| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Wed, 17 Jun 2020 19:06:25 -0700
From: Saravana Kannan <saravanak@...gle.com>
To: Will Deacon <will@...nel.org>
Cc: Ard Biesheuvel <ardb@...nel.org>,
Catalin Marinas <catalin.marinas@....com>,
Android Kernel Team <kernel-team@...roid.com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Linux ARM <linux-arm-kernel@...ts.infradead.org>,
Ard Biesheuvel <ard.biesheuvel@...aro.org>
Subject: Re: [PATCH v1] arm64/module: Optimize module load time by optimizing
PLT counting
On Wed, Jun 17, 2020 at 7:05 AM Will Deacon <will@...nel.org> wrote:
>
> Hi all,
>
> On Wed, Jun 17, 2020 at 10:17:33AM +0200, Ard Biesheuvel wrote:
> > On Tue, 16 Jun 2020 at 23:40, Will Deacon <will@...nel.org> wrote:
> > > On Fri, Jun 05, 2020 at 03:22:57PM -0700, Saravana Kannan wrote:
> > > > This gives significant reduction in module load time for modules with
> > > > large number of relocations with no measurable impact on modules with a
> > > > small number of relocations. In my test setup with CONFIG_RANDOMIZE_BASE
> > > > enabled, the load time for one module went down from 268ms to 100ms.
> > > > Another module went down from 143ms to 83ms.
> > >
> > > Whilst I can see that's a significant relative saving, what proportion of
> > > actual boot time are we talking about here? It would be interesting to
> > > know if there are bigger potential savings elsewhere.
> > >
> >
> > Also, 'some module' vs 'some other module' doesn't really say
> > anything. Please explain which modules and their sizes.
>
> I suspect these are all out-of-tree modules, but yes, some metadata such as
> sizes, nr or relocs etc would be good to have in the commit message.
>
> > > > diff --git a/arch/arm64/kernel/module-plts.c b/arch/arm64/kernel/module-plts.c
> > > > index 65b08a74aec6..bf5118b3b828 100644
> > > > --- a/arch/arm64/kernel/module-plts.c
> > > > +++ b/arch/arm64/kernel/module-plts.c
> > > > @@ -253,6 +253,36 @@ static unsigned int count_plts(Elf64_Sym *syms, Elf64_Rela *rela, int num,
> > > > return ret;
> > > > }
> > > >
> > > > +static bool rela_needs_dedup(Elf64_Rela *rela)
> > > > +{
> > > > + return ELF64_R_TYPE(rela->r_info) == R_AARCH64_JUMP26
> > > > + || ELF64_R_TYPE(rela->r_info) == R_AARCH64_CALL26;
> > > > +}
> > >
> >
> > Would it help to check the section index here as well? Call/jump
> > instructions within a section are never sent through a PLT entry.
>
> (I tried hacking this in below)
>
> > > Does this handle A53 erratum 843419 correctly? I'm worried that we skip
> > > the ADRP PLTs there.
> > >
> >
> > ADRP PLTs cannot be deduplicated, as they incorporate a relative jump
> > back to the caller.
>
> Duh yes, thanks. We can't trash the link register here.
>
> > > > +/* Group the CALL26/JUMP26 relas toward the beginning of the array. */
> > > > +static int partition_dedup_relas(Elf64_Rela *rela, int numrels)
> > > > +{
> > > > + int i = 0, j = numrels - 1;
> > > > + Elf64_Rela t;
> > > > +
> > > > + while (i < j) {
> > > > + while (rela_needs_dedup(rela + i) && i < j)
> > > > + i++;
> > > > + while (!rela_needs_dedup(rela + j) && i < j)
> > > > + j--;
> > > > + if (i < j) {
> > > > + t = *(rela + j);
> > > > + *(rela + j) = *(rela + i);
> > > > + *(rela + i) = t;
> > > > + }
> > > > + }
> > >
> > > This is very hard to read and I think some of the 'i < j' comparisons are
> > > redundant. Would it make more sense to assign a temporary rather than
> > > post-inc/decrement and recheck?
> > >
> >
> > Agreed.
> >
> > Also, what's wrong with [] array indexing?
>
> Saravana, since our stylistic objections are reasonably vague, I tried
> to clean this up so you can get an idea of how I'd prefer it to look (can't
> speak for Ard). I haven't tried running this, but please feel free to adapt
> it. Replacement diff below.
>
> Will
>
> --->8
>
> diff --git a/arch/arm64/kernel/module-plts.c b/arch/arm64/kernel/module-plts.c
> index 65b08a74aec6..204290314054 100644
> --- a/arch/arm64/kernel/module-plts.c
> +++ b/arch/arm64/kernel/module-plts.c
> @@ -253,6 +253,39 @@ static unsigned int count_plts(Elf64_Sym *syms, Elf64_Rela *rela, int num,
> return ret;
> }
>
> +static bool branch_rela_needs_plt(Elf64_Sym *syms, Elf64_Rela *rela,
> + Elf64_Word dstidx)
> +{
> +
> + Elf64_Sym *s = syms + ELF64_R_SYM(rela->r_info);
> +
> + if (s->st_shndx == dstidx)
> + return false;
> +
> + return ELF64_R_TYPE(rela->r_info) == R_AARCH64_JUMP26 ||
> + ELF64_R_TYPE(rela->r_info) == R_AARCH64_CALL26;
> +}
> +
> +static int partition_branch_plt_relas(Elf64_Sym *syms, Elf64_Rela *rela,
> + int numrels, Elf64_Word dstidx)
> +{
> + int i = 0, j = numrels - 1;
> +
> + if (!IS_ENABLED(CONFIG_RANDOMIZE_BASE))
> + return 0;
> +
> + while (i < j) {
> + if (branch_rela_needs_plt(syms, &rela[i], dstidx))
> + i++;
> + else if (branch_rela_needs_plt(syms, &rela[j], dstidx))
> + swap(rela[i], rela[j]);
> + else
> + j--;
> + }
> +
> + return i;
> +}
> +
> int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
> char *secstrings, struct module *mod)
> {
> @@ -290,7 +323,7 @@ int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
>
> for (i = 0; i < ehdr->e_shnum; i++) {
> Elf64_Rela *rels = (void *)ehdr + sechdrs[i].sh_offset;
> - int numrels = sechdrs[i].sh_size / sizeof(Elf64_Rela);
> + int nents, numrels = sechdrs[i].sh_size / sizeof(Elf64_Rela);
> Elf64_Shdr *dstsec = sechdrs + sechdrs[i].sh_info;
>
> if (sechdrs[i].sh_type != SHT_RELA)
> @@ -300,8 +333,14 @@ int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
> if (!(dstsec->sh_flags & SHF_EXECINSTR))
> continue;
>
> - /* sort by type, symbol index and addend */
> - sort(rels, numrels, sizeof(Elf64_Rela), cmp_rela, NULL);
> + /*
> + * sort branch relocations requiring a PLT by type, symbol index
> + * and addend
> + */
> + nents = partition_branch_plt_relas(syms, rels, numrels,
> + sechdrs[i].sh_info);
> + if (nents)
> + sort(rels, nents, sizeof(Elf64_Rela), cmp_rela, NULL);
>
> if (!str_has_prefix(secstrings + dstsec->sh_name, ".init"))
> core_plts += count_plts(syms, rels, numrels,
Thanks Will & Ard. I'll incorporate your feedback and send a v2 within
a few days.
-Saravana
Powered by blists - more mailing lists