Anaconda, parted, and geometry

[Chuck Anderson]

For reasons known only to IBM, Thinkpads have been shipping with hard 
disks partitioned with a geometry of 240 heads, 63 sectors.  Even my 
brand shiny new T61 does this.  However, since Linux has stopped 
asking the BIOS what geometry to use, it now defaults to 255 heads, 63 
sectors.  Further, it isn't clear to me how to override Linux's choice 
of hdd geometry in the new world order of libata, nor should that be 
necessary in the normal case to get a sane partition table IMO.

Why should I care, you ask?  Isn't disk geometry an anachronism from 
the days of DOS?  Well, yes, but the problem is for whatever reason, 
not everything ignores geometry...

For example, Anaconda/parted likes to force cylinder alignment.  
Windows uses the BIOS/partition table geometry which may have a 
different idea about where cylinders begin and end.  The reasons for 
these behaviors aren't entirely clear to me.

This leads to /really/ weird partitioning when different programs with 
different ideas of the geometry add/delete partitions from the disk, 
such as strange gaps of free space when you create a new partition in 
Anaconda, or situations where some partition table entries are stored 
using one set of C,H,S values, and others are using a different set. 
This of course also causes programs like "fdisk" and "sfdisk" to 
complain about cylinder boundaries and C,H,S values being incorrect.

What to do about it?  Can't we all agree to use the same geometry when 
dealing with the partition table?  When Anaconda/parted reads the 
table, shouldn't it deduce the most fitting C,H,S values to use for 
cylinder alignment and writing out new entries?  Or shouldn't it ask 
the BIOS what to use, since that seems to be what Windows does?

I used to work around issues like this by using fdisk in VT2 to 
partition things how I like, and then let Anaconda install to those.  
However, it now seems impossible to create a new encrypted LVM PV 
unless you let Anaconda's parted create the PV partition too.  Perhaps 
that could be improved upon.

I remember a bit of the fiasco of "I can't boot Windows anymore" that 
happened a few years ago, but I don't know what the outcome/solution 
was.  Did the Linux kernel and Anaconda just punt the whole issue of 
trying to match geometries and let things fall as they may?

Whatever was done, it just seems wrong and dirty to end up with a disk 
that has a schizophrenic idea of what geometry to use for its 
different partitions.

--Chuck, who had to create a spreadsheet just to figure out what 
happend to his partition table...