Build on arm64

[andrewbird]

I have an arm64 (raspberry pi5 8GB + SSD) that I'd like to see if I can get dosemu2/fdpp running on. First off I'm trying to build fdpp and so I'm installing the prerequisites, however I don't seem to be able to find nasm-segelf

Broken fdpp-build-deps:arm64 Depends on nasm-segelf:arm64 < none @un H >
  Removing fdpp-build-deps:arm64 because I can't find nasm-segelf:arm64
Done

Is this likely to be the first of such missing packages, in that they can't be made available for arm64, or should I persevere?

[stsp]

https://code.launchpad.net/~stsp-0/+archive/ubuntu/nasm-segelf/+packages
Its here and of course the arm64 arch
is supported.

[stsp]

So should now be fixed.

[andrewbird]

a new one, but I think it may have got further last time.

Dependency Extension: org.freedesktop.Sdk.Extension.toolchain-i386 24.08
Installing org.freedesktop.Sdk.Extension.toolchain-i386/aarch64/24.08 from flathub

Note that the directories 

'/var/lib/flatpak/exports/share'
'/home/ajb/.local/share/flatpak/exports/share'

are not in the search path set by the XDG_DATA_DIRS environment variable, so
applications installed by Flatpak may not appear on your desktop until the
session is restarted.

error: Nothing matches org.freedesktop.Sdk.Extension.toolchain-i386 in remote flathub
Error installing deps: running `flatpak --user install -y --noninteractive flathub org.freedesktop.Sdk.Extension.toolchain-i386/aarch64/24.08`: Child process exited with code 1
make: *** [Makefile:113: flatpak-build] Error 1
~~

[stsp]

Please do flatpak remote-ls --user flathub |grep toolchain

[stsp]

Ah forget it, I can do
flatpak --arch=aarch64 remote-ls --user
to see what you see.

[stsp]

So they do not provide the
toolchain-i386 for arm currently.
I'll ask at their github what to do.

[andrewbird]

Sorry, I stepped away from the test machine for a while.On 7 Dec 2024 9:09 pm, Stas Sergeev ***@***.***> wrote: So they do not provide the toolchain-i386 for arm currently. I'll ask at their github what to do. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>

[stsp]

OK and I don't think this can be
solved quickly anyway. :(

[stsp]

Should now be fixed, please re-check.

[andrewbird]

The guy suggested this diff to the test case and with the default 16kb kernel it still works.

ajb@raspberrypi:/local/src/dosemu2.git $ diff -urN mproc-orig.c mprot.c
--- mproc-orig.c	2024-12-10 13:30:13.310443983 +0000
+++ mprot.c	2024-12-10 16:18:14.508283049 +0000
@@ -3,20 +3,22 @@
 #include <unistd.h>
 #include <sys/mman.h>
 
+#define PAGE 16384
+
 int main(void)
 {
     int err;
     char buf[1024];
     char *addr2;
-    char *addr = mmap(NULL, 4096*3, PROT_READ | PROT_WRITE,
+    char *addr = mmap(NULL, PAGE*3, PROT_READ | PROT_WRITE,
             MAP_SHARED | MAP_ANONYMOUS, -1, 0);
     if (addr == MAP_FAILED) {
         perror("mmap()");
         return 1;
     }
     printf("mapped %p\n", addr);
-    addr2 = addr + 4096;
-    err = mprotect(addr2, 4096, PROT_NONE);
+    addr2 = addr + PAGE;
+    err = mprotect(addr2, PAGE, PROT_NONE);
     if (err) {
         perror("mprotect()");
         return 1;

[andrewbird]

it also works with int page = getpagesize(); to find the pagesize at runtime.

[andrewbird]

And apparently that function is deprecated, should now use long sz = sysconf(_SC_PAGESIZE); instead.

[stsp]

mprot.c.txt
Please try this test.
It is again trying "unaligned" mprotect,
but with the same protection as the
previously aligned one.

[andrewbird]

ajb@raspberrypi:/local/src/dosemu2.git $ getconf PAGE_SIZE
16384
ajb@raspberrypi:/local/src/dosemu2.git $ gcc -o mprot2 mprot2.c 
ajb@raspberrypi:/local/src/dosemu2.git $ ./mprot2
mapped 0x7fff7318c000
mprotect 0x7fff73190000 worked
mprotect(): Invalid argument

[stsp]

Off-topic: https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=3ca459eaba1b
So we'll get the group perms for
tun device working as expected
very soon. But it would be good
to get the packaged CI fixed in a
mean time. :)

[andrewbird]

But it would be good
to get the packaged CI fixed in a
mean time. :)

I've forgotten, what needs doing there?

[stsp]

Allow "ERROR: Native DPMI not compiled in"
on Focal (https://github.com/dosemu2/dosemu2/actions/runs/12110023736/job/33759822452)

[andrewbird]

So is the native DPMI missing just likely to affect the CPU tests that are failing or will it affect general Dosemu use where no DPMI preference is configured?

[stsp]

It should affect only a few tests
as native DPMI is not a default.
Even on KVM-less CI I think it
should use JIT by default, not
native.