On Thu, Jul 25, 2013 at 6:11 AM, Hans Petter Selasky <h...@bitfrost.no> wrote: > On 07/24/13 22:09, Damjan Jovanovic wrote: >> >> Hi >> >> The ioctl: >> #define USB_FS_INIT _IOW ('U', 195, struct usb_fs_init) >> when used by a COMPAT_32BIT libusb on amd64, fails (causing >> libusb_open() to fail) due to: >> >> freebsd32_ioctl(0x6,0x800c55c3,0xffffc710,0x0,0x0,0x0) ERR#25 >> 'Inappropriate ioctl for device' >> but when hacked a bit: >> freebsd32_ioctl(0x6,0x801055c3,0xffffc710,0x0,0x0,0x0) = 0 (0x0) >> >> because sizeof(struct usb_fs_init) is 12 bytes on i386, and 16 bytes on >> amd64. >> >> But the failure of even libusb_open() - a fundamental libusb function >> - must mean that the COMPAT_32BIT libusb never worked. Can we please >> take this opportunity to kill it and do a real 32 bit compatibility >> layer in the kernel that will work from 32 bit chroots and statically >> linked 32 bit binaries? I'll help. > > > Hi, > > How did you compile it? Is the structure the same size, compiled with GCC > and LLVM?
GCC 4.2.1 "gcc file.c -o file -lusb" in a 32 bit chroot = 12 bytes "gcc file.c -o file -lusb" 64 bit = 16 bytes Clang 3.1 "clang file.c -o file -lusb" in a 32 bit chroot = 12 bytes "clang file.c -o file -lusb" 64 bit = 16 bytes The "uint8_t ep_index_max" field at the end of struct usb_fs_init must be getting padded to a 4 byte boundary on 32 bit, and 8 byte boundary on 64 bit. > The LIB32 for USB has been tested. I did a quick test and found on 9-stable: > > cc -m32 -I . -L /usr/lib32 -lusb usbconfig.c dump.c > env LD_PRELOAD=/usr/lib32/libusb.so ./a.out > ugen0.1: <UHCI root HUB Intel> at usbus0, cfg=0 md=HOST spd=FULL (12Mbps) > pwr=SAVE (0mA) > ugen1.1: <UHCI root HUB Intel> at usbus1, cfg=0 md=HOST spd=FULL (12Mbps) > pwr=SAVE (0mA) > > ... But usbconfig uses the libusb20 API and never seems to get to that broken ioctl. > Your approach requires much more code. It is not just about some structures, > but also about code accessing those structures. Feel free to submit a patch > however. I'll see what I can do. > I would rather fix this by adding the proper __aligned() to the structures > in question. Do that in the meanwhile? > --HPS > Damjan _______________________________________________ email@example.com mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-usb To unsubscribe, send any mail to "freebsd-usb-unsubscr...@freebsd.org"