IDE Preempt Specification
#noprint VERSION 0.2 - for other versions, click the "info" button. '''Table Of Contents:''' [[TableOfContents]] #noprint
The IDE driver in the Linux kernel normally uses a series of busywait delays during its initialization. When the driver executes these busywaits, the kernel does nothing for the duration of the wait. The time spent in these waits could be used for other initialization activities, if they could be run concurrently with these waits.
More specifically, busywait-style delays such as udelay() in module init functions inhibit kernel preemption because the Big Kernel Lock is held, while yielding APIs such as schedule_timeout() allow preemption. This is true because the kernel handles the BKL specially and releases and reacquires it across reschedules allowed by the current thread.
This IDE-preempt specification requires that the driver eliminate these busywaits and replace them with a mechanism that allows other work to proceed while the IDE driver is initializing.
IDE modules are one of the major offenders with regard to executing non-preemptible (busywait) delays during initialization. The ide-probe-mod driver spends a great deal of time in repeated calls to ide_delay_50ms() during probe and drive identification. The driver busy waits in order to let the IDE controller make progress before polling for status or to allow previous operations to complete.
The total time spent in these busywait calls in some testing was on the order of 250 milliseconds. (Testing details are available on the page mentioned below in the References section.) By changing these calls to something that can be preempted, other initialization work can potentially proceed in parallel with the IDE driver init, and the total system boot time can be reduced.
- The Linux kernel SHOULD use preemptible waits during IDE driver initialization, rather than busywaits. 2. When this feature is supported, the IDE driver SHOULD NOT impose a maximum period of non-preemptibility in the kernel of greater than 20 milliseconds.
Notes (informational and non-normative)
- It is expected that using a preemptible wait in the kernel will have no benefit in the current (2.4) kernel driver initialization scheme unless the IDE driver is loaded as a module. 2. It is expected that the overhead of using a preemptible wait in the IDE driver may lengthen the total time for the IDE driver to complete it's initialization. However, the total CPU time available for system initialization should be increased by this feature. That is, more time should be available for other initialization tasks, which should result in a net decrease of boot time. 3. This feature is not required (that is, it is not specified as a "SHALL" requirement) because it is not clear that this feature is valuable for a large set of embedded configurations. The fact that the feature requires that the IDE driver be loaded as a module (in order to achieve concurrency gains) is a serious drawback. However, this feature offers some advantage for some situations, and points the way toward more general techniques for improved concurrency in driver initializations. 4. The value of 20 milliseconds in item 2 of the specification may seem somewhat arbitrary. Following is some rationale for the selection of this number.
Because of the structure of the IDE driver initialization for the 2.4 kernel, the preempt-off period is an indicator for the amount of time that the CPU is NOT available for other concurrent initialization activities. Thus, by recommending a maximum non-preemptibility value for the IDE driver init, the specification is really indicating that some significant amount of time, formerly wasted by IDE busywaits, should be made available for other initialization activities.
The non-preemptible period is normally equal to the sum of all IDE driver busywaits combined. With a preemptible wait, the non-preemptible period should be smaller than the duration of a single busywait.
In 2.4.20, the wait is executed in the routine ide_delay_50ms(), which, as its name suggests, delays for 50 milliseconds. More than half of the busywait interval should become available for other processes to use, if the feature is to have any real value.
Also, in order to be reasonable, the value chosen should be longer than the maximum preemption off period that has been measured in testing of this feature. This is about 10 milliseconds.
Therefore, 20 milliseconds represents a "rule-of-thumb" guideline for the maximum amount of time that the IDE driver waits should be non-preemptible.
A patch is available which implements this features. Please see the Preempt IDE Preempt page.