linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "liuqi (BA)" <liuqi115@huawei.com>
To: "Song Bao Hua (Barry Song)" <song.bao.hua@hisilicon.com>,
	Masami Hiramatsu <mhiramat@kernel.org>,
	Linuxarm <linuxarm@huawei.com>
Cc: "catalin.marinas@arm.com" <catalin.marinas@arm.com>,
	"will@kernel.org" <will@kernel.org>,
	"naveen.n.rao@linux.ibm.com" <naveen.n.rao@linux.ibm.com>,
	"anil.s.keshavamurthy@intel.com" <anil.s.keshavamurthy@intel.com>,
	"davem@davemloft.net" <davem@davemloft.net>,
	"linux-arm-kernel@lists.infradead.org" 
	<linux-arm-kernel@lists.infradead.org>,
	"Zengtao (B)" <prime.zeng@hisilicon.com>,
	"robin.murphy@arm.com" <robin.murphy@arm.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] arm64: kprobe: Enable OPTPROBE for arm64
Date: Mon, 2 Aug 2021 11:52:00 +0800	[thread overview]
Message-ID: <2f32fff3-6b58-583f-8e85-06ec1553d3f4@huawei.com> (raw)
In-Reply-To: <6a97dff6c33c4b84887223de2502bd3d@hisilicon.com>



On 2021/7/31 20:21, Song Bao Hua (Barry Song) wrote:
> 
> 
>> -----Original Message-----
>> From: Masami Hiramatsu [mailto:mhiramat@kernel.org]
>> Sent: Saturday, July 31, 2021 1:16 PM
>> To: Song Bao Hua (Barry Song) <song.bao.hua@hisilicon.com>
>> Cc: liuqi (BA) <liuqi115@huawei.com>; catalin.marinas@arm.com;
>> will@kernel.org; naveen.n.rao@linux.ibm.com; anil.s.keshavamurthy@intel.com;
>> davem@davemloft.net; linux-arm-kernel@lists.infradead.org; Zengtao (B)
>> <prime.zeng@hisilicon.com>; robin.murphy@arm.com; Linuxarm
>> <linuxarm@huawei.com>; linux-kernel@vger.kernel.org
>> Subject: Re: [PATCH] arm64: kprobe: Enable OPTPROBE for arm64
>>
>> On Fri, 30 Jul 2021 10:04:06 +0000
>> "Song Bao Hua (Barry Song)" <song.bao.hua@hisilicon.com> wrote:
>>
>>>>>>>
>>>>>>> Hi Qi,
>>>>>>>
>>>>>>> Thanks for your effort!
>>>>>>>
>>>>>>> On Mon, 19 Jul 2021 20:24:17 +0800
>>>>>>> Qi Liu <liuqi115@huawei.com> wrote:
>>>>>>>
>>>>>>>> This patch introduce optprobe for ARM64. In optprobe, probed
>>>>>>>> instruction is replaced by a branch instruction to detour
>>>>>>>> buffer. Detour buffer contains trampoline code and a call to
>>>>>>>> optimized_callback(). optimized_callback() calls opt_pre_handler()
>>>>>>>> to execute kprobe handler.
>>>>>>>
>>>>>>> OK so this will replace only one instruction.
>>>>>>>
>>>>>>>>
>>>>>>>> Limitations:
>>>>>>>> - We only support !CONFIG_RANDOMIZE_MODULE_REGION_FULL case to
>>>>>>>> guarantee the offset between probe point and kprobe pre_handler
>>>>>>>> is not larger than 128MiB.
>>>>>>>
>>>>>>> Hmm, shouldn't we depends on !CONFIG_ARM64_MODULE_PLTS? Or,
>>>>>>> allocate an intermediate trampoline area similar to arm optprobe
>>>>>>> does.
>>>>>>
>>>>>> Depending on !CONFIG_ARM64_MODULE_PLTS will totally disable
>>>>>> RANDOMIZE_BASE according to arch/arm64/Kconfig:
>>>>>> config RANDOMIZE_BASE
>>>>>> 	bool "Randomize the address of the kernel image"
>>>>>> 	select ARM64_MODULE_PLTS if MODULES
>>>>>> 	select RELOCATABLE
>>>>>
>>>>> Yes, but why it is required for "RANDOMIZE_BASE"?
