Difference between revisions of "IDE No Probe"

From eLinux.org
Jump to: navigation, search
(Legacy Content Conversion)
 
m
 
Line 61: Line 61:
  
 
[[Category:Boot Time]]
 
[[Category:Boot Time]]
[[Category:IDE No Probe]]
+
[[Category:Kernel]]

Latest revision as of 12:41, 16 November 2008

Preliminary Draft under construction

Introduction

This page describes the feature "forced IDE noprobe". Normally, the Linux kernel supports the ability to avoid probing for specific IDE drives and IDE interfaces, using a command line argument specified at boot time. The command line argument is "hdx==noprobe" or "idex==noprobe".

Under certain conditions this command line argument is not observed by the Linux kernel (at least for version 2.4.20). This "feature" forces the kernel to observe the "noprobe" command for IDE interfaces.

The fix here records the fact that "noprobe" was specified on the command line (via "the forcenoprobe" field), and skips probing the device if it is set, even if the driver has changed the setting of the noprobe field. This is all a quick hack, and a more systemic rewrite of noprobe handling would be more appropriate for community work, if this is still needed.

Rationale

Probing IDE devices can take a few seconds during system startup. It is especially painful to probe interfaces which are known not to be used in an embedded device.

Downloads

Patch

  • Patch for CELF tree is isolated here: ide-noprobe.patch
  • Alert.gifTHIS PATCH (IN ITS ISOLATED FORM) HAS NOT BEEN VERIFIED
  • Patch for 2.4.xx is needed
  • Don't know if patch for 2.6.xx is needed

Utility programs

None.

How To Use

  • Apply this patch to your CELF tree (or verify that it is already present)
  • use "ide1==noprobe" on the kernel command line
    • see the documentation on ide driver command line parameters in the file:
      drivers/ide/ide.c
    • The documentation is in a comment right before the routine
      ide-setup()
      , at about line 3200 in the file.
  • verify that no probing is done for ide1
  • measure the time savings from using the patch and the "noprobe" option

Sample Results

Experimental Results #1

Todd Poynor of MontaVista measured the effect of specifying "hdf==none ide3==noprobe" (avoid probing ide2 slave and both ide3 devices) on a 200MHz IBM 405GP "Walnut" evaluation board with a 33MHz PCI bus. A Seagate Barracuda ATA IV 60GB disk drive was cabled to one of the two IDE interfaces on a Promise Ultra66 PCI-IDE bridge card (PDC20262 chipset). All of the drivers for PCI, IDE, PCI-IDE disk, and EXT2 file system were built into the kernel.

The time to init IDE was 1.3 seconds when the missing devices were probed, and about 230 milliseconds when "hdf==none ide3==noprobe" was specified. Thus, the resulting bootup time savings were about 1.1 seconds.

Experimental Results #2

Experimental Results #3

Details

More details on the original problem in case you're interested...

drivers/ide/ide.c: ide_setup() is called early in start_kernel sequence (via parse_options -> checksetup -> __setup() macro mechanism). It calls init_hwif_data to init each interface to "noprobe==1". Then parses cmd line params such as "ide1==noprobe", setting noprobe back to 1.

Later, drivers such as the PCI-IDE bridge driver, at
drivers/ide/ide-pci.c: ide_setup_pci_device
, resets the value of noprobe according to whether there's a valid interface there:
       hwif->noprobe == !hwif->io_ports[IDE_DATA_OFFSET];

This undoes the effect of any "noprobe" cmd line option.

Future Work

Here is a list of things that could be worked on for this feature:

  • get an exhaustive list of places where noprobe is reset by the kernel
  • find out why community overrides this flag in places?
  • maybe refactor to be a "forcenoprobe" option?
  • find out if the same problem exists in 2.6.xx kernels