Hi,
On Mon, Aug 18, 2025 at 11:34:18PM -0700, panfan wrote:
> with the previous patch [PATCH v3 6/6] arm64: module: rework module VA
> range selection, the module region can use a full 2GB for large modules.
>
> On arm64 with CONFIG_DYNAMIC_FTRACE, due to the ±128 MB range limit of
> the `BL` instruction, ftrace uses a PLT entry to branch indirectly to
> ftrace_caller when modules may be placed far away.
>
> Currently, this PLT(.text.ftrace_trampoline) resides in MOD_TEXT,
> so call sites in .init.text cannot reach it by `BL` if .init.text
> and .text are allocated in different 128 MB regions.
Ouch; sorry about this. Given the failure mode below, this is a pretty
horrid bug.
> For example, init_moudle in tz_log_dlkm.ko can not reach PLT or
> ftrace_caller by `BL`:
>
> module_direct_base = 0xFFFFFFC07B270000 128M
> module_plt_base = 0xFFFFFFC003270000 2G
>
> mod = 0xFFFFFFC07FF65880 -> (
> state = MODULE_STATE_COMING,
> name = "tz_log_dlkm",
> init = 0xFFFFFFC00370F01C,
>
> mem = (
> (base = 0xFFFFFFC07E7E8000, size = 12288, // MOD_TEXT -- direct
> (base = 0xFFFFFFC07FF65000, size = 12288, // MOD_DATA
> (base = 0xFFFFFFC07FFFB000, size = 12288, // MOD_RODATA
> (base = 0xFFFFFFC07DDA9000, size = 4096, // MOD_RO_AFTER_INIT
> (base = 0xFFFFFFC00370F000, size = 4096, // MOD_INIT_TEXT -- plt
> (base = 0xFFFFFFC003711000, size = 12288, // MOD_INIT_DATA
>
> arch = (
> core = (plt_shndx = 8, plt_num_entries = 0, plt_max_entries = 35),
> init = (plt_shndx = 9, plt_num_entries = 1, plt_max_entries = 1),
> ftrace_trampolines = 0xFFFFFFC07E7EA730 -> ( //
> .text.ftrace_trampoline in MOD_TEXT
>
> PLT in .text.ftrace_trampoline:
> 0xFFFFFFC07E7EA730 adrp x16,0xFFFFFFC080014000
> 0xFFFFFFC07E7EA734 add x16,x16,#0xF64 ; x16,x16,#3940
> 0xFFFFFFC07E7EA738 br x16; ftrace_caller
>
> Here, init_module() in MOD_INIT_TEXT cannot branch to the PLT in MOD_TEXT
> because the offset exceeds 128 MB. As a result,
> ftrace fails to update `nop` to `BL` and inserts `brk #0x100` instead:
>
> 0xFFFFFFC00370F01C init_module: mov x9,x30
> 0xFFFFFFC00370F020 brk #0x100 ; #256
>
> [ 36.290790][ T835] label_imm_common: offset out of range
>
> [ 36.333765][ T835] Kernel text patching generated an invalid instruct
> ion at init_module+0x4/0xfe4 [tz_log_dlkm]!
>
> [ 36.335728][ T835] Call trace:
> [ 36.335735][ T835] init_module+0x4/0xfe4 [tz_log_dlkm]
> [ 36.335750][ T835] do_init_module+0x60/0x2cc
> [ 36.335761][ T835] load_module+0x10e0/0x12ac
> [ 36.335771][ T835] __arm64_sys_finit_module+0x240/0x348
> [ 36.335780][ T835] invoke_syscall+0x60/0x11c
> [ 36.335791][ T835] el0_svc_common+0xb4/0xf0
> [ 36.335801][ T835] do_el0_svc+0x24/0x30
> [ 36.335810][ T835] el0_svc+0x3c/0x74
> [ 36.335821][ T835] el0t_64_sync_handler+0x68/0xbc
> [ 36.335831][ T835] el0t_64_sync+0x1a8/0x1ac
>
> To fix this, introduce an additional `.init.text.ftrace_trampoline`
> section for .init.text. This provides a PLT within MOD_INIT_TEXT, ensuring
> that init functions can branch within range using `BL`. This section
> is freed after do_one_initcall, so there is no persistent cost.
> The core text continues to use the existing PLT in MOD_TEXT.
>
> Signed-off-by: panfan <pan...@qti.qualcomm.com>
Aside from one nit below, I think this is the right fix, though I think
it would be good to simplify/clarify the commit message.
How about:
| On arm64, it has been possible for a module's sections to be placed more
| than 128M away from each other since commit:
|
| 3e35d303ab7d ("arm64: module: rework module VA range selection")
|
| Due to this, an ftrace callsite in a module's .init.text section can be
| out of branch range for the module's ftrace PLT entry (in the module's
| .text section). Any attempt to enable tracing of that callsite will
| result in a BRK being patched into the callsite, resulting in a fatal
| exception when the callsite is later executed.
|
| Fix this by adding an additional trampoline for .init.text, which will
| be within range.
|
| No additional trampolines are necessary due to to the way a given
| module's executable sections are packed together. Any executable
| section beginning with ".init" or ".exit" will be placed in
| MOD_INIT_TEXT, and any other executable section will be placed in
| MOD_TEXT.
|
| Fixes: 3e35d303ab7d ("arm64: module: rework module VA range selection")
[...]
