proc: real CPU time, a tty number, and an exe that is a symlink
Four Gate 0 rows, and a fifth found while checking one of them.
**getrusage() did not distinguish RUSAGE_CHILDREN from RUSAGE_SELF**, which
undersells it: it ignored `who` entirely and reported the machine's *uptime*
as the caller's user time. Every process of every age answered with the same
number. Per-process CPU time is sampled at the timer tick now and split on the
interrupted privilege level; children's time is rolled up in wait() rather than
exit(), because RUSAGE_CHILDREN means *reaped* children and a running one has
not contributed yet.
**And wait4()'s fourth argument is a `struct rusage *` that was read off the
stack and thrown away.** That is the one shells actually use -- wait3() is the
same syscall with pid -1 -- so zsh's `time` reported
16777216.01s user 1073.78s system 927983249% cpu
for every command. 2^24 seconds is uninitialised memory, and it did not vary
with the workload because it was never written. Now:
busy loop 2.66s user 97% cpu 2.720 total
sleep 2 0.02s user 0% cpu 2.248 total
ls -lR / 0.70s user 65% cpu 1.070 total
**ru_stime is a structural zero and says so.** MSR_SFMASK includes FL_IF, so
SYSCALL clears interrupts and no tick can land while a process is inside a
syscall. Tick sampling cannot see kernel time at all here. Charging it needs
TSC deltas closed and reopened around every sched(), which is scheduler
accounting rather than sampling; the limitation is written down in
docs/architecture/kernel.md rather than papered over.
**/proc/<pid>/exe is a symlink.** It also needed a real st_size, and every
other procfs file deliberately reports zero -- readlink(2) bounds its copy by
it, so a symlink of size 0 resolves to the empty string, which looks like it
worked and is worse than the regular file was. Generating that one path in
ilock() costs one small render per stat() of one file.
**OVMF memory reporting is closed**: 1 GB reports 1017 MB usable, 2 GB reports
2041 MB. The firmware keeps about 7 MB either way, which is the right shape.
**And `ps` printed a header and nothing else**, which is not in any list
because nobody had run it. `ps` with no arguments shows only processes on
*this* terminal; /proc/<pid>/stat hardcoded tty_nr to 0, so nothing could ever
match. `ps -e` always worked, which is exactly what hid it. p->ctty was already
tracked -- procfs simply was not asked. Fields 14..17 were hardcoded zero too,
so TIME read 00:00:00 however long a process had run; they carry the new
counters now, and a shell that has burned 653 ticks reports 00:00:06.
Verified: utime climbs 18 -> 653 across a loop in the shell itself, `ps` lists
its own terminal, `ps -e` still lists everything, `readlink /proc/1/exe` gives
/bin/schinit, and sigtest/$()/rename/dmesg are all still green.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Nexoniarz
parent
7e88cea151
commit
a5ff83de46
8 changed files with 199 additions and 26 deletions
|
|
@ -237,6 +237,14 @@ allocation over 8 MB mapped straight into a neighbouring live mapping.
|
|||
|
||||
## What it does not do
|
||||
|
||||
- **`ru_stime` is effectively always zero.** CPU time is sampled at the timer
|
||||
tick and split on the interrupted privilege level, but `MSR_SFMASK` includes
|
||||
`FL_IF`, so `SYSCALL` clears interrupts and no tick can land while a process
|
||||
is inside a syscall. User time is real and discriminates properly — a busy
|
||||
loop reports 97%, `sleep 2` reports 0% — but kernel time needs TSC deltas
|
||||
closed and reopened around every `sched()`, which is scheduler accounting
|
||||
rather than sampling, and is not written.
|
||||
|
||||
- **No slab allocator.** One page at a time, and every subsystem's fixed-size
|
||||
table is a consequence. This is the single change that would most improve the
|
||||
rest of the kernel.
|
||||
|
|
|
|||
|
|
@ -215,12 +215,33 @@ right then.
|
|||
cross-compile cache. Autoconf's own default for that check is `yes`;
|
||||
omitting the line would have been right. See
|
||||
[../history/kernel.md](../history/kernel.md), section 7.
