From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:49414) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1X9Xms-0000DG-8j for qemu-devel@nongnu.org; Tue, 22 Jul 2014 06:59:20 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1X9Xmm-00079B-3T for qemu-devel@nongnu.org; Tue, 22 Jul 2014 06:59:14 -0400 Message-ID: <53CE43F9.6010403@suse.de> Date: Tue, 22 Jul 2014 12:59:05 +0200 From: Alexander Graf MIME-Version: 1.0 References: <1405935017-15390-1-git-send-email-hutao@cn.fujitsu.com> In-Reply-To: <1405935017-15390-1-git-send-email-hutao@cn.fujitsu.com> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH for 2.1] ppc: fix -mem-path failure List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Hu Tao Cc: "Shreyas B. Prabhu" , "" , qemu-devel@nongnu.org, Paolo Bonzini On 07/21/2014 11:30 AM, Hu Tao wrote: > commit e938ba0c tried to enable -mem-path for ppc but breaked some ppc > boards. > > The problems are: > > 1. it fails when allocating memory for rom, sram whose sizes are less > than huge page size: > > ./ppc-softmmu/qemu-system-ppc -m 512 -mem-path /hugepages/ \ > -kernel /home/hutao/Downloads/vmlinux-ppc -initrd \ > /home/hutao/Downloads/initrd-ppc.gz > qemu-system-ppc: /mnt/data/projects/qemu/exec.c:1184: qemu_ram_set_idstr: Assertion `new_block' failed. > > 2. if there is a numa node backed by memory backend object, qemu fails > with message: > > ./ppc-softmmu/qemu-system-ppc -m 512 \ > -object memory-backend-file,size=512M,mem-path=/hugepages,id=f0 \ > -numa node,nodeid=0,memdev=f0 \ > -kernel /home/hutao/Downloads/vmlinux-ppc \ > -initrd /home/hutao/Downloads/initrd-ppc.gz > qemu-system-ppc: memory backend f0 is used multiple times. Each -numa option must use a different memdev value. > > This patch does following: > > 1. replaces memory_region_allocate_system_memory() with > memory_region_init_ram() for rom, sram. Then only system memory > is backed by hugepages when specifying mem-path. > > 2. for memory banks, allocates all ram with > one memory_region_allocate_system_memory(), and use > memory_region_init_alias() to initialize memory banks. > > Tested machines: default(g3beige), mac99, taihu, bamboo, ref405ep. > > Signed-off-by: Hu Tao Thanks, this makes a lot of sense. I've applied this to ppc-next (for 2.1) and kicked off an autotest run. If that's successful, I'll send a pull request to Peter ASAP. Alex