[lkml]   [1998]   [Sep]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: 2.1.118 Tons of oopes (arms waving violently!!!)
    > Potentially. I've thought about it on and off, but it's painful to add the
    > entries to every single driver, and the win has never been very big.

    Did you see my message on this? You can do auto initialization in register_blkdev.
    Additionally, my suggestion changes the device arrays to simply hardwire invalid
    devices to a fops which simply returns 0 (or NULL or whatever) when it hits a bad

    The math is pretty simple:

    right now, we're loading the struct from the device array, checking a pointer in
    the structure, then we're executing that or returning NULL.

    My method was to simply preset all the NULL entries in register_blkdev to point
    to the generic case equivelent and blindly call the function with a simple inline

    #define function(dev) do { \
    device_array[MAJOR(dev)]->function(dev); \
    } while(0)

    It's simple, it's elegant and it leverages precomputed tables all over the place.
    I would reckon that we could gain a usec or more in benchmarks with this optimization.

    > Also note that many of the current NULL checks (not the ones for the fops
    > thing, but for other things) are superfluous: we check for NULL not
    > because we need to, but often because it's just an almost automatic thing
    > for me.
    > For example, there is sometimes code essentially like this (see
    > sys_lseek(), for example)
    > struct dentry * dentry;
    > struct inode * inode;
    > struct file * filp = fget(fd);
    > if (!filp)
    > return -EBADF;
    > dentry = filp->f_dentry;
    > if (!dentry)
    > return -EBADF;
    > inode = dentry->d_inode;
    > if (!inode)
    > return -BADF;
    > and the only check that needs to be done is really the filp one: if you
    > get a filp that is active, then the dentry and inode had better be there,
    > because it would be a major error (worthy of a kernel oops) if they
    > weren't, so checking for dentry/inode being non-NULL is unnecessary, but
    > often done simply because I and others have been anal about it.
    > Personally, I would suggest against changing all the file and inode
    > operations unless somebody can show cases where it makes a real-world
    > difference in performance..
    > Linus
    > -
    > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    > the body of a message to
    > Please read the FAQ at

    Perry Harrington Linux rules all OSes. APSoft ()
    email: Think Blue. /\

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    Please read the FAQ at

     \ /
      Last update: 2005-03-22 13:44    [W:0.025 / U:11.848 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site