On Fri, 10 May 2019 at 15:05, Alex Bennée <alex.ben...@linaro.org> wrote: > Only for the first one though.. that said I'm sure the write string is > leaking when we do gdb output with whatever lock_user_string is trying > to achieve.
Yes, there looks like there's a leak there. (The fix is complicated because we need to check whether the string buffer is required to hang around until the asynchronous gdb operation is finished and the arm_semi_cb is invoked, or whether we can free it as soon as arm_gdb_syscall() returns.) lock_user_string is basically "give me a host pointer to the string at this address in guest memory": * on softmmu, the 'lock' operation copies the contents of guest memory into a local buffer, and 'unlock' then frees the buffer (possibly copying the updated local buffer contents back to the guest) * on linux-user, 'lock' does the guest-addr-to-host-addr conversion, and if DEBUG_REMAP is defined then it will also copy it into a separate buffer (and unlock will copy it back) thanks -- PMM
