[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 2/2] add rpc_versions for vm types

From: Flavio Cruz
Subject: Re: [PATCH 2/2] add rpc_versions for vm types
Date: Tue, 6 Dec 2022 00:33:08 -0500

On Tue, Dec 06, 2022 at 02:10:52AM +0100, Samuel Thibault wrote:
Hello Flavio,

What do you think of that?

I think we could just define vm_size_t as typedef of __mach_uintptr_t then it will be unsigned int on 32 bits and unsigned long on 64 bits.

Regarding mach_msg_type_number_t, do you know if there is a need to copy more than 4GB using RPCs? I guess it could be useful, but servers can just
paginate the data. For now, we could just change some of the signatures to
use mach_msg_type_number_t instead of vm_size_t to make the interface
explicit about the limits. If we think there is a need to go beyond 4GB we can make mach_msg_type_number_t an unsigned long on pure 64 bits.


Samuel Thibault, le lun. 29 août 2022 01:21:23 +0200, a ecrit:

Luca Dariz, le dim. 03 avril 2022 16:59:55 +0200, a ecrit:
> @@ -88,13 +94,36 @@ typedef unsigned long long rpc_phys_addr_t;
>   * expressing the difference between two
>   * vm_offset_t entities.
>   */
> -#ifdef __x86_64__
>  typedef   unsigned long   vm_size_t;
> -#else
> -typedef   natural_t       vm_size_t;
> -#endif
>  typedef   vm_size_t *     vm_size_array_t;

Mmm, this has triggered a lot of warnings & errors about mismatches
between vm_size_t and mach_msg_type_number_t in userland hurd&glibc.

While a lot of them are real mistakes (passing mach_msg_type_number_t*
for a vm_size_t* parameter), others are questioning, for instance:

routine io_read (
        io_object: io_t;
        out data: data_t, dealloc;
        offset: loff_t;
        amount: vm_size_t);

amount is a vm_size_t, thus unsigned long, but the data_t array will
have a mach_msg_type_number_t size. On a 64bit userland, vm_size_t being
64bit would be useless if mach_msg_type_number_t is 32bit. Perhaps we
want to make mach_msg_type_number_t long? Or use another type for
the size of data_t?

That's tricky questions ahead. Perhaps we can for now use

#if defined(KERNEL) || defined(__x86_64__)
typedef unsigned long   vm_size_t;
typedef natural_t       vm_size_t;

to avoid exposing the new type to userland for now, so as to have the
time to fix such questions before exposing the switch?


reply via email to

[Prev in Thread] Current Thread [Next in Thread]