> Is running on real hardware really more stable than in qemu? What kind > of issue do you encounter? > We should be able to make things just work on qemu, that'll be much > simpler for people to try out than trying on $random hardware
Really, both are necessary. Qemu is a "ideal" environment for test. But, in real hardware, the standards are not always followed, components have irregular latencies, strange combinations (IDE CDROM drive + SATA HDD, by example)... So we can start the test on Qemu but, once we pass the test in it, we can repeat the same in different hardware configurations, to check how it works in "irregular" situations. By example, some years ago, I found a bug in video interface enumeration (if the machine had many video outputs, each video output was listed as a videocard, and Xorg failed), and I found it testing the Hurd in a real machine. In SMP, Damien discovered that AMD machine reserve for other tasks the memory address which the standard indicated for the startup routine. Again, testing the SMP's gnumach in a real machine with AMD cpu. So both are necessary. We cannot develop an OS without never test it in real hardware El lun, 22 jun 2026 a las 23:14, Samuel Thibault (<[email protected]>) escribió: > Hello, > > [email protected], le jeu. 18 juin 2026 21:17:39 +0000, a ecrit: > > I consistantly have difficulty running the Hurd on qemu, so I run it > > on real hardware: https://hurd.ion.nu/faq/drivers.html > > Is running on real hardware really more stable than in qemu? What kind > of issue do you encounter? > > We should be able to make things just work on qemu, that'll be much > simpler for people to try out than trying on $random hardware. > > Samuel > >
