NUMA Hotplug emulator
in [Kernel]
|
From: Haicheng Li on 13 May 2010 22:20 Dave Hansen wrote: > On Thu, 2010-05-13 at 09:55 -0700, Greg KH wrote: >>> Add a sysfs entry "probe" under /sys/devices/system/node/: >>> >>> - to show all fake offlined nodes: >>> $ cat /sys/devices/system/node/probe >>> >>> - to hotadd a fake offlined node, e.g. nodeid is N: >>> $ echo N > /sys/devices/system/node/probe >> As you are trying to add a new sysfs file, please create the matching >> Documentation/ABI/ file as well. >> >> Also note that sysfs files are "one value per file", which I don't think >> this file follows, right? > > I think in this case, it was meant to be a list of acceptable parameters > rather than a set of values, kinda like /sys/power/state. Right. > Instead, I guess we could have: > > /sys/devices/system/node/probeable/3 > /sys/devices/system/node/probeable/43 > /sys/devices/system/node/probeable/65 > /sys/devices/system/node/probeable/5145 > > and the knowledge that you need to pick one of those to echo > into /sys/devices/system/node/probe. I think this way would make things complex if we just want to show user which node could be hotadded. > But, it's a lot more self > explanatory if you 'cat /sys/devices/system/node/probe', and then pick > one of those to echo back into the file. agreed. > Seems like a decent place to violate the "rule". :) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KAMEZAWA Hiroyuki on 13 May 2010 22:30 On Thu, 13 May 2010 19:45:44 +0800 Shaohui Zheng <shaohui.zheng(a)intel.com> wrote: > x86: infrastructure of NUMA hotplug emulation > Hmm. do we have to create this for x86 only ? Can't we live with lmb ? as lmb_hide_node() or some. IIUC, x86-version lmb is now under development. THanks, -Kame > NUMA hotplug emulator introduces a new node state N_HIDDEN to > identify the fake offlined node. It firstly hides RAM via E820 > table and then emulates fake offlined nodes with the hidden RAM. > > After system bootup, user is able to hotplug-add these offlined > nodes, which is just similar to a real hardware hotplug behavior. > > Using boot option "numa=hide=N*size" to fake offlined nodes: > - N is the number of hidden nodes > - size is the memory size (in MB) per hidden node. > > OPEN: Kernel might use part of hidden memory region as RAM buffer, > now emulator directly hide 128M extra space to workaround > this issue. Any better way to avoid this conflict? > > Signed-off-by: Haicheng Li <haicheng.li(a)linux.intel.com> > Signed-off-by: Shaohui Zheng <shaohui.zheng(a)intel.com> > --- > diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c > index 8948f47..3e0d94d 100644 > --- a/arch/x86/mm/numa_64.c > +++ b/arch/x86/mm/numa_64.c > @@ -307,6 +307,87 @@ void __init numa_init_array(void) > } > } > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +static char *hp_cmdline __initdata; > +static struct bootnode *hidden_nodes; > +static u64 hp_start, hp_end; > +static long hidden_num, hp_size; > + > +int hotadd_hidden_nodes(int nid) > +{ > + int ret; > + > + if (!node_hidden(nid)) > + return -EINVAL; > + > + ret = add_memory(nid, hidden_nodes[nid].start, > + hidden_nodes[nid].end - hidden_nodes[nid].start); > + if (!ret) { > + node_clear_hidden(nid); > + return 0; > + } else { > + return -EEXIST; > + } > +} > + > +static void __init numa_hide_nodes(void) > +{ > + char *c; > + int ret; > + > + c = strchr(hp_cmdline, '*'); > + if (!c) > + return; > + else > + *c = '\0'; > + ret = strict_strtol(hp_cmdline, 0, &hidden_num); > + if (ret == -EINVAL) > + return; > + ret = strict_strtol(c + 1, 0, &hp_size); > + if (ret == -EINVAL) > + return; > + hp_size <<= 20; > + > + hp_start = e820_hide_mem(hidden_num * hp_size); > + if (hp_start <= 0) { > + printk(KERN_ERR "Hide too much memory, disable node hotplug emualtion."); > + hidden_num = 0; > + return; > + } > + > + hp_end = hp_start + hidden_num * hp_size; > + > + /* leave 128M