Necessary tools before going on:

download the kernel you want from kernel.org and put it in /usr/src

(the following example uses the 2.6.9 kernel)

now you have three possibilities:

1a) Upgrading the current kernel, for example you wish to add the v2.6.9 and you have 2.6.8 installed and working.

In order to prevent a kernel panic you're going to start from the current kernel's configuration taking the files in /boot . Assuming you are in /usr/src/linux, the commands are:

explain: the cp command copy into the new sources the 2.6.8 kernel's configuration by renaming the file from config-2.6.8 to .config; afterwards, the $ make oldconfig command will adapt, modify and make this .config compatible with the 2.6.9 kernel sources you're going to compile.

Obviously a newer kernel contains new features, thus this last command will ask you some little question on the terminal (for most them, just confirm the default proposed setting by pressing Enter). This happens according to news in the kernel and to the age of the previous configuration file (config-2.6.8)

2a) Recompiling right the same kernel version (this often happens, when the user wants to add a new support in the kernel, wants to change his picture, to make the kernel smaller, to add a module, etc...):

As an example we assume a 2.6.8 kernel, of course change the numbering for your needs.

Assuming the fact you are in /usr/src/linux type these commands:

what it does : the cp command copy into the sources the same current kernel configuration by simply renaming from config-2.6.8 in .config; afterwards the $ make menuconfig command will help you doing the desired changings to the kernel configuration through a ncurses interface. Once you added what you need or deleted what you don't want exit and save the file when the dialog window will ask you.

3a) Starting from scratch and configuring a new kernel by hand (very rare):
assuming the fact you are in /usr/src/linux type this command:

Anyway YOU configured your kernle I suggest you to apply this features to have more performance by your machine:

1. In «General Setup» choose:

   • -Support for paging of anonymous memory (swap)

     • --Support for prefetching swapped memory

2. In «Processor type and features»:

   • In «Processor family» choose ypur CPU model (for me is Dual Core Duo/Xeon).

   • In Preemption Model-> Voluntary Kernel Preemption (Desktop)

   • In «High Memory Support» choose:

     • off if you have < of 1 GB of RAM

     • 1GB Low Memory Support if you have = 1GB of RAM

     • 4GB if you have > of 1GB of RAM (up to 64 GB)

   • In «Timer frequency» choose : 1000 Hz

3. In «Block layer -> IO Schedulers» chose ONLY «CFQ I/O scheduler», you ''ll have more performance.

4. In «Kernel hacking» deselect «Kernel debugging».

5. In «Device Drivers -> Character devices» select «Virtual terminal».

Now the following steps are the same:

N.B.: the initrd image (RamDisk) is a portion of memory allocated as a partition, that is, it takes the memory and uses it as an hard disk, saving there some files. 

This technique is usually used to load some modules before installing the definitive filesystem, for example when the kernel requires a special module to access a filesystem; sometimes this happens, expecially on a scsi disk or raid controller.

Thus you'd better add --initrd when you create the kernel image, so ue$ make-kpkg --initrd kernel_image insted of $ make-kpkg kernel_image that I wrote above. You have also the possibility to create the initrd image later with the mkinitrd tool contained in initrd-tools, just type:

that create a subversion of the kernel and it's very comfortable. Indeed, it's enough to manually remove the old modules before installing kernel-image.deb (with dpkg -i) with:

$rm -rf /lib/modules/2.6.x