Starting with Oracle Solaris 11.3, when a system crashes, the crash dump files might be preserved in memory until after the system reboots. When the system is rebooting, the crash dump files are extracted from memory to the file system that is defined in the dump configuration. Once these files are written, the system automatically reboots to normal multi-user configuration. This process is referred to as a deferred dump. Deferred dumps enable systems to return to a running state more quickly after a kernel panic. In addition, deferred dumps specifically benefit systems, such as the SPARC M7 series servers, which may ship without a local disk.
![]() | Caution - The first reboot performs the deferred dump. On x86 systems, a second reboot might automatically occur to eliminate any possible performance issues caused by memory fragmentation. Do not interrupt this process. The system is ready for use after the second reboot. |
The deferred dump process includes the following system messages:
When the system is performing a deferred dump to memory, a Preserving kernel image in RAM message with a progress bar appears.
When the system reboots after the panic, a Verified previous kernel image notice followed by a Reconciling deferred dump message with a progress bar appears.
When savecore is saving the crash dump files to the file system, an Extracting crash dump message with a progress bar appears.
On x86 systems, deferred dumps work in conjunction with the default fast reboot feature in Oracle Solaris. Therefore, a deferred dump on an x86 system can function only on platforms supporting fast reboot. For further information about using fast reboots, see Accelerating the Reboot Process in Booting and Shutting Down Oracle Solaris 11.3 Systems.
Deferred dumps are enabled by default on SPARC T4 systems that support at least Sys Firmware 8.7 or Sys Firmware 9.4 and later, and on Fujitsu M10 systems that support XCP 2290 or later with sufficient memory to preserve crash dump files until the system reboots.
Even with deferred dump, you should still ensure that the configured dump device has sufficient space to store crash dump information. This is in case a live system crash dump using savecore -L is necessary. Also, crash dump information will be automatically written to the dump device if possible without being held in memory. The writing operation occurs in the following situations:
The platform does not support a fast reboot, or the fast reboot fails.
The savecore utility is disabled or fails.
The system is using kernel zones.
For further information about preparing for and responding to system crashes, review this chapter and see the dumpadm(1M) man page.