space for possible RAM buffer usage later > + * any other better way to avoid this conflict? > + */ > + e820_hide_mem(128*1024*1024); > +} > + > +static void __init numa_hotplug_emulation(void) > +{ > + int i, num_nodes = 0; > + > + for_each_online_node(i) > + if (i > num_nodes) > + num_nodes = i; > + > + i = num_nodes + hidden_num; > + if (!hidden_nodes) { > + hidden_nodes = alloc_bootmem(sizeof(struct bootnode) * i); > + memset(hidden_nodes, 0, sizeof(struct bootnode) * i); > + } > + > + if (hidden_num) > + for (i = 0; i < hidden_num; i++) { > + int nid = num_nodes + i + 1; > + node_set(nid, node_possible_map); > + hidden_nodes[nid].start = hp_start + hp_size * i; > + hidden_nodes[nid].end = hp_start + hp_size * (i+1); > + node_set_hidden(nid); > + } > +} > +#endif /* CONFIG_NODE_HOTPLUG_EMU */ > + > #ifdef CONFIG_NUMA_EMU > /* Numa emulation */ > static struct bootnode nodes[MAX_NUMNODES] __initdata; > @@ -661,7 +742,7 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > > #ifdef CONFIG_NUMA_EMU > if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8)) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > @@ -669,14 +750,14 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > #ifdef CONFIG_ACPI_NUMA > if (!numa_off && acpi && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, > last_pfn << PAGE_SHIFT)) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > > #ifdef CONFIG_K8_NUMA > if (!numa_off && k8 && !k8_scan_nodes()) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > @@ -696,6 +777,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > numa_set_node(i, 0); > e820_register_active_regions(0, start_pfn, last_pfn); > setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT); > +done: > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + if (hidden_num) > + numa_hotplug_emulation(); > +#endif > + return; > } > > unsigned long __init numa_free_all_bootmem(void) > @@ -723,6 +810,12 @@ static __init int numa_setup(char *opt) > if (!strncmp(opt, "fake=", 5)) > cmdline = opt + 5; > #endif > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + if (!strncmp(opt, "hide=", 5)) { > + hp_cmdline = opt + 5; > + numa_hide_nodes(); > + } > +#endif > #ifdef CONFIG_ACPI_NUMA > if (!strncmp(opt, "noacpi", 6)) > acpi_numa = -1; > diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h > index dba35e4..ba0f82d 100644 > --- a/include/linux/nodemask.h > +++ b/include/linux/nodemask.h > @@ -371,6 +371,10 @@ static inline void __nodes_fold(nodemask_t *dstp, const nodemask_t *origp, > */ > enum node_states { > N_POSSIBLE, /* The node could become online at some point */ > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + N_HIDDEN, /* The node is hidden at booting time, could be > + * onlined in run time */ > +#endif > N_ONLINE, /* The node is online */ > N_NORMAL_MEMORY, /* The node has regular memory */ > #ifdef CONFIG_HIGHMEM > @@ -470,6 +474,13 @@ static inline int num_node_state(enum node_states state) > #define node_online(node) node_state((node), N_ONLINE) > #define node_possible(node) node_state((node), N_POSSIBLE) > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +#define node_set_hidden(node) node_set_state((node), N_HIDDEN) > +#define node_clear_hidden(node) node_clear_state((node), N_HIDDEN) > +#define node_hidden(node) node_state((node), N_HIDDEN) > +extern int hotadd_hidden_nodes(int nid); > +#endif > + > #define for_each_node(node) for_each_node_state(node, N_POSSIBLE) > #define for_each_online_node(node) for_each_node_state(node, N_ONLINE) > > -- > Thanks & Regards, > Shaohui > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Haicheng Li on 13 May 2010 23:40 Shaohui Zheng wrote: > Userland interface to hotplug-add fake offlined nodes. > > Add a sysfs entry "probe" under /sys/devices/system/node/: > > - to show all fake offlined nodes: > $ cat /sys/devices/system/node/probe > > - to hotadd a fake offlined node, e.g. nodeid