linux-kernel - Re: [patch V2 11/16] x86/ioperm: Share I/O bitmap if identical

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]

Message-ID: <20191112071406.GC100264@gmail.com>
Date:   Tue, 12 Nov 2019 08:14:06 +0100
From:   Ingo Molnar <mingo@...nel.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     LKML <linux-kernel@...r.kernel.org>, x86@...nel.org,
        Linus Torvalds <torvalds@...uxfoundation.org>,
        Andy Lutomirski <luto@...nel.org>,
        Stephen Hemminger <stephen@...workplumber.org>,
        Willy Tarreau <w@....eu>, Juergen Gross <jgross@...e.com>,
        Sean Christopherson <sean.j.christopherson@...el.com>,
        "H. Peter Anvin" <hpa@...or.com>
Subject: Re: [patch V2 11/16] x86/ioperm: Share I/O bitmap if identical


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

> The I/O bitmap is duplicated on fork. That's wasting memory and slows down
> fork. There is no point to do so. As long as the bitmap is not modified it
> can be shared between threads and processes.
> 
> Add a refcount and just share it on fork. If a task modifies the bitmap
> then it has to do the duplication if and only if it is shared.
> 
> Signed-off-by: Thomas Gleixner <tglx@...utronix.de>
> ---
> V2: New patch
> ---
>  arch/x86/include/asm/iobitmap.h |    5 +++++
>  arch/x86/kernel/ioport.c        |   38 ++++++++++++++++++++++++++++++++------
>  arch/x86/kernel/process.c       |   39 ++++++---------------------------------
>  3 files changed, 43 insertions(+), 39 deletions(-)
> 
> --- a/arch/x86/include/asm/iobitmap.h
> +++ b/arch/x86/include/asm/iobitmap.h
> @@ -2,10 +2,12 @@
>  #ifndef _ASM_X86_IOBITMAP_H
>  #define _ASM_X86_IOBITMAP_H
>  
> +#include <linux/refcount.h>
>  #include <asm/processor.h>
>  
>  struct io_bitmap {
>  	u64			sequence;
> +	refcount_t		refcnt;
>  	unsigned int		io_bitmap_max;
>  	union {
>  		unsigned long	bits[IO_BITMAP_LONGS];

> +void io_bitmap_share(struct task_struct *tsk)
> + {
> +	/*
> +	 * Take a refcount on current's bitmap. It can be used by
> +	 * both tasks as long as none of them changes the bitmap.
> +	 */
> +	refcount_inc(&current->thread.io_bitmap->refcnt);
> +	tsk->thread.io_bitmap = current->thread.io_bitmap;
> +	set_tsk_thread_flag(tsk, TIF_IO_BITMAP);
> +}

Ok, this is really neat. I suspect there might be some pathological cases 
on ancient NUMA systems with a really high NUMA factor and bad caching 
where this new sharing might regress performance, but I doubt this 
matters, as both the hardware and this software functionality is legacy.


> +	/*
> +	 * If the bitmap is not shared, then nothing can take a refcount as
> +	 * current can obviously not fork at the same time. If it's shared
> +	 * duplicate it and drop the refcount on the original one.
> +	 */
> +	if (refcount_read(&iobm->refcnt) > 1) {
> +		iobm = kmemdup(iobm, sizeof(*iobm), GFP_KERNEL);
> +		if (!iobm)
> +			return -ENOMEM;
> +		io_bitmap_exit();
>  	}
>  
> +	/* Set the tasks io_bitmap pointer (might be the same) */

speling nit:

s/tasks
 /task's

Thanks,

	Ingo