Samuel Thibault, le mer. 01 nov. 2023 01:50:40 +0100, a ecrit:
> Samuel Thibault, le mar. 31 oct. 2023 04:40:43 +0100, a ecrit:
> > Samuel Thibault, le lun. 30 oct. 2023 18:35:03 +0100, a ecrit:
> > > Samuel Thibault, le dim. 29 oct. 2023 23:27:22 +0100, a ecrit:
> > > > Samuel Thibault, le ven. 27 oct. 2023 08:48:19 +0200, a ecrit:
> > > > > while [ "$(echo -n `echo internal/reflectlite.s-gox | sed -e 's/s-gox/gox/' ` )" = internal/reflectlite.gox ] ; do : ; done
> > > >
> > > > For now, I could reproduce with
> > > >
> > > > time while [ "$(echo -n `echo a` )" = a ] ; do : ; done
> > > >
> > > > by running two of them in parallel, along with an apt install loop in
> > > > parallel. It takes a few hours to reproduce (sometimes 1, sometimes
> > > > 3...)
> > >
> > > It seems to happen more often when running inside a chroot (possibly
> > > because of the intermediate firmlink redirection?), and possibly
> > > eatmydata also makes it more frequent.
> >
> > (it looks like there are memory leaks in proc, its vminfo keeps
> > increasing).
>
> It seems 64bit-specific: the program below makes proc leak memory, 100
> vminfo lines at a time. Possibly __mach_msg_destroy doesn't actually
> properly parse messages to be destroyed, so that in the error case the
> server leaks non-inline data? Flavio, perhaps you have an idea?
Realizing only now by reading the __mach_msg_destroy assembly...
unsigned int msgt_inline : 1,
msgt_longform : 1,
msgt_deallocate : 1,
msgt_name : 8,
msgt_size : 16,
msgt_unused : 5;
This field ordering makes reading them overly complex... It'll be a pain
to rebootstrap, but perhaps we do want to put msgt_size and msgt_name
first?
I only moved msgt_size to the end of the struct since a small number of (deprecated) RPC types
require msgt_size to be 2 bytes long and those can be shortened to be under 1 byte.
We could end up with a much larger contiguous msgt_unused that could be used for
other things in the future.
In relation to other changes we have to do for finishing the ABI... I think we have a reasonable
ABI now. However, I wonder if it would be simpler to just ask the user land to pass port names using
the following struct:
#ifdef KERNEL
union mach_rpc_port {
mach_port_name_t name;
mach_port_t kernel_port;
};
#else
struct mach_rpc_port {
mach_port_name_t name;
int unused;
};
#endif
It would make the kernel simpler since no message resizing is necessary and most of the code using this
would be MiG generated.
Flavio
Samuel