is N: > $ echo N > /sys/devices/system/node/probe > > Signed-off-by: Haicheng Li <haicheng.li(a)linux.intel.com> > Signed-off-by: Shaohui Zheng <shaohui.zheng(a)intel.com> > --- > diff --git a/drivers/base/node.c b/drivers/base/node.c > index 057979a..a0be257 100644 > --- a/drivers/base/node.c > +++ b/drivers/base/node.c > @@ -535,6 +535,26 @@ void unregister_one_node(int nid) > unregister_node(&node_devices[nid]); > } > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +static ssize_t store_nodes_probe(struct sysdev_class *class, > + struct sysdev_class_attribute *attr, > + const char *buf, size_t count) > +{ > + long nid; > + int ret; > + > + strict_strtol(buf, 0, &nid); > + if (nid < 0 || nid > nr_node_ids - 1) { Shaohui, In fact, no need to do such check here, hotadd_hidden_nodes() can handle such invalid parameter by itself. > + printk(KERN_ERR "Invalid NUMA node id: %d (0 <= nid < %d).\n", > + nid, nr_node_ids); > + return -EPERM; Per Andi's earlier review comments, -EPERM is odd, we'd fix it. > + } > + hotadd_hidden_nodes(nid); > + > + return count; > +} > +#endif Pls. replace it with following code: +#ifdef CONFIG_NODE_HOTPLUG_EMU +static ssize_t store_nodes_probe(struct sysdev_class *class, + struct sysdev_class_attribute *attr, + const char *buf, size_t count) +{ + long nid; + int ret; + + ret = strict_strtol(buf, 0, &nid); + if (ret == -EINVAL) + return ret; + + ret = hotadd_hidden_nodes(nid); + if (!ret) + return count; + else + return -EIO; +} +#endif -haicheng -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Wu Fengguang on 14 May 2010 00:20 > Pls. replace it with following code: > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +static ssize_t store_nodes_probe(struct sysdev_class *class, > + struct sysdev_class_attribute *attr, > + const char *buf, size_t count) > +{ > + long nid; > + int ret; > + > + ret = strict_strtol(buf, 0, &nid); > + if (ret == -EINVAL) > + return ret; > + > + ret = hotadd_hidden_nodes(nid); > + if (!ret) > + return count; > + else > + return -EIO; > +} > +#endif How about this? err = strict_strtol(buf, 0, &nid); if (err < 0) return err; err = hotadd_hidden_nodes(nid); if (err < 0) return err; return count; Thanks, Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Haicheng Li on 14 May 2010 00:50
Shaohui Zheng wrote: > x86: infrastructure of NUMA hotplug emulation > > NUMA hotplug emulator introduces a new node state N_HIDDEN to > identify the fake offlined node. It firstly hides RAM via E820 > table and then emulates fake offlined nodes with the hidden RAM. > > After system bootup, user is able to hotplug-add these offlined > nodes, which is just similar to a real hardware hotplug behavior. > > Using boot option "numa=hide=N*size" to fake offlined nodes: > - N is the number of hidden nodes > - size is the memory size (in MB) per hidden node. > > OPEN: Kernel might use part of hidden memory region as RAM buffer, > now emulator directly hide 128M extra space to workaround > this issue. Any better way to avoid this conflict? I'd like to hear some advices on this OPEN. I'm not sure if it's an undebugged bug, but it might be. Now, removing extra 128M is just a workaround. I used to find kernel might map the hidden address space to RAM buffer. For example: 1) I have removed the mem range "0x2c0000000 ~ 0x33fffffff" from e820 table. 