so what if I'm not running on a PC?

[canadian_worm]

geekish post

taken from the ramdisk block driver code, deep in the 2.6 kernel:

static int rd_open(struct inode *inode, struct file *filp)
{
unsigned unit = iminor(inode);

if (rd_bdev[unit] == NULL) {
struct block_device *bdev = inode->i_bdev;
struct address_space *mapping;
unsigned bsize;
gfp_t gfp_mask;

inode = igrab(bdev->bd_inode);
rd_bdev[unit] = bdev;
bdev->bd_openers++;
bsize = bdev_hardsect_size(bdev);
bdev->bd_block_size = bsize;
inode->i_blkbits = blksize_bits(bsize);
inode->i_size = get_capacity(bdev->bd_disk)<<9;

mapping = inode->i_mapping;
mapping->a_ops = &ramdisk_aops;
mapping->backing_dev_info = &rd_backing_dev_info;
bdev->bd_inode_backing_dev_info = &rd_file_backing_dev_info;

/*
* Deep badness. rd_blkdev_pagecache_IO() needs to allocate
* pagecache pages within a request_fn. We cannot recur back
* into the filesytem which is mounted atop the ramdisk, because
* that would deadlock on fs locks. And we really don't want
* to reenter rd_blkdev_pagecache_IO when we're already within
* that function.
*
* So we turn off __GFP_FS and __GFP_IO.
*
* And to give this thing a hope of working, turn on __GFP_HIGH.
* Hopefully, there's enough regular memory allocation going on
* for the page allocator emergency pools to keep the ramdisk
* driver happy.
*/
gfp_mask = mapping_gfp_mask(mapping);
gfp_mask &= ~(__GFP_FS|__GFP_IO);
gfp_mask |= __GFP_HIGH;
mapping_set_gfp_mask(mapping, gfp_mask);
}

return 0;
}

so, what happens if I am not running this code on a PC, and don't happen
to have extra high mem (GFP_HIGH) kicking about to rescue the the page
allocator emergency pools?

really, imho, the file system and page caching should be designed to
work with or without it. no assumptions about architecture should be made
at this level!