diff options
Diffstat (limited to 'TRICKS')
| -rw-r--r-- | TRICKS | 26 |
1 files changed, 13 insertions, 13 deletions
@@ -4,6 +4,9 @@ might be worth pointing out in a longer explanation or in class. --- +[2009/07/12: No longer relevant; forkret1 changed +and this is now cleaner.] + forkret1 in trapasm.S is called with a tf argument. In order to use it, forkret1 copies the tf pointer into %esp and then jumps to trapret, which pops the @@ -45,21 +48,21 @@ always. There is a (harmless) race in pushcli, which does - eflags = read_eflags(); + eflags = readeflags(); cli(); - if(cpus[cpu()].ncli++ == 0) - cpus[cpu()].intena = eflags & FL_IF; + if(c->ncli++ == 0) + c->intena = eflags & FL_IF; Consider a bottom-level pushcli. If interrupts are disabled already, then the right thing happens: read_eflags finds that FL_IF is not set, -and intena = 1. If interrupts are enabled, then +and intena = 0. If interrupts are enabled, then it is less clear that the right thing happens: -the read_eflags can execute, then the process +the readeflags can execute, then the process can get preempted and rescheduled on another cpu, and then once it starts running, perhaps with interrupts disabled (can happen since the scheduler -only disables interrupts once per scheduling loop, +only enables interrupts once per scheduling loop, not every time it schedules a process), it will incorrectly record that interrupts *were* enabled. This doesn't matter, because if it was safe to be @@ -112,17 +115,13 @@ processors will need it. --- -The code in sys_fork needs to read np->pid before +The code in fork needs to read np->pid before setting np->state to RUNNABLE. int - sys_fork(void) + fork(void) { - int pid; - struct proc *np; - - if((np = copyproc(cp)) == 0) - return -1; + ... pid = np->pid; np->state = RUNNABLE; return pid; @@ -134,3 +133,4 @@ get reused for a different process (with a new pid), all before the return statement. So it's not safe to just do "return np->pid;". +This works because proc.h marks the pid as volatile. |