NRAO Home  >  Green Bank  |  Wiki Topic:    GB > Software > TWikiUsers > JoeBrandt > DcrRTAIMagmaNotes

Changes | Index | Contents | Search | Statistics | Go

DCR on RHE4 Configuration Notes

Getting the Software

• The www.gna.org site hosts (or mirrors) the RTAI CVS site.
• The main distribution site for kernel sources (Note generic, not Redhat) is www.kernel.org

Driver 2.4 to 2.6 Porting Notes:

• copy_to/from_user() return value must not be ignored, and typically should return -EFAULT (Now defined with attribute warn_unused_result.)
• read{b,w,l} and write{b,w,l} become ioread{8,16,32} and iowrite{8,16,32} respectively
• pci_present(), pcibios_present() are obsolete. Use CONFIG_PCI token. (What about non-pci systems running pci-enabled kernels?
• check_region() obsolete, use return value of request_region()
• cli/sti must be removed (see (kernel release)/Documentation/cli-sti-removal.txt) new functions are:
  • local_irq_disable() => turn local IRQs off
  • local_irq_enable() => turn local IRQs on
  • local_save_flags(flags) => save the current IRQ state into flags. The state can be on or off. (on some architectures there's even more bits in it.)
  • local_irq_save(flags) => save the current IRQ state into flags and disable interrupts.
  • local_irq_restore(flags) => restore the IRQ state from flags.
• 'old style' PCI probing is still defined (e.g. pci_find_class()) is defined in RHE4, but not in 2.6.14.3
• remap_page_range() still in RHE4, should be remap_pfn_range() (see http://lwn.net/Articles/28746)
• interrupt handler return value type (see http://lwn.net/Articles/23625/)

Interactions with do_gettimeofday()

The perscription is to always check for the locked condition prior to calling do_gettimeofday(). If we find it locked, use the tsc timer to estimate and fake a time. This is all nicely packaged in a RTAI safe function 'rt_do_gettimeofday()'.

In 2.4 kernels, the lock type was a rwlock_t, in 2.6 this has changed to a seqlock_t, which is faster. The write lock increments a sequence number, as does the write_unlock. In order to detect a write-lock condition, we look for an odd (as in 1,3,5,7) return value. Works fine.

Misc. To Do Items:

• reconfigure the kernel for 4GB address space, right now it only "sees" the first GB of memory.
• load/unload scripts
• ntp (time deamon) integration

OS Build and Configuration

• The hyperthreading issue. The 2.6 series appears to have solid support for hyperthreading CPU's. However the developers of RTAI recommend disabling the feature and using non-SMP kernels (on single chip systems).
• The test system is a SATA only system, which is well supported under 2.6 (the current DCR has PATA disks installed). Getting the SATA and required SCSI code configured in as modules is the key.
• Disabled ACPI power management, though it is still used for APIC routing/management.
• Since the custom kernel has the NPTL patch, nothing needs to be done to diable it. (To clarify, on Redhat 2.4 systems, the 2.6 NPTL patch was back-ported, and the system had NPTL ready libraries. When using a kernel.org+RTAI custom kernel, NPTL was not supported and required a bit of work to disable it.)

Quick-start

Start by downloading the latest release of RTAI:

• cvs -d:pserver:anonymous@cvs.gna.org:/cvs/rtai co modulename
  • Where "modulename" is [ "magma" | "vulcano" | ... ]

Unpack the source look in .../base/arch/i386/patches to see what kernel versions are supported, then go get that source version from kernel.org.

Unpack and patch the source like so:

• tar jxf linux-(version).bz2
• cd linux-(version)
• patch -p1 < patch_file_from_rtai

Configure the kernel ... build, install, Reboot, test normal operation ...

Configure RTAI ... build, install, run the kernel and user-space testsuites

-- JoeBrandt - 02 Dec 2005

Attachment:	Action:	Size:	Date:	Who:	Comment:
config-2.6.14.3-RTAI-Magma	action	51502	02 Dec 2005 - 21:11	JoeBrandt	Initial working version for 'tank'.