On 10/20/25 11:03, Alex Bennée wrote: > Sean Anderson <[email protected]> writes: > >> While discussing [1], it came to my attention that QEMU does not >> properly truncate/error SYS_FLEN on 32-bit systems. > > TIL that semihostingfs was a thing! > >> Fix this, and some >> other bugs with GDB File I/O that I found while working on this series. >> That said, GDB File I/O has been substantially broken for two years now, >> so it makes me wonder if anyone actually uses it! > > I suspect this is at the upper end of things to use semihosting for as > its real purpose is to help bootstrap things on the barest of metal > until you have enough bits going to selfhost. In QEMU land it is a > convenient way to do host calls for test cases. > > We don't have much actual testing of semihosting in the tree although I > do run Peter's semihosting tests from time to time: > > https://git.linaro.org/people/peter.maydell/semihosting-tests.git/ > > the tests do include flen() but obviously don't cover the extreme > filesize cases or overflow. > >> It would certainly >> simplify the implementation if we didn't have to support it. > > While semihosting does have the concept of optional extensions SYS_FLEN > is not one of them.
The above comments refer to GDB File I/O, which currently cannot open files due to the problem fixed in patch 1/3. FLEN is not really broken for most use-cases (as long as the user doesn't access large files). --Sean