>>>>> Does that imply the module call might need to use PLT in
>>>>> some cases?
>>>>>
>>>>>>
>>>>>> Depending on !RANDOMIZE_MODULE_REGION_FULL seems to be still
>>>>>> allowing RANDOMIZE_BASE via avoiding long jump according to:
>>>>>> arch/arm64/Kconfig:
>>>>>>
>>>>>> config RANDOMIZE_MODULE_REGION_FULL
>>>>>> 	bool "Randomize the module region over a 4 GB range"
>>>>>> 	depends on RANDOMIZE_BASE
>>>>>> 	default y
>>>>>> 	help
>>>>>> 	  Randomizes the location of the module region inside a 4 GB window
>>>>>> 	  covering the core kernel. This way, it is less likely for modules
>>>>>> 	  to leak information about the location of core kernel data structures
>>>>>> 	  but it does imply that function calls between modules and the core
>>>>>> 	  kernel will need to be resolved via veneers in the module PLT.
>>>>>>
>>>>>> 	  When this option is not set, the module region will be randomized
>> over
>>>>>> 	  a limited range that contains the [_stext, _etext] interval of the
>>>>>> 	  core kernel, so branch relocations are always in range.
>>>>>
>>>>> Hmm, this dependency looks strange. If it always in range, don't we need
>>>>> PLT for modules?
>>>>>
>>>>> Cataline, would you know why?
>>>>> Maybe it's a KASLR's Kconfig issue?
>>>>
>>>> I actually didn't see any problem after making this change:
>>>>
>>>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>>>> index e07e7de9ac49..6440671b72e0 100644
>>>> --- a/arch/arm64/Kconfig
>>>> +++ b/arch/arm64/Kconfig
>>>> @@ -1781,7 +1781,6 @@ config RELOCATABLE
>>>>
>>>>   config RANDOMIZE_BASE
>>>>          bool "Randomize the address of the kernel image"
>>>> -       select ARM64_MODULE_PLTS if MODULES
>>>>          select RELOCATABLE
>>>>          help
>>>>            Randomizes the virtual address at which the kernel image is
>>>> @@ -1801,6 +1800,7 @@ config RANDOMIZE_BASE
>>>>   config RANDOMIZE_MODULE_REGION_FULL
>>>>          bool "Randomize the module region over a 4 GB range"
>>>>          depends on RANDOMIZE_BASE
>>>> +       select ARM64_MODULE_PLTS if MODULES
>>>>          default y
>>>>          help
>>>>            Randomizes the location of the module region inside a 4 GB window
>>>>
>>>> and having this config:
>>>> # zcat /proc/config.gz | grep RANDOMIZE_BASE
>>>> CONFIG_RANDOMIZE_BASE=y
>>>>
>>>> # zcat /proc/config.gz | grep RANDOMIZE_MODULE_REGION_FULL
>>>> # CONFIG_RANDOMIZE_MODULE_REGION_FULL is not set
>>>>
>>>> # zcat /proc/config.gz | grep ARM64_MODULE_PLTS
>>>> # CONFIG_ARM64_MODULE_PLTS is not set
>>>>
>>>> Modules work all good:
>>>> # lsmod
>>>> Module                  Size  Used by
>>>> btrfs                1355776  0
>>>> blake2b_generic        20480  0
>>>> libcrc32c              16384  1 btrfs
>>>> xor                    20480  1 btrfs
>>>> xor_neon               16384  1 xor
>>>> zstd_compress         163840  1 btrfs
>>>> raid6_pq              110592  1 btrfs
>>>> ctr                    16384  0
>>>> md5                    16384  0
>>>> ip_tunnel              32768  0
>>>> ipv6                  442368  28
>>>>
>>>>
>>>> I am not quite sure if there is a corner case. If no,
>>>> I would think the kconfig might be some improper.
