R_L_TLS_IE_PCALA_12_SJ20
    B_R_L_TLS_IE64_PC_LO20  B_R_L_TLS_IE_PCALA_32_SJ20
    B_R_L_TLS_IE64_PC_HI12  B_R_L_TLS_IE_PCALA_52_SK12
3. B_R_L_TLS_IE_LO12       B_R_L_TLS_IE_ABS_0_SK12
4. B_R_L_TLS_IE_HI20       B_R_L_TLS_IE_ABS_12_SJ20
    B_R_L_TLS_IE64_LO20     B_R_L_TLS_IE_ABS_32_SJ20
    B_R_L_TLS_IE64_HI12     B_R_L_TLS_IE_ABS_52_SK12
    B_R_L_TLS_LD_PC_HI20    B_R_L_TLS_LD_PCALA_12_SJ20
5. B_R_L_TLS_LD_HI20       B_R_L_TLS_LD_ABS_12_SJ20
    B_R_L_TLS_GD_PC_HI20    B_R_L_TLS_GD_PCALA_12_SJ20
6. B_R_L_TLS_GD_HI20       B_R_L_TLS_GD_ABS_12_SJ20


  



More information about the Binutils mailing list