public inbox for binutils@sourceware.org
 help / color / mirror / Atom feed
From: mengqinggang <mengqinggang@loongson.cn>
To: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Cc: binutils@sourceware.org, Lulu Cai <cailulu@loongson.cn>,
	chenglulu@loongson.cn, hejinyang@loongson.cn,
	i.swmail@xen0n.name, liuzhensong@loongson.cn,
	luweining@loongson.cn, Fangrui Song <maskray@google.com>,
	nickc@redhat.com, wanglei@loongson.cn, xry111@xry111.site,
	xuchenghua@loongson.cn
Subject: Re: [PATCH] LoongArch: Update comment about bottom bit usage in TLS GOT construction
Date: Fri, 29 Dec 2023 17:59:04 +0800	[thread overview]
Message-ID: <6ef9bbd3-8bd6-44b1-22ca-757c4108f883@loongson.cn> (raw)
In-Reply-To: <2E548645-1BD7-4B21-B1D0-D5961A11FC0D@gmail.com>

These comments mainly explain the issue of another implementation method.


If the code like below, TLS IE and GD processed separately, we need two 
bits as flags.

case  R_LARCH_TLS_IE_PC_HI20:
     if ((got_off & 1) == 0)
         {
             ...
         }

case R_LARCH_TLS_GD_PC_HI20:
     if ((got_off & 2) == 0)
         {
             ...
         }


在 2023/12/29 下午5:11, Tatsuyuki Ishi 写道:
>> On Dec 29, 2023, at 17:50, mengqinggang <mengqinggang@loongson.cn> wrote:
>>
>> At the beginning of implementation, we try to IE and GD only generate their own GOT entry,
>> so we need two bits to do the flags.
> I checked the tree back at the date the comment was introduced (6d13722a: "bfd: Add supported for LoongArch new relocations.”). Even at that time, there was no |= 2 accesses, so if my understanding is correct, this comment was wrong from the beginning.
>
> Could you clarify what you mean here, in particular what “At the beginning of implementation” refer to?
>
>> 在 2023/12/28 下午10:58, Tatsuyuki Ishi 写道:
>>> The GOT assignment logic, likely copied from another backend, does not in
>>> fact require multiple flags to construct multiple type of slots. Instead,
>>> all slots are initialized on the first relocation encountered. Update
>>> comment to avoid confusion.
>>> ---
>>>   bfd/elfnn-loongarch.c | 17 +++++++----------
>>>   1 file changed, 7 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/bfd/elfnn-loongarch.c b/bfd/elfnn-loongarch.c
>>> index 64c34e99261..d66dcee1100 100644
>>> --- a/bfd/elfnn-loongarch.c
>>> +++ b/bfd/elfnn-loongarch.c
>>> @@ -3656,16 +3656,13 @@ loongarch_elf_relocate_section (bfd *output_bfd, struct bfd_link_info *info,
>>>   	  relocation -= elf_hash_table (info)->tls_sec->vma;
>>>   	  break;
>>>   -	/* TLS IE LD/GD process separately is troublesome.
>>> -	   When a symbol is both ie and LD/GD, h->got.off |= 1
>>> -	   make only one type be relocated.  We must use
>>> -	   h->got.offset |= 1 and h->got.offset |= 2
>>> -	   diff IE and LD/GD.  And all (got_off & (~(bfd_vma)1))
>>> -	   (IE LD/GD and reusable GOT reloc) must change to
>>> -	   (got_off & (~(bfd_vma)3)), beause we use lowest 2 bits
>>> -	   as a tag.
>>> -	   Now, LD and GD is both GOT_TLS_GD type, LD seems to
>>> -	   can be omitted.  */
>>> +	/* The bottom bit of h->got.offset is (ab)used as a flag.
>>> +	   Upon encountering the first TLS relocation, we initialize all GOT
>>> +	   slots for the corresponding symbol and set the bottom bit to 1.
>>> +
>>> +	   The second and subsequent relocations will check the flag, see that
>>> +	   the slot is already initialized, and move on to just relocating the
>>> +	   entry.  */
>>>   	case R_LARCH_TLS_IE_PC_HI20:
>>>   	case R_LARCH_TLS_IE_HI20:
>>>   	case R_LARCH_TLS_LD_PC_HI20:


  reply	other threads:[~2023-12-29  9:59 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-12-28 14:58 [PATCH] LoongArch: Do not add DF_STATIC_TLS for TLS LE Tatsuyuki Ishi
2023-12-28 14:58 ` [PATCH] LoongArch: Use tab to indent assembly in TLSDESC test suite Tatsuyuki Ishi
2024-01-22  9:47   ` mengqinggang
2023-12-28 14:58 ` [PATCH] LoongArch: Update comment about bottom bit usage in TLS GOT construction Tatsuyuki Ishi
2023-12-29  8:50   ` mengqinggang
2023-12-29  9:11     ` Tatsuyuki Ishi
2023-12-29  9:59       ` mengqinggang [this message]
2024-01-22  5:45 ` [PATCH] LoongArch: Do not add DF_STATIC_TLS for TLS LE Tatsuyuki Ishi
2024-01-22  5:48   ` Fangrui Song
2024-01-22  6:51     ` Xi Ruoyao
2024-01-22  7:11       ` Fangrui Song
2024-01-22  8:27   ` mengqinggang
2024-01-22  9:46 ` mengqinggang

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6ef9bbd3-8bd6-44b1-22ca-757c4108f883@loongson.cn \
    --to=mengqinggang@loongson.cn \
    --cc=binutils@sourceware.org \
    --cc=cailulu@loongson.cn \
    --cc=chenglulu@loongson.cn \
    --cc=hejinyang@loongson.cn \
    --cc=i.swmail@xen0n.name \
    --cc=ishitatsuyuki@gmail.com \
    --cc=liuzhensong@loongson.cn \
    --cc=luweining@loongson.cn \
    --cc=maskray@google.com \
    --cc=nickc@redhat.com \
    --cc=wanglei@loongson.cn \
    --cc=xry111@xry111.site \
    --cc=xuchenghua@loongson.cn \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).