more questions about UAPI content -- why some usages of "__KERNEL__"?

Robert P. J. Day rpjday at crashcourse.ca
Mon Jan 23 06:26:08 EST 2017


  a couple more questions about UAPI in the kernel source, just to
clarify that there is still some cleanup that could be done.

  first, i get that whole idea behind the include/uapi/ directory ...
that it's way cleaner to factor out the userspace content to a single
location, and use that as the basis for a subsequent "kernel-headers"
package or something to that effect. but even having done that, there
is of course some necessary usage of testing the __KERNEL__ macro to
see if one is preprocessing in kernel space or not.

  one of these uses is, say, in a uapi header file, there might be a
test to include a structure member depending on whether you're in
kernel space or user space. but consider the following example.

  here's include/linux/mtd/inftl.h, wherein an entire chunk of that
header file is protected by a "__KERNEL__" test:

#ifndef __MTD_INFTL_H__
#define __MTD_INFTL_H__

... snip ...

#ifdef __KERNEL__              <--- start test

struct INFTLrecord {
        struct mtd_blktrans_dev mbd;
        __u16 MediaUnit;
        __u32 EraseSize;
        struct INFTLMediaHeader MediaHdr;
        int usecount;
        unsigned char heads;
        unsigned char sectors;
        unsigned short cylinders;
        __u16 numvunits;
        __u16 firstEUN;
        __u16 lastEUN;
        __u16 numfreeEUNs;
        __u16 LastFreeEUN;              /* To speed up finding a free EUN */
        int head,sect,cyl;
        __u16 *PUtable;                 /* Physical Unit Table */
        __u16 *VUtable;                 /* Virtual Unit Table */
        unsigned int nb_blocks;         /* number of physical blocks */
        unsigned int nb_boot_blocks;    /* number of blocks used by the bios */
        struct erase_info instr;
};

int INFTL_mount(struct INFTLrecord *s);
int INFTL_formatblock(struct INFTLrecord *s, int block);

void INFTL_dumptables(struct INFTLrecord *s);
void INFTL_dumpVUchains(struct INFTLrecord *s);

int inftl_read_oob(struct mtd_info *mtd, loff_t offs, size_t len,
                   size_t *retlen, uint8_t *buf);
int inftl_write_oob(struct mtd_info *mtd, loff_t offs, size_t len,
                    size_t *retlen, uint8_t *buf);

#endif /* __KERNEL__ */              <--- end test

#endif /* __MTD_INFTL_H__ */


  is that test really necessary? this is a header file that should be
included *only* from within kernel space -- how could it possibly end
up being used from user space? or am i missing something?

  i can see the value in uapi header files still needing to test
__KERNEL__, in case there are minor tweaks to be done. but it seems
that there should be precious little need for testing __KERNEL__ in
*kernel* header files once userspace content has been factored out and
moved to uapi, no?

  thoughts?

rday

-- 

========================================================================
Robert P. J. Day                                 Ottawa, Ontario, CANADA
                        http://crashcourse.ca

Twitter:                                       http://twitter.com/rpjday
LinkedIn:                               http://ca.linkedin.com/in/rpjday
========================================================================




More information about the Kernelnewbies mailing list