From mboxrd@z Thu Jan 1 00:00:00 1970 From: Troy Kisky Date: Tue, 14 Aug 2012 11:47:56 -0700 Subject: [U-Boot] [PATCH 1/1] USB: Fix strict aliasing in ohci-hcd In-Reply-To: <201208141933.29344.marex@denx.de> References: <1344721749-25644-1-git-send-email-troy.kisky@boundarydevices.com> <201208141933.29344.marex@denx.de> Message-ID: <502A9D5C.9060804@boundarydevices.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de On 8/14/2012 10:33 AM, Marek Vasut wrote: > Dear Troy Kisky, > >> commit 5f6aa03fda2a0a79940765865c1e4266be8a75f8 >> USB: Fix complaints about strict aliasing in OHCI-HCD >> >> tried to fix this, but gcc4.4 still complains. So, this >> patch basically reverts the above and does a simpler fix. >> >> also, the above commit incorrectly changed >> /* corresponds to data_buf[4-7] */ >> datab [1] = 0; >> to >> >> /* corresponds to databuf.u8[4-7] */ >> databuf.u8[1] = 0; >> >> This patch also fixes that. >> >> Signed-off-by: Troy Kisky >> --- >> drivers/usb/host/ohci-hcd.c | 70 >> +++++++++++++++++++------------------------ 1 file changed, 31 >> insertions(+), 39 deletions(-) >> >> diff --git a/drivers/usb/host/ohci-hcd.c b/drivers/usb/host/ohci-hcd.c >> index d24f2f1..9f47351 100644 >> --- a/drivers/usb/host/ohci-hcd.c >> +++ b/drivers/usb/host/ohci-hcd.c >> @@ -1261,19 +1261,11 @@ static int ohci_submit_rh_msg(struct usb_device >> *dev, unsigned long pipe, int leni = transfer_len; >> int len = 0; >> int stat = 0; >> - __u32 datab[4]; >> - union { >> - void *ptr; >> - __u8 *u8; >> - __u16 *u16; >> - __u32 *u32; >> - } databuf; >> __u16 bmRType_bReq; >> __u16 wValue; >> __u16 wIndex; >> __u16 wLength; >> - >> - databuf.u32 = (__u32 *)datab; >> + ALLOC_ALIGN_BUFFER(__u8, databuf, 16, sizeof(u32)); > Ok, pardon my sloppiness in reviews ... but why not alloc this cache aligned? > That way we'd also circumvent the issues with caches we'd eventually hit anyway. > > [...] > > The rest is OK. > Line 1464 has if (data != databuf) memcpy(data, databuf, len); So as far as I can tell, databuf is always copied and there are no cache issues to circumvent. But I have no complaint if you desire to cacheline aligned anyway... Troy