>>>
>>> The corner case is that even CONFIG_RANDOMIZE_MODULE_REGION_FULL
>>> is not enabled, but if CONFIG_ARM64_MODULE_PLTS is enabled, when
>>> we can't get memory from the 128MB area in case the area is exhausted,
>>> we will fall back in module_alloc() to a 2GB area as long as either
>>> of the below two conditions is met:
>>>
>>> 1. KASAN is not enabled
>>> 2. KASAN is enabled and CONFIG_KASAN_VMALLOC is also enabled.
>>>
>>> void *module_alloc(unsigned long size)
>>> {
>>> 	u64 module_alloc_end = module_alloc_base + MODULES_VSIZE;
>>> 	gfp_t gfp_mask = GFP_KERNEL;
>>> 	void *p;
>>>
>>> 	/* Silence the initial allocation */
>>> 	if (IS_ENABLED(CONFIG_ARM64_MODULE_PLTS))
>>> 		gfp_mask |= __GFP_NOWARN;
>>>
>>> 	if (IS_ENABLED(CONFIG_KASAN_GENERIC) ||
>>> 	    IS_ENABLED(CONFIG_KASAN_SW_TAGS))
>>> 		/* don't exceed the static module region - see below */
>>> 		module_alloc_end = MODULES_END;
>>>
>>> 	p = __vmalloc_node_range(size, MODULE_ALIGN, module_alloc_base,
>>> 				module_alloc_end, gfp_mask, PAGE_KERNEL, 0,
>>> 				NUMA_NO_NODE, __builtin_return_address(0));
>>>
>>> 	if (!p && IS_ENABLED(CONFIG_ARM64_MODULE_PLTS) &&
>>> 	    (IS_ENABLED(CONFIG_KASAN_VMALLOC) ||
>>> 	     (!IS_ENABLED(CONFIG_KASAN_GENERIC) &&
>>> 	      !IS_ENABLED(CONFIG_KASAN_SW_TAGS))))
>>> 		/*
>>> 		 * KASAN without KASAN_VMALLOC can only deal with module
>>> 		 * allocations being served from the reserved module region,
>>> 		 * since the remainder of the vmalloc region is already
>>> 		 * backed by zero shadow pages, and punching holes into it
>>> 		 * is non-trivial. Since the module region is not randomized
>>> 		 * when KASAN is enabled without KASAN_VMALLOC, it is even
>>> 		 * less likely that the module region gets exhausted, so we
>>> 		 * can simply omit this fallback in that case.
>>> 		 */
>>> 		p = __vmalloc_node_range(size, MODULE_ALIGN, module_alloc_base,
>>> 				module_alloc_base + SZ_2G, GFP_KERNEL,
>>> 				PAGE_KERNEL, 0, NUMA_NO_NODE,
>>> 				__builtin_return_address(0));
>>>
>>> 	if (p && (kasan_module_alloc(p, size) < 0)) {
>>> 		vfree(p);
>>> 		return NULL;
>>> 	}
>>>
>>> 	return p;
>>> }
>>>
>>> This should be happening quite rarely. But maybe arm64's document
>>> needs some minor fixup, otherwise, it is quite confusing.
>>
>> OK, so CONFIG_KASAN_VLALLOC=y and CONFIG_ARM64_MODULE_PLTS=y, the
>> module_alloc()
>> basically returns the memory in 128MB region, but can return the memory in 2GB
>> region. (This is OK because optprobe can filter it out)
>> But CONFIG_RANDOMIZE_MODULE_REGION_FULL=y, there is almost no chance to get
>> the memory in 128MB region.
>>
>> Hmm, for the optprobe in kernel text, maybe we can define 'optinsn_alloc_start'
>> by 'module_alloc_base - (SZ_2G - MODULES_VADDR)' and use __vmalloc_node_range()
>> to avoid this issue. But that is only for the kernel. For the modules, we may
>> always out of 128MB region.
> 
> If we can have some separate PLT entries in each module for optprobe,
> we should be able to short-jump to the PLT entry and then PLT entry
> will further long-jump to detour out of the range. That is exactly
> the duty of PLT.
> 
> Right now, arm64 has support on dynamic_ftrace by adding a
> section in module for ftrace PLT.
