linux-kernel - Re: [patch 3/3] x86/paravirt: Replace paravirt patch asm magic

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 25 Apr 2019 08:52:09 +0200
From:   Ingo Molnar <mingo@...nel.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     LKML <linux-kernel@...r.kernel.org>, x86@...nel.org,
        Juergen Gross <jgross@...e.com>,
        Andi Kleen <ak@...ux.intel.com>
Subject: Re: [patch 3/3] x86/paravirt: Replace paravirt patch asm magic


* Thomas Gleixner <tglx@...utronix.de> wrote:

> The magic macro DEF_NATIVE() in the paravirt patching code uses inline
> assembly to generate a data table for patching in the native instructions.
> 
> While clever this is falling apart with LTO and even aside of LTO the
> construct is just working by chance according to GCC folks.
> 
> Aside of that the tables are constant data and not some form of magic
> text.
> 
> As these constructs are not subject to frequent changes it is not a
> maintenance issue to convert them to regular data tables which are
> initialized with hex bytes.
> 
> Create a new set of macros and data structures to store the instruction
> sequences and convert the code over.
> 
> Reported-by: Andi Kleen <ak@...ux.intel.com>
> Signed-off-by: Thomas Gleixner <tglx@...utronix.de>

> -# ifdef CONFIG_PARAVIRT_XXL
> -DEF_NATIVE(irq, irq_disable, "cli");
> -DEF_NATIVE(irq, irq_enable, "sti");
> -DEF_NATIVE(irq, restore_fl, "push %eax; popf");
> -DEF_NATIVE(irq, save_fl, "pushf; pop %eax");
> -DEF_NATIVE(cpu, iret, "iret");
> -DEF_NATIVE(mmu, read_cr2, "mov %cr2, %eax");
> -DEF_NATIVE(mmu, write_cr3, "mov %eax, %cr3");
> -DEF_NATIVE(mmu, read_cr3, "mov %cr3, %eax");

> +static const struct patch_xxl patch_data_xxl = {
> +	.irq_irq_disable	= { 0xfa },		// cli
> +	.irq_irq_enable		= { 0xfb },		// sti
> +	.irq_save_fl		= { 0x9c, 0x58 },	// pushf; pop %[re]ax
> +	.mmu_read_cr2		= { 0x0f, 0x20, 0xd0 },	// mov %cr2, %[re]ax
> +	.mmu_read_cr3		= { 0x0f, 0x20, 0xd8 },	// mov %cr3, %[re]ax
> +# ifdef CONFIG_X86_64
> +	.irq_restore_fl		= { 0x57, 0x9d },	// push %rdi; popfq
> +	.mmu_write_cr3		= { 0x0f, 0x22, 0xdf },	// mov %rdi, %cr3
> +	.cpu_wbinvd		= { 0x0f, 0x09 },	// wbinvd
> +	.cpu_usergs_sysret64	= { 0x0f, 0x01, 0xf8,
> +				    0x48, 0x0f, 0x07 },	// swapgs; sysretq
> +	.cpu_swapgs		= { 0x0f, 0x01, 0xf8 },	// swapgs
> +	.mov64			= { 0x48, 0x89, 0xf8 },	// mov %rdi, %rax
> +# else
> +	.irq_restore_fl		= { 0x50, 0x9d },	// push %eax; popf
> +	.mmu_write_cr3		= { 0x0f, 0x22, 0xd8 },	// mov %eax, %cr3
> +	.cpu_iret		= { 0xcf },		// iret
> +# endif

I think these open-coded hexa versions are somewhat fragile as well, how 
about putting these into a .S file and controlling the sections in an LTO 
safe manner there?

That will also allow us to write proper asm, and global labels can be 
used to extract the patchlets and their length?

Thanks,

	Ingo