> -static struct plt_entry *get_ftrace_plt(struct module *mod)
> +static struct plt_entry *get_ftrace_plt(struct module *mod, unsigned long
> addr)
> {
> #ifdef CONFIG_MODULES
> - struct plt_entry *plt = mod->arch.ftrace_trampolines;
> + struct plt_entry *plt = NULL;
> +
> + if (within_module_mem_type(addr, mod, MOD_INIT_TEXT))
> + plt = mod->arch.init_ftrace_trampolines;
> + else
> + plt = mod->arch.ftrace_trampolines;
>
> return &plt[FTRACE_PLT_IDX];
It would be good if we could explicitly verify that the region is either
MOD_TEXT or MOD_INIT_TEXT, and if not, return NULL, e.g.
if (within_module_mem_type(addr, mod, MOD_INIT_TEXT))
plt = mod->arch.init_ftrace_trampolines;
else if (within_module_mem_type(addr, mod, MOD_TEXT))
plt = mod->arch.ftrace_trampolines;
else
return NULL;
That last case should never happen, but if it does it'll cause
ftrace_find_callable_addr() to log a warning and return false, such that
its callers can return an error.
Other than that, I think this is fine. With those changes:
Acked-by: Mark Rutland <mark.rutl...@arm.com>
Mark.
> #else
> @@ -332,7 +337,7 @@ static bool ftrace_find_callable_addr(struct dyn_ftrace
> *rec,
> if (WARN_ON(!mod))
> return false;
>
> - plt = get_ftrace_plt(mod);
> + plt = get_ftrace_plt(mod, pc);
> if (!plt) {
> pr_err("ftrace: no module PLT for %ps\n", (void *)*addr);
> return false;
> diff --git a/arch/arm64/kernel/module-plts.c b/arch/arm64/kernel/module-plts.c
> index bde32979c06a..7afd370da9f4 100644
> --- a/arch/arm64/kernel/module-plts.c
> +++ b/arch/arm64/kernel/module-plts.c
> @@ -283,7 +283,7 @@ int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr
> *sechdrs,
> unsigned long core_plts = 0;
> unsigned long init_plts = 0;
> Elf64_Sym *syms = NULL;
> - Elf_Shdr *pltsec, *tramp = NULL;
> + Elf_Shdr *pltsec, *tramp = NULL, *init_tramp = NULL;
> int i;
>
> /*
> @@ -298,6 +298,9 @@ int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr
> *sechdrs,
> else if (!strcmp(secstrings + sechdrs[i].sh_name,
> ".text.ftrace_trampoline"))
> tramp = sechdrs + i;
> + else if (!strcmp(secstrings + sechdrs[i].sh_name,
> + ".init.text.ftrace_trampoline"))
> + init_tramp = sechdrs + i;
> else if (sechdrs[i].sh_type == SHT_SYMTAB)
> syms = (Elf64_Sym *)sechdrs[i].sh_addr;
> }
> @@ -363,5 +366,12 @@ int module_frob_arch_sections(Elf_Ehdr *ehdr, Elf_Shdr
> *sechdrs,
> tramp->sh_size = NR_FTRACE_PLTS * sizeof(struct plt_entry);
> }
>
> + if (init_tramp) {
> + init_tramp->sh_type = SHT_NOBITS;
> + init_tramp->sh_flags = SHF_EXECINSTR | SHF_ALLOC;
> + init_tramp->sh_addralign = __alignof__(struct plt_entry);
> + init_tramp->sh_size = NR_FTRACE_PLTS * sizeof(struct plt_entry);
> + }
> +
> return 0;
> }
> diff --git a/arch/arm64/kernel/module.c b/arch/arm64/kernel/module.c
> index 40148d2725ce..d6d443c4a01a 100644
> --- a/arch/arm64/kernel/module.c
> +++ b/arch/arm64/kernel/module.c
> @@ -466,6 +466,17 @@ static int module_init_ftrace_plt(const Elf_Ehdr *hdr,
> __init_plt(&plts[FTRACE_PLT_IDX], FTRACE_ADDR);
>
> mod->arch.ftrace_trampolines = plts;
> +
> + s = find_section(hdr, sechdrs, ".init.text.ftrace_trampoline");
> + if (!s)
> + return -ENOEXEC;
> +
> + plts = (void *)s->sh_addr;
> +
> + __init_plt(&plts[FTRACE_PLT_IDX], FTRACE_ADDR);
> +
> + mod->arch.init_ftrace_trampolines = plts;
> +
> #endif
> return 0;
> }
> --
> 2.34.1
>
