From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([140.186.70.92]:45760) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1QHyUf-0007Q0-Ox for qemu-devel@nongnu.org; Thu, 05 May 2011 09:21:26 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1QHyUb-0004G9-U3 for qemu-devel@nongnu.org; Thu, 05 May 2011 09:21:25 -0400 Received: from mx1.redhat.com ([209.132.183.28]:38888) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1QHyUb-0004G2-Ly for qemu-devel@nongnu.org; Thu, 05 May 2011 09:21:21 -0400 Date: Thu, 5 May 2011 16:21:07 +0300 From: "Michael S. Tsirkin" Message-ID: <20110505132106.GG30119@redhat.com> References: <20110503182039.28430.26530.stgit@s20.home> <20110503183652.28430.83897.stgit@s20.home> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20110503183652.28430.83897.stgit@s20.home> Subject: Re: [Qemu-devel] [PATCH v2 3/3] CPUPhysMemoryClient: Batch contiguous addresses when playing catchup List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Alex Williamson Cc: jan.kiszka@siemens.com, qemu-devel@nongnu.org, armbru@redhat.com On Tue, May 03, 2011 at 12:36:58PM -0600, Alex Williamson wrote: > When a phys memory client registers and we play catchup by walking > the page tables, we can make a huge improvement in the number of > times the set_memory callback is called by batching contiguous > pages together. With a 4G guest, this reduces the number of callbacks > at registration from 1048866 to 296. > > Signed-off-by: Alex Williamson > --- > > exec.c | 38 ++++++++++++++++++++++++++++++++------ > 1 files changed, 32 insertions(+), 6 deletions(-) > > diff --git a/exec.c b/exec.c > index bbd5c86..a0678a4 100644 > --- a/exec.c > +++ b/exec.c > @@ -1741,14 +1741,21 @@ static int cpu_notify_migration_log(int enable) > return 0; > } > > +struct last_map { > + target_phys_addr_t start_addr; > + ram_addr_t size; A bit worried that ram_addr_t size might thinkably overflow (it's just a long, could be a 4G ram). Break it out when it fills up? > + ram_addr_t phys_offset; > +}; > + > /* The l1_phys_map provides the upper P_L1_BITs of the guest physical > * address. Each intermediate table provides the next L2_BITs of guest > * physical address space. The number of levels vary based on host and > * guest configuration, making it efficient to build the final guest > * physical address by seeding the L1 offset and shifting and adding in > * each L2 offset as we recurse through them. */ > -static void phys_page_for_each_1(CPUPhysMemoryClient *client, > - int level, void **lp, target_phys_addr_t addr) > +static void phys_page_for_each_1(CPUPhysMemoryClient *client, int level, > + void **lp, target_phys_addr_t addr, > + struct last_map *map) > { > int i; > > @@ -1760,15 +1767,29 @@ static void phys_page_for_each_1(CPUPhysMemoryClient *client, > addr <<= L2_BITS + TARGET_PAGE_BITS; > for (i = 0; i < L2_SIZE; ++i) { > if (pd[i].phys_offset != IO_MEM_UNASSIGNED) { > - client->set_memory(client, addr | i << TARGET_PAGE_BITS, > - TARGET_PAGE_SIZE, pd[i].phys_offset); > + target_phys_addr_t start_addr = addr | i << TARGET_PAGE_BITS; > + > + if (map->size && > + start_addr == map->start_addr + map->size && > + pd[i].phys_offset == map->phys_offset + map->size) { > + > + map->size += TARGET_PAGE_SIZE; > + continue; > + } else if (map->size) { > + client->set_memory(client, map->start_addr, > + map->size, map->phys_offset); > + } > + > + map->start_addr = start_addr; > + map->size = TARGET_PAGE_SIZE; > + map->phys_offset = pd[i].phys_offset; > } > } > } else { > void **pp = *lp; > for (i = 0; i < L2_SIZE; ++i) { > phys_page_for_each_1(client, level - 1, pp + i, > - (addr << L2_BITS) | i); > + (addr << L2_BITS) | i, map); > } > } > } > @@ -1776,9 +1797,14 @@ static void phys_page_for_each_1(CPUPhysMemoryClient *client, > static void phys_page_for_each(CPUPhysMemoryClient *client) > { > int i; > + struct last_map map = { 0 }; > + Nit: just {} is enough. > for (i = 0; i < P_L1_SIZE; ++i) { > phys_page_for_each_1(client, P_L1_SHIFT / L2_BITS - 1, > - l1_phys_map + i, i); > + l1_phys_map + i, i, &map); > + } > + if (map.size) { > + client->set_memory(client, map.start_addr, map.size, map.phys_offset); > } > } >