On 10/10/20 11:46 pm, Jakub Kicinski wrote: > On Sat, 10 Oct 2020 23:34:51 +0530 Anant Thazhemadam wrote: >> On 10/10/20 10:29 pm, Jakub Kicinski wrote: >>> On Sat, 10 Oct 2020 12:14:59 +0530 Anant Thazhemadam wrote: >>>> get_registers() directly returns the return value of >>>> usb_control_msg_recv() - 0 if successful, and negative error number >>>> otherwise. >>> Are you expecting Greg to take this as a part of some USB subsystem >>> changes? I don't see usb_control_msg_recv() in my tree, and the >>> semantics of usb_control_msg() are not what you described. >> No, I'm not. usb_control_msg_recv() is an API that was recently >> introduced, and get_registers() in rtl8150.c was also modified to >> use it in order to prevent partial reads. >> >> By your tree, I assume you mean >> https://git.kernel.org/pub/scm/linux/kernel/git/kuba/linux.git/ >> (it was the only one I could find). >> >> I don't see the commit that this patch is supposed to fix in your >> tree either... :/ >> >> Nonetheless, this commit fixes an issue that was applied to the >> networking tree, and has made its way into linux-next as well, if >> I'm not mistaken. > I mean the networking tree, what's the commit ID in linux-next? > > Your fixes tag points to f45a4248ea4c, but looks like the code was > quite correct at that point. Ah, my apologies. You're right. It doesn't look like those helpers have made their way into the networking tree yet. (This gets mentioned here as well, https://www.mail-archive.com/netdev@xxxxxxxxxxxxxxx/msg357843.html) The commit ID pointed to by the fixes tag is correct. The change introduced by said commit looks right, but is logically incorrect. get_registers() directly returns the return value of usb_control_msg_recv(), and usb_control_msg_recv() returns 0 on success and negative error number otherwise. (You can find more about the new helpers here https://lore.kernel.org/alsa-devel/20200914153756.3412156-1-gregkh@xxxxxxxxxxxxxxxxxxx/ ) The commit ID mentioned introduces a change that is supposed to copy over the ethernet only when get_registers() succeeds, i.e., a complete read occurs, and generate and set a random ethernet address otherwise (reading the commit message should give some more insight). The condition that checks if get_registers() succeeds (as specified in f45a4248ea4c) was, ret == sizeof(node_id) where ret is the return value of get_registers(). However, ret will never equal sizeof(node_id), since ret can only be equal to 0 or a negative number. Thus, even in case where get_registers() succeeds, a randomly generated MAC address would get copied over, instead of copying the appropriate ethernet address, which is logically incorrect and not optimal. Hence, we need to modify this to check if (ret == 0), and copy over the correct ethernet address in that case, instead of randomly generating one and assigning that. Hope this helps. Thanks, Anant
