On Mon, 2018-07-30 at 15:17 -0400, Theodore Y. Ts'o wrote:
+AD4- I've been looking at what's the best way to make sure everything gets
+AD4- cleanly flushed out to disk on a powerdown. Right now in
+AD4- +AF8AXw-orderly+AF8-poweroff(), we call emergency+AF8-sync() which kicks a workqueue
+AD4- to flush all file systems and block devices --- and then we
+AD4- immediately power down the system, before the scheduler even has a
+AD4- chance to schedule the workqueue thread. Hopefully userspace has the
+AD4- unmounted all file systems, which will has implicitly issued a cache
+AD4- flush command, but if we have a userspace program writing to a block
+AD4- device directly, currently there's nothing to make sure things will
+AD4- get flushed out to the device.
+AD4-
+AD4- Beyond that, though, I'm interested in figuring out how to make sure
+AD4- that all SCSI devices will receive (and acknowledge) SHUTDOWN command
+AD4- so that the disks can be spun down and heads retracted to a safe
+AD4- landing zone before we power down the system.
+AD4-
+AD4- It appears the best way to do this is to call sd+AF8-shutdown(), since we
+AD4- don't seem to have a high-level +ACI-shutdown+ACI- concept recognized in the
+AD4- block layer (the way we currently, have, say support for +ACI-discard+ACI-).
+AD4-
+AD4- So the question is, what's the best way to architect something like
+AD4- this. I could implement a hacky interator loop in the SCSI subsystem,
+AD4- and call it directly from +AF8AXw-orderly+AF8-poweroff in kernel/reboot.c. But
+AD4- I'm pretty sure that would never get accepted upstream, and so it
+AD4- would remain a Google data center hack.
+AD4-
+AD4- What do people think would be the best way of implementing something
+AD4- that would be upstream acceptable?
Hi Ted,
Isn't that behavior a bug in +AF8AXw-orderly+AF8-poweroff()? My understanding is that
+AF8AXw-orderly+AF8-poweroff() calls run+AF8-cmd(poweroff+AF8-cmd). If the poweroff command
gets the chance to run then it will execute the reboot() system call to power
off the system. The reboot() system call then calls kernel+AF8-power+AF8-off(). That
last function then calls device+AF8-shutdown(). device+AF8-shutdown() calls the
.shutdown() method for all known devices. As you know the sd driver
implements that method. More in detail:
architecture code or ACPI code
-+AD4- orderly+AF8-poweroff()
-+AD4- poweroff+AF8-work+AF8-func()
-+AD4- +AF8AXw-orderly+AF8-poweroff()
-+AD4- run+AF8-cmd(poweroff+AF8-cmd)
-+AD4- call+AF8-usermodehelper(+ACI-/sbin/poweroff+ACI-)
>From the systemd source file src/core/shutdown.c:
(void) reboot(RB+AF8-POWER+AF8-OFF)+ADs-
>From kernel/reboot.c:
SYSCALL+AF8-DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
void +AF8AXw-user +ACo-, arg)
+AHs-
...
case LINUX+AF8-REBOOT+AF8-CMD+AF8-POWER+AF8-OFF:
kernel+AF8-power+AF8-off()+ADs-
do+AF8-exit(0)+ADs-
break+ADs-
...
+AH0-
void kernel+AF8-power+AF8-off(void)
+AHs-
kernel+AF8-shutdown+AF8-prepare(SYSTEM+AF8-POWER+AF8-OFF)+ADs-
if (pm+AF8-power+AF8-off+AF8-prepare)
pm+AF8-power+AF8-off+AF8-prepare()+ADs-
migrate+AF8-to+AF8-reboot+AF8-cpu()+ADs-
syscore+AF8-shutdown()+ADs-
pr+AF8-emerg(+ACI-Power down+AFw-n+ACI-)+ADs-
kmsg+AF8-dump(KMSG+AF8-DUMP+AF8-POWEROFF)+ADs-
machine+AF8-power+AF8-off()+ADs-
+AH0-
static void kernel+AF8-shutdown+AF8-prepare(enum system+AF8-states state)
+AHs-
blocking+AF8-notifier+AF8-call+AF8-chain(+ACY-reboot+AF8-notifier+AF8-list,
(state +AD0APQ- SYSTEM+AF8-HALT) ? SYS+AF8-HALT : SYS+AF8-POWER+AF8-OFF, NULL)+ADs-
system+AF8-state +AD0- state+ADs-
usermodehelper+AF8-disable()+ADs-
device+AF8-shutdown()+ADs-
+AH0-
>From drivers/base/core.c:
void device+AF8-shutdown(void)
+AHs-
...
if (dev-+AD4-class +ACYAJg- dev-+AD4-class-+AD4-shutdown+AF8-pre) +AHs-
if (initcall+AF8-debug)
dev+AF8-info(dev, +ACI-shutdown+AF8-pre+AFw-n+ACI-)+ADs-
dev-+AD4-class-+AD4-shutdown+AF8-pre(dev)+ADs-
+AH0-
if (dev-+AD4-bus +ACYAJg- dev-+AD4-bus-+AD4-shutdown) +AHs-
if (initcall+AF8-debug)
dev+AF8-info(dev, +ACI-shutdown+AFw-n+ACI-)+ADs-
dev-+AD4-bus-+AD4-shutdown(dev)+ADs-
+AH0- else if (dev-+AD4-driver +ACYAJg- dev-+AD4-driver-+AD4-shutdown) +AHs-
if (initcall+AF8-debug)
dev+AF8-info(dev, +ACI-shutdown+AFw-n+ACI-)+ADs-
dev-+AD4-driver-+AD4-shutdown(dev)+ADs-
+AH0-
...
+AH0-
Bart.
