From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: by sourceware.org (Postfix, from userid 48) id A9F713857416; Thu, 14 Oct 2021 04:15:57 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org A9F713857416 From: "jon@solid-run.com" To: glibc-bugs@sourceware.org Subject: [Bug libc/28432] Aarch64 memcpy used on device-memory Date: Thu, 14 Oct 2021 04:15:57 +0000 X-Bugzilla-Reason: CC X-Bugzilla-Type: changed X-Bugzilla-Watch-Reason: None X-Bugzilla-Product: glibc X-Bugzilla-Component: libc X-Bugzilla-Version: 2.32 X-Bugzilla-Keywords: X-Bugzilla-Severity: normal X-Bugzilla-Who: jon@solid-run.com X-Bugzilla-Status: UNCONFIRMED X-Bugzilla-Resolution: X-Bugzilla-Priority: P2 X-Bugzilla-Assigned-To: unassigned at sourceware dot org X-Bugzilla-Target-Milestone: --- X-Bugzilla-Flags: X-Bugzilla-Changed-Fields: Message-ID: In-Reply-To: References: Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Bugzilla-URL: http://sourceware.org/bugzilla/ Auto-Submitted: auto-generated MIME-Version: 1.0 X-BeenThere: glibc-bugs@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Glibc-bugs mailing list List-Unsubscribe: , List-Archive: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 14 Oct 2021 04:15:57 -0000 https://sourceware.org/bugzilla/show_bug.cgi?id=3D28432 --- Comment #5 from Jon Nettleton --- (In reply to Szabolcs Nagy from comment #4) > the issue may be how the memory was mapped: it is invalid > to map with normal memory attributes if overlapping writes > are not supported the way hw expects normal memory to work. > (i'm unfortunately not an expert on this and i don't know > what's going on in the drivers, but any further information > would be helpful so others can understand the issue better) This is not a new issue. It was first discussed in 2018 on this very long kernel mailing list thread.=20 https://lore.kernel.org/lkml/87h8k7h8q9.fsf@linux.ibm.com/T/ It was originally written off as a PCIe implementation problem. Well I have now seen the issue personally on three different combinations of Cortex-a72, CCN's, and PCIe IP, from multiple manufacturer's. The only common part of a= ll the implementations is the Arm implementation of the Cortex-A72. I will no= te, but can not personally verify that most likely the Kunpeng SOC also exhibits this behavior because they also have similar patches to the ones I have written. Someone could spend another 4 years trying to find the exact cause of this behavior, since it has just sat around for the last 4 years and ignored my guess is that this is not going to happen. Half of the engineers claim that mapping device memory as uncached on Arm is the same as write-combine and another half that claim that device memory can never be treated as normal memory. Truth be told unless we are going to rework the entire Linux graph= ics stack we need some sort of compatibility with how the software currently wo= rks. If we don't want to patch the main aarch64 memcpy then I will be happy to submit a ifunc implementation. --=20 You are receiving this mail because: You are on the CC list for the bug.=