> arch/arm64/include/asm/module.lds.h:
> SECTIONS {
> #ifdef CONFIG_ARM64_MODULE_PLTS
> 	.plt 0 (NOLOAD) : { BYTE(0) }
> 	.init.plt 0 (NOLOAD) : { BYTE(0) }
> 	.text.ftrace_trampoline 0 (NOLOAD) : { BYTE(0) }
> #endif
> ...
> }
> 
> arch/arm64/kernel/module.c will initialize some PLT entries
> for ftrace:
> 
> static int module_init_ftrace_plt(const Elf_Ehdr *hdr,
> 				  const Elf_Shdr *sechdrs,
> 				  struct module *mod)
> {
> #if defined(CONFIG_ARM64_MODULE_PLTS) && defined(CONFIG_DYNAMIC_FTRACE)
> 	const Elf_Shdr *s;
> 	struct plt_entry *plts;
> 
> 	s = find_section(hdr, sechdrs, ".text.ftrace_trampoline");
> 	if (!s)
> 		return -ENOEXEC;
> 
> 	plts = (void *)s->sh_addr;
> 
> 	__init_plt(&plts[FTRACE_PLT_IDX], FTRACE_ADDR);
> 
> 	if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_REGS))
> 		__init_plt(&plts[FTRACE_REGS_PLT_IDX], FTRACE_REGS_ADDR);
> 
> 	mod->arch.ftrace_trampolines = plts;
> #endif
> 	return 0;
> }
> 
> Ftrace will then use those PLT entries in arch/arm64/kernel/ftrace.c:
> static struct plt_entry *get_ftrace_plt(struct module *mod, unsigned long addr)
> {
> #ifdef CONFIG_ARM64_MODULE_PLTS
> 	struct plt_entry *plt = mod->arch.ftrace_trampolines;
> 
> 	if (addr == FTRACE_ADDR)
> 		return &plt[FTRACE_PLT_IDX];
> 	if (addr == FTRACE_REGS_ADDR &&
> 	    IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_REGS))
> 		return &plt[FTRACE_REGS_PLT_IDX];
> #endif
> 	return NULL;
> }
> 
> /*
>   * Turn on the call to ftrace_caller() in instrumented function
>   */
> int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
> {
> 	unsigned long pc = rec->ip;
> 	u32 old, new;
> 	long offset = (long)pc - (long)addr;
> 
> 	if (offset < -SZ_128M || offset >= SZ_128M) {
> 		struct module *mod;
> 		struct plt_entry *plt;
> 
> 		if (!IS_ENABLED(CONFIG_ARM64_MODULE_PLTS))
> 			return -EINVAL;
> 
> 		/*
> 		 * On kernels that support module PLTs, the offset between the
> 		 * branch instruction and its target may legally exceed the
> 		 * range of an ordinary relative 'bl' opcode. In this case, we
> 		 * need to branch via a trampoline in the module.
> 		 *
> 		 * NOTE: __module_text_address() must be called with preemption
> 		 * disabled, but we can rely on ftrace_lock to ensure that 'mod'
> 		 * retains its validity throughout the remainder of this code.
> 		 */
> 		preempt_disable();
> 		mod = __module_text_address(pc);
> 		preempt_enable();
> 
> 		if (WARN_ON(!mod))
> 			return -EINVAL;
> 
> 		plt = get_ftrace_plt(mod, addr);
> 		if (!plt) {
> 			pr_err("ftrace: no module PLT for %ps\n", (void *)addr);
> 			return -EINVAL;
> 		}
> 
> 		addr = (unsigned long)plt;
> 	}
> 
> 	old = aarch64_insn_gen_nop();
> 	new = aarch64_insn_gen_branch_imm(pc, addr, AARCH64_INSN_BRANCH_LINK);
> 
> 	return ftrace_modify_code(pc, old, new, true);
> }
> 
> This might be the direction to go later. Anyway, "Rome wasn't built
> in a day", for this stage, we might focus on optprobe for the case
> of non-randomized module region :-).
