From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga03.intel.com (mga03.intel.com [134.134.136.65]) by dpdk.org (Postfix) with ESMTP id 3A9197F74 for ; Thu, 13 Nov 2014 13:48:46 +0100 (CET) Received: from orsmga001.jf.intel.com ([10.7.209.18]) by orsmga103.jf.intel.com with ESMTP; 13 Nov 2014 04:56:18 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.07,377,1413270000"; d="scan'208";a="607215463" Received: from bricha3-mobl3.ger.corp.intel.com ([10.237.220.93]) by orsmga001.jf.intel.com with SMTP; 13 Nov 2014 04:58:42 -0800 Received: by (sSMTP sendmail emulation); Thu, 13 Nov 2014 12:58:41 +0025 Date: Thu, 13 Nov 2014 12:58:41 +0000 From: Bruce Richardson To: "Burakov, Anatoly" Message-ID: <20141113125840.GA9932@bricha3-MOBL3> References: <1415619272-8281-1-git-send-email-anatoly.burakov@intel.com> <1415700565-19157-1-git-send-email-anatoly.burakov@intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Organization: Intel Shannon Ltd. User-Agent: Mutt/1.5.23 (2014-03-12) Cc: "dev@dpdk.org" Subject: Re: [dpdk-dev] [PATCH v8] eal: map PCI memory resources after hugepages X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 13 Nov 2014 12:48:47 -0000 On Thu, Nov 13, 2014 at 11:34:22AM +0000, Burakov, Anatoly wrote: > Hi Thomas and all > > Are there any objections to this patch? If there are no objections to it, could someone perhaps ack it? > No objections, just a quick request for clarification below. /Bruce > Thanks, > Anatoly > > -----Original Message----- > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Anatoly Burakov > Sent: Tuesday, November 11, 2014 10:09 AM > To: dev@dpdk.org > Subject: [dpdk-dev] [PATCH v8] eal: map PCI memory resources after hugepages > > Multi-process DPDK application must mmap hugepages and PCI resources into the same virtual address space. By default the virtual addresses are chosen by the primary process automatically when calling the mmap. > But sometimes the chosen virtual addresses aren't usable in secondary process - for example, secondary process is linked with more libraries than primary process, and the library occupies the same address space that the primary process has requested for PCI mappings. > > This patch makes EAL try and map PCI BARs right after the hugepages (instead of location chosen by mmap) in virtual memory, so that PCI BARs have less chance of ending up in random places in virtual memory. > > Signed-off-by: Liang Xu > Signed-off-by: Anatoly Burakov > --- > lib/librte_eal/linuxapp/eal/eal_pci.c | 30 ++++++++++++++++------ > lib/librte_eal/linuxapp/eal/eal_pci_uio.c | 13 ++++++++-- > lib/librte_eal/linuxapp/eal/eal_pci_vfio.c | 19 +++++++++++--- > lib/librte_eal/linuxapp/eal/include/eal_pci_init.h | 6 +++++ > 4 files changed, 55 insertions(+), 13 deletions(-) > > diff --git a/lib/librte_eal/linuxapp/eal/eal_pci.c b/lib/librte_eal/linuxapp/eal/eal_pci.c > index 5fe3961..79fbbb8 100644 > --- a/lib/librte_eal/linuxapp/eal/eal_pci.c > +++ b/lib/librte_eal/linuxapp/eal/eal_pci.c > @@ -97,6 +97,25 @@ error: > return -1; > } > > +void * > +pci_find_max_end_va(void) > +{ > + const struct rte_memseg *seg = rte_eal_get_physmem_layout(); > + const struct rte_memseg *last = seg; > + unsigned i = 0; > + > + for (i = 0; i < RTE_MAX_MEMSEG; i++, seg++) { > + if (seg->addr == NULL) > + break; > + > + if (seg->addr > last->addr) > + last = seg; > + > + } > + return RTE_PTR_ADD(last->addr, last->len); } > + > + > /* map a particular resource from a file */ void * pci_map_resource(void *requested_addr, int fd, off_t offset, size_t size) @@ -106,21 +125,16 @@ pci_map_resource(void *requested_addr, int fd, off_t offset, size_t size) > /* Map the PCI memory resource of device */ > mapaddr = mmap(requested_addr, size, PROT_READ | PROT_WRITE, > MAP_SHARED, fd, offset); > - if (mapaddr == MAP_FAILED || > - (requested_addr != NULL && mapaddr != requested_addr)) { Why has this check been removed from here. I assume it is replaced by a new check in secondary processes that I see added below, but perhaps you could explain the reason for the change? > + if (mapaddr == MAP_FAILED) { > RTE_LOG(ERR, EAL, "%s(): cannot mmap(%d, %p, 0x%lx, 0x%lx): %s (%p)\n", > __func__, fd, requested_addr, > (unsigned long)size, (unsigned long)offset, > strerror(errno), mapaddr); > - goto fail; > + } else { > + RTE_LOG(DEBUG, EAL, " PCI memory mapped at %p\n", mapaddr); > } > > - RTE_LOG(DEBUG, EAL, " PCI memory mapped at %p\n", mapaddr); > - > return mapaddr; > - > -fail: > - return NULL; > } > > /* parse the "resource" sysfs file */ > diff --git a/lib/librte_eal/linuxapp/eal/eal_pci_uio.c b/lib/librte_eal/linuxapp/eal/eal_pci_uio.c > index 7e62266..e53f06b 100644 > --- a/lib/librte_eal/linuxapp/eal/eal_pci_uio.c > +++ b/lib/librte_eal/linuxapp/eal/eal_pci_uio.c > @@ -35,6 +35,7 @@ > #include > #include > #include > +#include > > #include > #include > @@ -48,6 +49,8 @@ > > static int pci_parse_sysfs_value(const char *filename, uint64_t *val); > > +void *pci_map_addr = NULL; > + > > #define OFF_MAX ((uint64_t)(off_t)-1) > static int > @@ -371,10 +374,16 @@ pci_uio_map_resource(struct rte_pci_device *dev) > if (maps[j].addr != NULL) > fail = 1; > else { > - mapaddr = pci_map_resource(NULL, fd, (off_t)offset, > + /* try mapping somewhere close to the end of hugepages */ > + if (pci_map_addr == NULL) > + pci_map_addr = pci_find_max_end_va(); > + > + mapaddr = pci_map_resource(pci_map_addr, fd, (off_t)offset, > (size_t)maps[j].size); > - if (mapaddr == NULL) > + if (mapaddr == MAP_FAILED) > fail = 1; > + > + pci_map_addr = RTE_PTR_ADD(mapaddr, (size_t) maps[j].size); > } > > if (fail) { > diff --git a/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c b/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c > index c776ddc..c1246e8 100644 > --- a/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c > +++ b/lib/librte_eal/linuxapp/eal/eal_pci_vfio.c > @@ -37,6 +37,7 @@ > #include > #include > #include > +#include > > #include > #include > @@ -720,10 +721,22 @@ pci_vfio_map_resource(struct rte_pci_device *dev) > if (i == msix_bar) > continue; > > - bar_addr = pci_map_resource(maps[i].addr, vfio_dev_fd, reg.offset, > - reg.size); > + if (internal_config.process_type == RTE_PROC_PRIMARY) { > + /* try mapping somewhere close to the end of hugepages */ > + if (pci_map_addr == NULL) > + pci_map_addr = pci_find_max_end_va(); > + > + bar_addr = pci_map_resource(pci_map_addr, vfio_dev_fd, reg.offset, > + reg.size); > + pci_map_addr = RTE_PTR_ADD(bar_addr, (size_t) reg.size); > + } else { > + bar_addr = pci_map_resource(maps[i].addr, vfio_dev_fd, reg.offset, > + reg.size); > + } > > - if (bar_addr == NULL) { > + if (bar_addr == MAP_FAILED || > + (internal_config.process_type == RTE_PROC_SECONDARY && > + bar_addr != maps[i].addr)) { > RTE_LOG(ERR, EAL, " %s mapping BAR%i failed: %s\n", pci_addr, i, > strerror(errno)); > close(vfio_dev_fd); > diff --git a/lib/librte_eal/linuxapp/eal/include/eal_pci_init.h b/lib/librte_eal/linuxapp/eal/include/eal_pci_init.h > index d758bee..1070eb8 100644 > --- a/lib/librte_eal/linuxapp/eal/include/eal_pci_init.h > +++ b/lib/librte_eal/linuxapp/eal/include/eal_pci_init.h > @@ -59,6 +59,12 @@ struct mapped_pci_resource { TAILQ_HEAD(mapped_pci_res_list, mapped_pci_resource); extern struct mapped_pci_res_list *pci_res_list; > > +/* > + * Helper function to map PCI resources right after hugepages in > +virtual memory */ extern void *pci_map_addr; void > +*pci_find_max_end_va(void); > + > void *pci_map_resource(void *requested_addr, int fd, off_t offset, > size_t size); > > -- > 1.8.1.4 >