|
||||
- [ ] **`getrusage` does not distinguish `RUSAGE_CHILDREN`** from `RUSAGE_SELF`.
|
||||
- [ ] **`/proc/<pid>/exe` is a regular file, not a symlink**, so `readlink()`
|
||||
fails and `ps -o comm` prints `[init]`. Symlinks themselves now work —
|
||||
`open()` and `stat()` follow them, which they never did — so what is left
|
||||
is procfs growing a symlink node type, not symlink support.
|
||||
- [ ] **OVMF memory reporting**, if the region-list fix did not fully close it.
|
||||
- [x] ~~**`getrusage` does not distinguish `RUSAGE_CHILDREN`** from
|
||||
`RUSAGE_SELF`.~~ It did not distinguish anything: it ignored `who` and
|
||||
reported the machine's *uptime* as the caller's user time, so every
|
||||
process of every age answered with the same number. Per-process CPU time
|
||||
is sampled at the timer tick now, children's time is rolled up on reap,
|
||||
and `wait4()`'s fourth argument — a `struct rusage *` that was read off
|
||||
the stack and thrown away — is filled. That last one is why zsh's `time`
|
||||
reported `16777216.01s user` for everything: 2^24 seconds is
|
||||
uninitialised memory, and it never varied because it was never written.
|
||||
`time` now says 97% cpu for a busy loop and 0% for `sleep 2`.
|
||||
`ru_stime` is a known zero; see
|
||||
[../architecture/kernel.md](../architecture/kernel.md).
|
||||
- [x] ~~**`/proc/<pid>/exe` is a regular file, not a symlink.**~~
|
||||
`lrwxrwxrwx ... exe -> /bin/schinit`, and `readlink` returns it. It also
|
||||
needed a real `st_size`, which every other procfs file deliberately
|
||||
reports as zero: `readlink(2)` bounds its copy by it, so a symlink of
|
||||
size 0 resolves to the empty string — which looks like it worked and is
|
||||
worse than the regular file was.
|
||||
- [x] ~~**OVMF memory reporting.**~~ The region-list fix did close it. 1 GB
|
||||
reports 1017 MB usable, 2 GB reports 2041 MB, which is the right shape:
|
||||
the firmware keeps about 7 MB either way.
|
||||
- [x] ~~**`ps` printed a header and nothing else.**~~ Found while checking the
|
||||
`exe` fix. `ps` with no arguments shows only processes on *this*
|
||||
terminal, and `/proc/<pid>/stat` hardcoded `tty_nr` to 0, so nothing
|
||||
could ever match — `ps -e` always worked, which is what hid it. Fields
|
||||
14-17 (utime/stime/cutime/cstime) were hardcoded zero too, so `TIME` was
|
||||
`00:00:00` for everything however long it had run.
|
||||
|
||||
**Work already done and not connected:**
|
||||
|
||||
|
|
|
|||
|
|
@ -536,7 +536,7 @@ void scheduler(void) __attribute__((noreturn));
|
|||
void sched(void);
|
||||
void sleep(void*, struct spinlock*);
|
||||
int wait(void);
|
||||
int waitpid(int, int*, int);
|
||||
int waitpid(int, int*, int, uint64*, uint64*);
|
||||
void wakeup(void*);
|
||||
void yield(void);
|
||||
void sinit(void);
|
||||
|
|
|
|||
|
|
@ -166,6 +166,9 @@ struct procinfo {
|
|||
int state; // enum procstate
|
||||
uint64 sz;
|
||||
int killed;
|
||||
int ctty; // controlling terminal, 1-based; 0 = none
|
||||
uint64 utime, stime; // CPU time in ticks, this process
|
||||
uint64 cutime, cstime; // and its reaped children
|
||||
char name[16];
|
||||
char exe[64]; // full path, for /proc/<pid>/exe
|
||||
};
|
||||
|
|
@ -213,6 +216,16 @@ struct proc {
|
|||
struct ioport_range ioport[NIOPORT];
|
||||
int nioport;
|
||||
|
||||
// CPU time, in timer ticks, charged by trap() at each tick to whoever was
|
||||
// running. utime/stime split on the interrupted privilege level -- cs & 3 --
|
||||
// which is the only information the tick carries about what it interrupted.
