OK, so a user loads fglrx or some other c++ kernel module. Now they have relocation sections with long names, including such beauties as 115 .gnu.linkonce.r._ZZN6AsicSII20ucode_class_pitcairnE24programApertureRegistersEjP14APERTURE_STATEE15UPPER_BOUND_GAP 116 .gnu.linkonce.t._ZN6AsicSII20ucode_class_pitcairnE23initializeApertureStateEP14APERTURE_STATEP18SURFACE_PROPERTIESm 136 .gnu.linkonce.t._ZNK25CMMHeap_FRAME_BUFFER_COREI18CMMHeap_ASIC_LOCALE20get_addr_restrictionER16MSF_SURF_ATTRIBSP21MEMHEAP_ADDR_RESTRICT 138 .gnu.linkonce.t._ZN18VMDomainManager_NI17CreateVmDomainsNII14VMDomainNI_PIO14VMDomainNI_PM417VMDomainNI_DRMDMAEEb22_VM_PROGRAMMING_METHOD 138 .gnu.linkonce.t._ZN18VMDomainManager_NI17CreateVmDomainsNII14VMDomainNI_PIO14VMDomainNI_PM417VMDomainSI_DRMDMAEEb22_VM_PROGRAMMING_METHOD for a glorious maximum "section name" length of 138. With bug #12612, we've once bumped up the symbol length limits, but not enough. Nothing is ever enough, when you have enough C++ templates! So we need some combination of relaxing limits more and/or keeping staprun from giving up so pessimistically in case of an overlong reloc name.
staprun promises to chill in the future. commit 0243596