On 2/20/24 06:06, Peter Maydell wrote:
Currently the qemu_register_reset() API permits the reset handler functions
registered with it to remove themselves from within the callback function.
This is fine with our current implementation, but is a bit odd, because
generally reset is supposed to be idempotent, and doesn't fit well in a
three-phase-reset world where a resettable object will get multiple
callbacks as the system is reset.
We now have only one user of qemu_register_reset() which makes use of
the ability to unregister itself within the callback:
restore_boot_order(). We want to change our implementation of
qemu_register_reset() to something where it would be awkward to
maintain the "can self-unregister" feature. Rather than making that
reimplementation complicated, change restore_boot_order() so that it
doesn't unregister itself but instead returns doing nothing for any
calls after it has done the "restore the boot order" work.
Signed-off-by: Peter Maydell<peter.mayd...@linaro.org>
---
It would be nicer not to use reset at all, especially since I'm not
a fan of conflating "system is reset" with "system boots", but I
didn't have a good idea for how to do that.
---
system/bootdevice.c | 25 ++++++++++++++-----------
1 file changed, 14 insertions(+), 11 deletions(-)
Acked-by: Richard Henderson <richard.hender...@linaro.org>
r~