|
||||
//
|
||||
// cutime/cstime accumulate *reaped* children, and only reaped ones: that is
|
||||
// what RUSAGE_CHILDREN means, and it is why they are rolled up in wait()
|
||||
// rather than in exit(). A child still running has not contributed yet.
|
||||
uint64 utime, stime;
|
||||
uint64 cutime, cstime;
|
||||
|
||||
uint64 signal_mask;
|
||||
uint64 pending_signals;
|
||||
int alarm_ticks;
|
||||
|
|
|
|||
|
|
@ -90,6 +90,26 @@ trap(struct trapframe *tf)
|
|||
wakeup(&ticks);
|
||||
release(&tickslock);
|
||||
}
|
||||
// Charge the tick to whoever it interrupted. Every CPU does this, not just
|
||||
// CPU 0 -- the global clock advances once, but the *work* happened on
|
||||
// whichever processor was running that process.
|
||||
//
|
||||
// The split is by the interrupted privilege level, because that is the
|
||||
// only thing the tick knows about what it interrupted. It is sampling, not
|
||||
// measurement, which is what Linux does too and why these are ticks rather
|
||||
// than microseconds.
|
||||
//
|
||||
// **stime is therefore almost always zero here, and that is structural.**
|
||||
// MSR_SFMASK includes FL_IF (see syscallinit), so SYSCALL clears interrupts
|
||||
// and nothing re-enables them for the duration -- a timer tick cannot land
|
||||
// while a process is executing a syscall. The only ticks that ever reach
|
||||
// the `else` are ones that arrive in a nested trap. Charging real system
|
||||
// time needs TSC deltas closed and reopened around every sched(), which is
|
||||
// scheduler accounting rather than sampling, and is not written.
|
||||
if(myproc()){
|
||||
if((tf->cs & 3) == DPL_USER) myproc()->utime++;
|
||||
else myproc()->stime++;
|
||||
}
|
||||
// Debug registers are per-CPU, and there is no general IPI here, so this
|
||||
// is how a watchpoint armed on one processor reaches the others: each
|
||||
// notices the table's generation has moved and reloads. One comparison
|
||||
|
|
|
|||
|
|
@ -175,15 +175,29 @@ gen_pid_stat(struct sb *s, const struct procinfo *pi)
|
|||
sbputi(s, pi->pgid); // 5 pgrp
|
||||
sbputc(s, ' ');
|
||||
sbputi(s, pi->sid); // 6 session
|
||||
sbputs(s, " 0 -1 0"); // 7 tty_nr, 8 tpgid, 9 flags
|
||||
// 7 tty_nr. Linux's dev_t: major 12 bits above minor's 8. /dev/ttyN is
|
||||
// major 1, minor 10+N in devfs, so a process whose ctty is 1 reports
|
||||
// (1 << 8) | 11. This was hardcoded 0, and that is why plain `ps` printed a
|
||||
// header and nothing else: ps with no arguments shows only the processes on
|
||||
// *this* terminal, and nothing could ever match. `ps -e` always worked.
|
||||
sbputc(s, ' ');
|
||||
sbputi(s, pi->ctty ? ((1 << 8) | (10 + pi->ctty)) : 0);
|
||||
sbputs(s, " -1 0"); // 8 tpgid, 9 flags
|
||||
|
||||
// 10..17: minflt cminflt majflt cmajflt utime stime cutime cstime.
|
||||
// No per-process accounting yet, so these are honestly zero rather than
|
||||
// invented -- a fabricated CPU time is worse than none, because top would
|
||||
// draw a plausible graph of a number nobody measured.
|
||||
for (i = 0; i < 8; i++)
|
||||
// 10..13: minflt cminflt majflt cmajflt. There is no fault accounting, and
|
||||
// a fabricated number is worse than none -- top would draw a plausible graph
|
||||
// of something nobody measured.
