On Fri, Oct 22, 2021 at 11:21:10AM +0200, Martin Kaiser wrote: > Thus wrote Michael Straube (straube.linux@xxxxxxxxx): > > > > > + !is_broadcast_ether_addr(GetAddr1Ptr(pframe))) > > > Hi Martin, > > > I'm not an expert regarding alignment. Is GetAddr1Ptr(pframe) always > > __aligned(2) as required by is_broadcast_ether_addr? > > Hi Michael, > > thanks for spotting this. To be honest, I didn't look at this in much > detail when I wrote the patch. > > I suppose the pframe comes from recvbuf2recvframe(). > precvframe = rtw_alloc_recvframe(pfree_recv_queue); > with > struct __queue *pfree_recv_queue = &precvpriv->free_recv_queue; > > This should be initialised in _rtw_init_recv_priv(). > rtw_init_queue(&precvpriv->free_recv_queue); > ... > precvpriv->precv_frame_buf = (u8 *)N_BYTE_ALIGMENT((size_t)(precvpriv->pallocated_frame_buf), RXFRAME_ALIGN_SZ); > precvframe = (struct recv_frame *)precvpriv->precv_frame_buf; > and the frames are added to the free_recv_queue. > RXFRAME_ALIGN_SZ is 1<<8. > > So pframe should be 256-byte aligned. > GetAddr1Ptr adds 4 to the start of pframe. > > I guess we're safe here. Greg already applied this patch. I'm not sure why you're looking at precvpriv->precv_frame_buf instead of "precv_frame->rx_data"? Am I missing something? This is actually a bit tricky for me to analyze because it gets set in two places: drivers/staging/r8188eu/core/rtw_recv.c | recvframe_pull | (struct recv_frame)->rx_data | 0-u64max drivers/staging/r8188eu/hal/usb_ops_linux.c | recvbuf2recvframe | (struct recv_frame)->rx_data | 0-u64max But the fact that we're using GetAddr1Ptr() on it at all suggests that it must be aligned and hopefully we've verified that the ->rx_data has enough data etc... ? This code is much trickier than I like it to be. Anyway, this patch doesn't introduce bugs that weren't present in the original. > > > > > @@ -841,7 +840,7 @@ void odm_RSSIMonitorCheck(struct odm_dm_struct *pDM_Odm) > > > > psta = pDM_Odm->pODM_StaInfo[i]; > > > > if (IS_STA_VALID(psta) && > > > > (psta->state & WIFI_ASOC_STATE) && > > > > - memcmp(psta->hwaddr, bcast_addr, ETH_ALEN) && > > > > + !is_broadcast_ether_addr(psta->hwaddr) && > > > Perhaps we should add __aligned(2) to the hwaddr variable in struct > > sta_info to be safe? > > > u8 hwaddr[ETH_ALEN] __aligned(2); > > Hmm, some of those arrays in other parts of the kernel have > __aligned(2), others don't... > > Can anyone else give some guidance? This array comes after a u32 and it's not packed so it's going to __aligned(4) already. regards, dan carpenter
