From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:34696) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ad2lr-0002uB-4Q for qemu-devel@nongnu.org; Mon, 07 Mar 2016 16:32:56 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ad2ln-0007Ky-PV for qemu-devel@nongnu.org; Mon, 07 Mar 2016 16:32:54 -0500 Received: from relay1.mentorg.com ([192.94.38.131]:44484) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ad2ln-0007KK-JG for qemu-devel@nongnu.org; Mon, 07 Mar 2016 16:32:51 -0500 Message-ID: <1457386368.3827.5.camel@mentor.com> From: Hollis Blanchard Date: Mon, 7 Mar 2016 13:32:48 -0800 In-Reply-To: <1456949575-1633-2-git-send-email-hollis_blanchard@mentor.com> References: <1456949575-1633-1-git-send-email-hollis_blanchard@mentor.com> <1456949575-1633-2-git-send-email-hollis_blanchard@mentor.com> Content-Type: multipart/alternative; boundary="=-8AXl2Paaym7rhBQVrx0P" MIME-Version: 1.0 Subject: Re: [Qemu-devel] [PATCH v2 2/2] trace: separate MMIO tracepoints from TB-access tracepoints List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: stefanha@redhat.com Cc: qemu-devel@nongnu.org --=-8AXl2Paaym7rhBQVrx0P Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8bit On Wed, 2016-03-02 at 12:12 -0800, Hollis Blanchard wrote: > Memory accesses to code which has previously been translated into a > TB show up > in the MMIO path, so that they may invalidate the TB. It's extremely > confusing > to mix those in with device MMIOs, so split them into their own > tracepoint. > > Signed-off-by: Hollis Blanchard > Reviewed-by: Stefan Hajnoczi > --- > It took many hours to figure out why some RAM accesses were coming > through the > MMIO path instead of being handled inline in the TBs. > > On IRC, Paolo expressed some concern about performance, but > ultimately agreed > that adding one conditional to an already heavy codepath wouldn't > have much > impact. > > v2: rename trace_memory_region_ops_tb_read/write to > trace_memory_region_tb_read/write > Stefan, any further comment here? You sent a Reviewed-by on the original, but unlike the "include CPU index" patch, you didn't mention adding this one to https://github.com/stefanha/qemu/commits/tracing. -- Hollis Blanchard Mentor Graphics Emulation Division --=-8AXl2Paaym7rhBQVrx0P Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: 7bit

On Wed, 2016-03-02 at 12:12 -0800, Hollis Blanchard wrote:

Memory accesses to code which has previously been translated into a TB show up
in the MMIO path, so that they may invalidate the TB. It's extremely confusing
to mix those in with device MMIOs, so split them into their own tracepoint.

Signed-off-by: Hollis Blanchard <hollis_blanchard@mentor.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
---
It took many hours to figure out why some RAM accesses were coming through the
MMIO path instead of being handled inline in the TBs.

On IRC, Paolo expressed some concern about performance, but ultimately agreed
that adding one conditional to an already heavy codepath wouldn't have much
impact.

v2: rename trace_memory_region_ops_tb_read/write to
trace_memory_region_tb_read/write

Stefan, any further comment here? You sent a Reviewed-by on the original, but unlike the "include CPU index" patch, you didn't mention adding this one to https://github.com/stefanha/qemu/commits/tracing.

--
Hollis Blanchard <hollis_blanchard@mentor.com>
Mentor Graphics Emulation Division

--=-8AXl2Paaym7rhBQVrx0P--