|
||||
for (i = 0; i < 4; i++)
|
||||
sbputs(s, " 0");
|
||||
|
||||
// 14..17: utime stime cutime cstime, in ticks, which is the unit the field
|
||||
// is defined in. These were zero too, which is why `ps` and `top` showed
|
||||
// 00:00:00 for everything no matter how long it had been running.
|
||||
sbputc(s, ' '); sbputi(s, (int64)pi->utime);
|
||||
sbputc(s, ' '); sbputi(s, (int64)pi->stime);
|
||||
sbputc(s, ' '); sbputi(s, (int64)pi->cutime);
|
||||
sbputc(s, ' '); sbputi(s, (int64)pi->cstime);
|
||||
|
||||
sbputs(s, " 20 0 1 0 0 "); // priority nice num_threads itrealvalue starttime
|
||||
sbputi(s, (int64)pi->sz); // 23 vsize, in bytes
|
||||
sbputc(s, ' ');
|
||||
|
|
@ -278,6 +292,27 @@ procfs_iload(struct inode *ip)
|
|||
return;
|
||||
}
|
||||
|
||||
// /proc/<pid>/exe is a symlink, as it is on Linux.
|
||||
//
|
||||
// It was a regular file holding the path as text, which reads the same from
|
||||
// `cat` and is wrong to everything that matters: readlink() refused it with
|
||||
// EINVAL, and `ps -o comm` -- which readlinks it -- fell back to the process
|
||||
// name in brackets and printed `[init]` for everything.
|
||||
//
|
||||
// This one file has to report a real size, and every other procfs file
|
||||
// deliberately reports zero. readlink(2) bounds its copy by ip->size, so a
|
||||
// symlink whose size is 0 resolves to the empty string -- which is worse
|
||||
// than the regular file was, because it looks like it worked. Generating it
|
||||
// here costs one small render per stat() of one file.
|
||||
if (!INO_IS_GLOBAL(ino) && INO_FILE(ino) == PF_EXE) {
|
||||
char path[64];
|
||||
int len = gen_file(ino, path, sizeof(path));
|
||||
ip->type = T_SYMLINK;
|
||||
ip->size = len > 0 ? (uint64)len : 0;
|
||||
ip->nlink = 1;
|
||||
return;
|
||||
}
|
||||
|
||||
ip->type = T_FILE;
|
||||
// Size is reported as zero.
|
||||
//
|
||||
|
|
|
|||
|
|
@ -310,6 +310,7 @@ found:
|
|||
memset(p->sig_handler, 0, sizeof(p->sig_handler));
|
||||
memset(p->sig_restorer, 0, sizeof(p->sig_restorer));
|
||||
p->pending_signals = 0;
|
||||
p->utime = p->stime = p->cutime = p->cstime = 0;
|
||||
p->signal_mask = 0;
|
||||
p->in_signal = 0;
|
||||
p->sigsusp_oldmask = 0;
|
||||
|
|
@ -659,7 +660,7 @@ exit(int status)
|
|||
int
|
||||
wait(void)
|
||||
{
|
||||
return waitpid(-1, 0, 0);
|
||||
return waitpid(-1, 0, 0, 0, 0);
|
||||
}
|
||||
|
||||
// waitpid(2)/wait4(2) proper: a specific pid (-1 = any child, matching
|
||||
|
|
@ -676,7 +677,7 @@ wait(void)
|
|||
// this kernel can't wake for a real SIGCHLD (no signal delivery), so it
|
||||
// silently retried forever: a real, low-CPU hang with no error printed.
|
||||
int
|
||||
waitpid(int pid, int *status, int options)
|
||||
waitpid(int pid, int *status, int options, uint64 *ut, uint64 *st)
|
||||
{
|
||||
struct proc *p;
|
||||
int havekids;
|
||||
|
|
@ -699,6 +700,16 @@ waitpid(int pid, int *status, int options)
|
|||
if(p->state == ZOMBIE){
|
||||
int rpid = p->pid;
|
||||
int rstatus = p->xstate; // copy out before the slot is recycled
|
||||
// RUSAGE_CHILDREN: the child's own time *and* whatever it had already
|
||||
// reaped, which is what makes the total transitive down a whole tree
|
||||
// of processes rather than one generation deep.
