Re: [PATCH 1/2] Fix for Haiku

[Thomas Huth <thuth@redhat.com>]

Meta comment: Please avoid
ƃuıʇsod uʍop ǝpısdn
... it makes threads much harder to read.

On 04/07/2021 16.20, Richard Zak wrote:
> Exactly. One of the developers on the Haiku forum shared with me the patch 
> that Haiku uses for qemu, and it has a few lines concerning error codes. 
> I'll look into this.
> https://github.com/haikuports/haikuports/blob/14c2cab5428145b93232cb69683a67bbe68a9f06/app-emulation/qemu/patches/qemu-3.1.1.1.patchset 

So something like the change in target/xtensa/xtensa-semi.c is the right way 
to go - using a switch-case statement looks like the right fix instead of 
disabling the targets in the configure script.

> This is a stopper for the configure script changes?

Well, if anyhow possible, we should avoid hacks like disabling a target in 
the configure script. Since this problem seems to be understood, you should 
aim for the right fix instead.

  Thomas

> În dum., 4 iul. 2021 la 05:30, Philippe Mathieu-Daudé <f4bug@amsat.org 
> <mailto:f4bug@amsat.org>> a scris:
> 
>     On 7/4/21 11:27 AM, Philippe Mathieu-Daudé wrote:
>      > On 7/4/21 11:03 AM, Philippe Mathieu-Daudé wrote:
>      >> On 7/4/21 12:06 AM, Richard Zak wrote:
>      >>> For MIPS (all sub-targets, 64-bit and EL) & xtensa(eb), the compiler
>      >>> complains about running out of memory. Best I can see, that's not what
>      >>> actually happens, but that's the error message. I was going to
>      >>> investigate this later, but this was the error which was causing the
>      >>> test with the Haiku VM with that corresponding make target. My
>     desktop &
>      >>> laptop have 64 GB, and I'm pretty sure it didn't get to that point.
>      >>>
>      >>>
>      >
>     /boot/system/develop/tools/bin/../lib/gcc/x86_64-unknown-haiku/8.3.0/../../../../x86_64-unknown-haiku/bin/ld:
>      >>> final link failed: memory exhausted
>      >
>      >> See how Haiku handles POSIX errno:
>      >>
>      >>
>     https://github.com/haiku/haiku/blob/master/headers/os/support/Errors.h
>     <https://github.com/haiku/haiku/blob/master/headers/os/support/Errors.h>
>      >>
>      >> #define B_GENERAL_ERROR_BASE   INT_MIN
>      >>
>      >> #define B_POSIX_ERROR_BASE     (B_GENERAL_ERROR_BASE + 0x7000)
>      >>
>      >> #define B_POSIX_ENOMEM  B_TO_POSIX_ERROR(B_POSIX_ERROR_BASE + 0)
>      >> #define E2BIG           B_TO_POSIX_ERROR(B_POSIX_ERROR_BASE + 1)
>      >> #define ECHILD          B_TO_POSIX_ERROR(B_POSIX_ERROR_BASE + 2)
>      >> ...
>      >>
>      >
>      > Same problem with Xtensa:
>      >
>      > static uint32_t errno_h2g(int host_errno)
>      > {
>      >     static const uint32_t guest_errno[] = {
>      >         [EPERM]         = TARGET_EPERM,
>      >         [ENOENT]        = TARGET_ENOENT,
>      >         [ESRCH]         = TARGET_ESRCH,
>      >         [EINTR]         = TARGET_EINTR,
>      >         [EIO]           = TARGET_EIO,
>      >         [ENXIO]         = TARGET_ENXIO,
>      >         [E2BIG]         = TARGET_E2BIG,
>      >         [ENOEXEC]       = TARGET_ENOEXEC,
>      > ...
> 
>     Annoyingly enough this is also how linux-user/syscall.c does
>     (thinking about code re-use):
> 
>     /*
>       * This list is the union of errno values overridden in asm-<arch>/errno.h
>       * minus the errnos that are not actually generic to all archs.
>       */
>     static uint16_t host_to_target_errno_table[ERRNO_TABLE_SIZE] = {
>          [EAGAIN]            = TARGET_EAGAIN,
>          [EIDRM]             = TARGET_EIDRM,
>          [ECHRNG]            = TARGET_ECHRNG,
>          [EL2NSYNC]          = TARGET_EL2NSYNC,
>          [EL3HLT]            = TARGET_EL3HLT,
>          [EL3RST]            = TARGET_EL3RST,
>     ...
> 
> 
> 
> -- 
> Regards,
> 
> Richard J. Zak
> Professional Genius
> PGP Key: https://keybase.io/rjzak/key.asc <https://keybase.io/rjzak/key.asc>