guenter geiger <[EMAIL PROTECTED]> writes: > I just wanted to suggest a possible solution to the problem: > > > > > So far, no one is sure why CAP_SYS_RESOURCE is needed, but we find > > > > that mlockall() fails without it. > > I am just wondering if mlockall(MCL_CURRENT) is possible without > CAP_SYS_RESOURCE, because its potentially less dangerous than MCL_FUTURE.
Are you suggesting that the need for CAP_SYS_RESOURCE is caused by the use of MCL_FUTURE? That seems plausible, although I see no sign of it in the 2.4.19 kernel sources for my system. The JACK server still has to allocate additional storage after the initial call to mlockall(). We either have to mlock() each block when it is allocated (error-prone), or use MCL_FUTURE. In either case, the same amount of memory gets locked. The only difference is whether it gets done reliably. -- joq