|
||||
myproc()->cutime += p->utime + p->cutime;
|
||||
myproc()->cstime += p->stime + p->cstime;
|
||||
// wait4()'s fourth argument wants the *child's* usage, not the
|
||||
// parent's running total, so it has to be read here -- three lines
|
||||
// further down the slot is UNUSED and these are zero.
|
||||
if (ut) *ut = p->utime + p->cutime;
|
||||
if (st) *st = p->stime + p->cstime;
|
||||
p->kstack = 0; // the slot keeps its stack; see kstacks[]
|
||||
// Give back anything shared before freevm() decides it owns the
|
||||
// pages. See proc_release_shared() in kernel/sys/sys_mem.c.
|
||||
|
|
@ -1317,6 +1328,11 @@ proc_snapshot(int idx, struct procinfo *out)
|
|||
out->state = p->state;
|
||||
out->sz = p->sz;
|
||||
out->killed = p->killed;
|
||||
out->ctty = p->ctty;
|
||||
out->utime = p->utime;
|
||||
out->stime = p->stime;
|
||||
out->cutime = p->cutime;
|
||||
out->cstime = p->cstime;
|
||||
safestrcpy(out->name, p->name, sizeof(out->name));
|
||||
safestrcpy(out->exe, p->exe, sizeof(out->exe));
|
||||
release(&ptable.lock);
|
||||
|
|
|
|||
|
|
@ -52,6 +52,9 @@ sys_kill(void)
|
|||
return sigsend(pid, sig) < 0 ? -ESRCH : 0;
|
||||
}
|
||||
|
||||
// Defined with getrusage() further down; wait4() needs it too.
|
||||
static void rusage_fill(struct rusage *ru, uint64 ut, uint64 st);
|
||||
|
||||
addr_t sys_getpid(void) { return myproc()->pid; }
|
||||
addr_t sys_getppid(void) { return myproc()->parent ? myproc()->parent->pid : 1; }
|
||||
addr_t sys_gettid(void) { return myproc()->pid; } // single-threaded: tid == pid
|
||||
|
|
@ -67,26 +70,45 @@ sys_waitpid(void)
|
|||
if (argint(0, &pid) < 0 || argaddr(1, &statusp) < 0 || argint(2, &options) < 0)
|
||||
return -1;
|
||||
int status;
|
||||
int wpid = waitpid(pid, &status, options);
|
||||
int wpid = waitpid(pid, &status, options, 0, 0);
|
||||
if (wpid > 0 && statusp)
|
||||
put_user(status, statusp);
|
||||
return wpid;
|
||||
}
|
||||
|
||||
// wait4(2) -- and its fourth argument is a `struct rusage *`.
|
||||
//
|
||||
// It was read off the stack and thrown away. So a caller that passed a buffer
|
||||
// got it back untouched, and read whatever had been there before: zsh's
|
||||
// `time` reported 16777216.01s user for every command, which is 2^24 seconds,
|
||||
// which is uninitialised memory that happened to look like a number. The
|
||||
// figure did not vary with the workload because it was never written.
|
||||
//
|
||||
// wait3() is the same syscall with pid -1, and it is the one shells actually
|
||||
// call, so this is the path every `time` goes through.
|
||||
addr_t
|
||||
sys_wait4(void)
|
||||
{
|
||||
int pid, options;
|
||||
addr_t statusp;
|
||||
addr_t statusp, rup;
|
||||
uint64 ut = 0, st = 0;
|
||||
|
||||
argint(0, &pid);
|
||||
argaddr(1, &statusp);
|
||||
argint(2, &options);
|
||||
argaddr(3, &rup);
|
||||
|
||||
int status;
|
||||
int wpid = waitpid(pid, &status, options);
|
||||
int wpid = waitpid(pid, &status, options, &ut, &st);
|
||||
if (wpid <= 0)
|
||||
return wpid;
|
||||
if (statusp)
|
||||
put_user(status, statusp);
|
||||
if (rup) {
|
||||
struct rusage *ru = (struct rusage*)user_ptr(rup, sizeof(*ru));
|
||||
if (ru)
|
||||
rusage_fill(ru, ut, st);
|
||||
}
|
||||
return wpid;
|
||||
}
|
||||
|
||||
|
|
@ -385,27 +407,65 @@ addr_t sys_pause(void)
|
|||
return -EINTR;
|
||||
}
|
||||
|
||||
// getrusage(2).
