Hi, mjt > 2026年6月5日 04:52，Miao Wang <[email protected]> 写道： > > Hi, > >> 2026年6月5日 00:06，Miao Wang <[email protected]> 写道： >> >> Hi, >> >>> 2026年6月4日 23:44，Michael Tokarev <[email protected]> 写道： >>> >>> On 04.06.2026 11:56, Miao Wang wrote: >>>> Hi, all >>>> I can reproduce this test failure locally. I believe the reason is that the >>>> implementation of the pvscsi device in qemu is broken when the endianness >>>> of >>>> the guest and the host differs. >>>> It can be easily seen in the driver code of pvscsi that the command data is >>>> sent through a register in units of 32bits in the native endianness. In the >>>> qemu emulated device code, however, although qemu has done the endianness >>>> translation of the registers, the command data is received as a data stream >>>> and filled into a buffer. When a field of a command data struct is 64bits >>>> long, >>>> the upper 32 bits and the lower 32 bits will be flipped resulting the >>>> mis-interpretation of the command data. >>> >>> Oh wow. >>> >>> Can you post this to the qemu mailing list please? >>> That sounds like a big issue. >> >> Posted on their issue tracker. >> >> https://gitlab.com/qemu-project/qemu/-/work_items/3530 >> > > Can you try the attached patch for qemu please? I have > tested basic functions and found it basically works. >
I have posted my patch fixing pvscsi emulation on big-endian hosts to the qemu-devel mailing list [1]. However, I have not received any response till now. Cheers, Miao Wang [1]: https://lore.kernel.org/qemu-devel/[email protected]/