> 
> BTW, @liuqi, if users set "nokaslr" in bootargs, will your optprobe
> always work and not fall back to normal kprobe even we remove the
> dependency on RANDOMIZED_MODULE_REGION_FULL?
> 
Hi Barry,

I do some tests on Hip08 platform, using nokaslr in booting cmdline and 
remove dependency on RANDOMIZED_MODULE_REGION_FULL, optprobe seems work.
Here is the log:

estuary:/$ uname -a
Linux (none) 5.13.0-rc4+ #37 SMP PREEMPT Mon Aug 2 08:13:37 CST 2021 
aarch64 GNU/Linux
estuary:/$ zcat /proc/config.gz | grep RANDOMIZE_MODULE_REGION
CONFIG_RANDOMIZE_MODULE_REGION_FULL=y
estuary:/$ zcat /proc/config.gz | grep OPTPROBE
CONFIG_OPTPROBES=y
CONFIG_HAVE_OPTPROBES=y
estuary:/$ cat /proc/cmdline
console=ttyAMA0,115200 earlycon=pl011,0x9000000 kpti=off nokaslr
estuary:/$ cat /sys/bus/platform/devices/hello_driver/kprobe_test
[   61.304143] do_empty returned 0 and took 200 ns to execute
[   61.304662] do_empty returned 0 and took 110 ns to execute
[   61.305196] do_empty returned 0 and took 100 ns to execute
[   61.305745] do_empty returned 0 and took 90 ns to execute
[   61.306262] do_empty returned 0 and took 90 ns to execute
[   61.306781] do_empty returned 0 and took 90 ns to execute
[   61.307286] do_empty returned 0 and took 90 ns to execute
[   61.307798] do_empty returned 0 and took 90 ns to execute
[   61.308314] do_empty returned 0 and took 90 ns to execute
[   61.308828] do_empty returned 0 and took 90 ns to execute
[   61.309323] do_empty returned 0 and took 80 ns to execute
[   61.309832] do_empty returned 0 and took 80 ns to execute
[   61.310357] do_empty returned 0 and took 80 ns to execute
[   61.310871] do_empty returned 0 and took 80 ns to execute
[   61.311361] do_empty returned 0 and took 80 ns to execute
[   61.311851] do_empty returned 0 and took 90 ns to execute
[   61.312358] do_empty returned 0 and took 90 ns to execute
[   61.312879] do_empty returned 0 and took 80 ns to execute

Thanks,
Qi

>>
>> Thank you,
>>
>> --
>> Masami Hiramatsu <mhiramat@kernel.org>
> 
> Thanks
> Barry
> .
> 


  reply	other threads:[~2021-08-02  3:52 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-19 12:24 Qi Liu
2021-07-21  8:41 ` Masami Hiramatsu
2021-07-22 10:24   ` Song Bao Hua (Barry Song)
2021-07-22 15:03     ` Masami Hiramatsu
2021-07-23  2:43       ` Song Bao Hua (Barry Song)
2021-07-30 10:04       ` Song Bao Hua (Barry Song)
2021-07-31  1:15         ` Masami Hiramatsu
2021-07-31 12:21           ` Song Bao Hua (Barry Song)
2021-08-02  3:52             ` liuqi (BA) [this message]
2021-08-02  3:59               ` liuqi (BA)
2021-08-02 12:02               ` liuqi (BA)
2021-07-30  3:32   ` liuqi (BA)

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=2f32fff3-6b58-583f-8e85-06ec1553d3f4@huawei.com \
    --to=liuqi115@huawei.com \
    --cc=anil.s.keshavamurthy@intel.com \
    --cc=catalin.marinas@arm.com \
    --cc=davem@davemloft.net \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linuxarm@huawei.com \
    --cc=mhiramat@kernel.org \
    --cc=naveen.n.rao@linux.ibm.com \
    --cc=prime.zeng@hisilicon.com \
    --cc=robin.murphy@arm.com \
    --cc=song.bao.hua@hisilicon.com \
    --cc=will@kernel.org \
    --subject='Re: [PATCH] arm64: kprobe: Enable OPTPROBE for arm64' \
    /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

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 NNTP newsgroup(s).