|
||||
//
|
||||
// This used to ignore `who` entirely and report the machine's *uptime* as the
|
||||
// caller's user time. Every process of every age answered with the same
|
||||
// number, and RUSAGE_CHILDREN answered with it too -- so a shell timing a
|
||||
// command got the age of the machine, and `time` on a pipeline was
|
||||
// meaningless rather than merely imprecise.
|
||||
//
|
||||
// Ticks, converted at the boundary. The kernel counts whole timer ticks
|
||||
// because that is what the sampling gives it; microseconds are what the
|
||||
// interface asks for, so the multiply happens here rather than pretending to a
|
||||
// resolution the counter does not have.
|
||||
static void
|
||||
rusage_fill(struct rusage *ru, uint64 ut, uint64 st)
|
||||
{
|
||||
memset(ru, 0, sizeof(*ru));
|
||||
ru->ru_utime.tv_sec = (long)(ut / HZ);
|
||||
ru->ru_utime.tv_usec = (long)((ut % HZ) * (1000000 / HZ));
|
||||
ru->ru_stime.tv_sec = (long)(st / HZ);
|
||||
ru->ru_stime.tv_usec = (long)((st % HZ) * (1000000 / HZ));
|
||||
}
|
||||
|
||||
addr_t sys_getrusage(void)
|
||||
{
|
||||
int who;
|
||||
struct rusage *ru;
|
||||
argint(0, &who);
|
||||
if (argptr(1, (char**)&ru, sizeof(*ru)) < 0) return -1;
|
||||
memset(ru, 0, sizeof(*ru));
|
||||
uint64 rsec, rusec;
|
||||
monotonic(&rsec, &rusec);
|
||||
ru->ru_utime.tv_sec = (long)rsec;
|
||||
ru->ru_utime.tv_usec = (long)rusec;
|
||||
return 0;
|
||||
if (argptr(1, (char**)&ru, sizeof(*ru)) < 0) return -EFAULT;
|
||||
|
||||
switch (who) {
|
||||
case RUSAGE_SELF:
|
||||
rusage_fill(ru, myproc()->utime, myproc()->stime);
|
||||
return 0;
|
||||
case RUSAGE_CHILDREN:
|
||||
rusage_fill(ru, myproc()->cutime, myproc()->cstime);
|
||||
return 0;
|
||||
default:
|
||||
// RUSAGE_THREAD is -1 and would be RUSAGE_SELF here, since a process
|
||||
// has exactly one thread. It is refused rather than aliased: answering
|
||||
// it would imply threads exist.
|
||||
return -EINVAL;
|
||||
}
|
||||
}
|
||||
|
||||
addr_t sys_times(void)
|
||||
{
|
||||
addr_t bufaddr;
|
||||
argaddr(0, &bufaddr);
|
||||
// The tms fields are in clock ticks, which is what the kernel counts, so
|
||||
// this one needs no conversion -- USER_HZ and HZ are both 100. They used
|
||||
// to be memset to zero, which is why `times()` agreed with nothing.
|
||||
if (bufaddr) {
|
||||
struct tms *t = (struct tms*)user_ptr(bufaddr, sizeof(*t));
|
||||
if (t) memset(t, 0, sizeof(*t));
|
||||
if (t) {
|
||||
t->tms_utime = (long)myproc()->utime;
|
||||
t->tms_stime = (long)myproc()->stime;
|
||||
t->tms_cutime = (long)myproc()->cutime;
|
||||
t->tms_cstime = (long)myproc()->cstime;
|
||||
}
|
||||
}
|
||||
// USER_HZ units, measured from the TSC.
|
||||
//
|
||||
|
|
|
|||
Loading…
Reference in a new issue