Re: [PATCH] mm,hugetlb: compute page_size_log properly

[Aneesh Kumar K.V]

Matthew Wilcox <willy@xxxxxxxxxxxxx> writes:

> On Tue, Mar 28, 2017 at 09:55:13AM -0700, Davidlohr Bueso wrote:
>> Do we have any consensus here? Keeping SHM_HUGE_* is currently
>> winning 2-1. If there are in fact users out there computing the
>> value manually, then I am ok with keeping it and properly exporting
>> it. Michal?
>
> Well, let's see what it looks like to do that.  I went down the rabbit
> hole trying to understand why some of the SHM_ flags had the same value
> as each other until I realised some of them were internal flags, some
> were flags to shmat() and others were flags to shmget().  Hopefully I
> disambiguated them nicely in this patch.  I also added 8MB and 16GB sizes.
> Any more architectures with a pet favourite huge/giant page size we
> should add convenience defines for?
>
> diff --git a/include/linux/shm.h b/include/linux/shm.h
> index 04e881829625..cd95243efd1a 100644
> --- a/include/linux/shm.h
> +++ b/include/linux/shm.h
> @@ -24,26 +24,13 @@ struct shmid_kernel /* private to the kernel */
>  	struct list_head	shm_clist;	/* list by creator */
>  };
>  
> -/* shm_mode upper byte flags */
> -#define	SHM_DEST	01000	/* segment will be destroyed on last detach */
> -#define SHM_LOCKED      02000   /* segment will not be swapped */
> -#define SHM_HUGETLB     04000   /* segment will use huge TLB pages */
> -#define SHM_NORESERVE   010000  /* don't check for reservations */
> -
> -/* Bits [26:31] are reserved */
> -
>  /*
> - * When SHM_HUGETLB is set bits [26:31] encode the log2 of the huge page size.
> - * This gives us 6 bits, which is enough until someone invents 128 bit address
> - * spaces.
> - *
> - * Assume these are all power of twos.
> - * When 0 use the default page size.
> + * These flags are used internally; they cannot be specified by the user.
> + * They are masked off in newseg().  These values are used by IPC_CREAT
> + * and IPC_EXCL when calling shmget().
>   */
> -#define SHM_HUGE_SHIFT  26
> -#define SHM_HUGE_MASK   0x3f
> -#define SHM_HUGE_2MB    (21 << SHM_HUGE_SHIFT)
> -#define SHM_HUGE_1GB    (30 << SHM_HUGE_SHIFT)
> +#define	SHM_DEST	01000	/* segment will be destroyed on last detach */
> +#define SHM_LOCKED      02000   /* segment will not be swapped */
>  
>  #ifdef CONFIG_SYSVIPC
>  struct sysv_shm {
> diff --git a/include/uapi/linux/shm.h b/include/uapi/linux/shm.h
> index 1fbf24ea37fd..44b36cb228d7 100644
> --- a/include/uapi/linux/shm.h
> +++ b/include/uapi/linux/shm.h
> @@ -40,15 +40,34 @@ struct shmid_ds {
>  /* Include the definition of shmid64_ds and shminfo64 */
>  #include <asm/shmbuf.h>
>  
> -/* permission flag for shmget */
> +/* shmget() shmflg values. */
> +/* The bottom nine bits are the same as open(2) mode flags */
>  #define SHM_R		0400	/* or S_IRUGO from <linux/stat.h> */
>  #define SHM_W		0200	/* or S_IWUGO from <linux/stat.h> */
> +/* Bits 9 & 10 are IPC_CREAT and IPC_EXCL */
> +#define SHM_HUGETLB     (1 << 11) /* segment will use huge TLB pages */
> +#define SHM_NORESERVE   (1 << 12) /* don't check for reservations */
>  
> -/* mode for attach */
> -#define	SHM_RDONLY	010000	/* read-only access */
> -#define	SHM_RND		020000	/* round attach address to SHMLBA boundary */
> -#define	SHM_REMAP	040000	/* take-over region on attach */
> -#define	SHM_EXEC	0100000	/* execution access */
> +/*
> + * When SHM_HUGETLB is set bits [26:31] encode the log2 of the huge page size.
> + * This gives us 6 bits, which is enough until someone invents 128 bit address
> + * spaces.  These match MAP_HUGE_SHIFT and MAP_HUGE_MASK.
> + *
> + * Assume these are all powers of two.
> + * When 0 use the default page size.
> + */
> +#define SHM_HUGE_SHIFT	26
> +#define SHM_HUGE_MASK	0x3f
> +#define SHM_HUGE_2MB	(21 << SHM_HUGE_SHIFT)
> +#define SHM_HUGE_8MB	(23 << SHM_HUGE_SHIFT)
> +#define SHM_HUGE_1GB	(30 << SHM_HUGE_SHIFT)
> +#define SHM_HUGE_16GB	(34 << SHM_HUGE_SHIFT)


This should be in arch/uapi like MAP_HUGE_2M ? That will let arch add
#defines based on the hugepae size supported by them.

> +
> +/* shmat() shmflg values */
> +#define	SHM_RDONLY	(1 << 12) /* read-only access */
> +#define	SHM_RND		(1 << 13) /* round attach address to SHMLBA boundary */
> +#define	SHM_REMAP	(1 << 14) /* take-over region on attach */
> +#define	SHM_EXEC	(1 << 15) /* execution access */
>  

-aneesh