2) however, after kernel bootup, I might find: # cat /proc/iomem 2c0000000-2c5ffffffff : RAM buffer This is just an example, the space range (2c5ffffffff) is not the exact number which I saw in my testing. This issue might come from commit 45fbe3ee01b8e463b28c2751b5dcc0cbdc142d90 Author: Linus Torvalds <torvalds(a)linux-foundation.org> Date: Wed May 6 08:06:44 2009 -0700 x86, e820, pci: reserve extra free space near end of RAM The point is to take all RAM resources we have, and _after_ we've added all the resources we've seen in the E820 tree, we then _also_ try to add fake reserved entries for any "round up to X" at the end of the RAM resources. [ Impact: improve PCI mem-resource allocation robustness, protect "stolen RA M" ] How could I make a better fix for this OPEN? -haicheng > Signed-off-by: Haicheng Li <haicheng.li(a)linux.intel.com> > Signed-off-by: Shaohui Zheng <shaohui.zheng(a)intel.com> > --- > diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c > index 8948f47..3e0d94d 100644 > --- a/arch/x86/mm/numa_64.c > +++ b/arch/x86/mm/numa_64.c > @@ -307,6 +307,87 @@ void __init numa_init_array(void) > } > } > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +static char *hp_cmdline __initdata; > +static struct bootnode *hidden_nodes; > +static u64 hp_start, hp_end; > +static long hidden_num, hp_size; > + > +int hotadd_hidden_nodes(int nid) > +{ > + int ret; > + > + if (!node_hidden(nid)) > + return -EINVAL; > + > + ret = add_memory(nid, hidden_nodes[nid].start, > + hidden_nodes[nid].end - hidden_nodes[nid].start); > + if (!ret) { > + node_clear_hidden(nid); > + return 0; > + } else { > + return -EEXIST; > + } > +} > + > +static void __init numa_hide_nodes(void) > +{ > + char *c; > + int ret; > + > + c = strchr(hp_cmdline, '*'); > + if (!c) > + return; > + else > + *c = '\0'; > + ret = strict_strtol(hp_cmdline, 0, &hidden_num); > + if (ret == -EINVAL) > + return; > + ret = strict_strtol(c + 1, 0, &hp_size); > + if (ret == -EINVAL) > + return; > + hp_size <<= 20; > + > + hp_start = e820_hide_mem(hidden_num * hp_size); > + if (hp_start <= 0) { > + printk(KERN_ERR "Hide too much memory, disable node hotplug emualtion."); > + hidden_num = 0; > + return; > + } > + > + hp_end = hp_start + hidden_num * hp_size; > + > + /* leave 128M space for possible RAM buffer usage later > + * any other better way to avoid this conflict? > + */ > + e820_hide_mem(128*1024*1024); > +} > + > +static void __init numa_hotplug_emulation(void) > +{ > + int i, num_nodes = 0; > + > + for_each_online_node(i) > + if (i > num_nodes) > + num_nodes = i; > + > + i = num_nodes + hidden_num; > + if (!hidden_nodes) { > + hidden_nodes = alloc_bootmem(sizeof(struct bootnode) * i); > + memset(hidden_nodes, 0, sizeof(struct bootnode) * i); > + } > + > + if (hidden_num) > + for (i = 0; i < hidden_num; i++) { > + int nid = num_nodes + i + 1; > + node_set(nid, node_possible_map); > + hidden_nodes[nid].start = hp_start + hp_size * i; > + hidden_nodes[nid].end = hp_start + hp_size * (i+1); > + node_set_hidden(nid); > + } > +} > +#endif /* CONFIG_NODE_HOTPLUG_EMU */ > + > #ifdef CONFIG_NUMA_EMU > /* Numa emulation */ > static struct bootnode nodes[MAX_NUMNODES] __initdata; > @@ -661,7 +742,7 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > > #ifdef CONFIG_NUMA_EMU > if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8)) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > @@ -669,14 +750,14 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > #ifdef CONFIG_ACPI_NUMA > if (!numa_off && acpi && !acpi_scan_nodes(start_pfn << PAGE_SHIFT, > last_pfn << PAGE_SHIFT)) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > > #ifdef CONFIG_K8_NUMA > if (!numa_off && k8 && !k8_scan_nodes()) > - return; > + goto done; > nodes_clear(node_possible_map); > nodes_clear(node_online_map); > #endif > @@ -696,6 +777,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn, > numa_set_node(i, 0); > e820_register_active_regions(0, start_pfn, last_pfn); > setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT); > +done: > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + if (hidden_num) > + numa_hotplug_emulation(); > +#endif > + return; > } > > unsigned long __init numa_free_all_bootmem(void) > @@ -723,6 +810,12 @@ static __init int numa_setup(char *opt) > if (!strncmp(opt, "fake=", 5)) > cmdline = opt + 5; > #endif > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + if (!strncmp(opt, "hide=", 5)) { > + hp_cmdline = opt + 5; > + numa_hide_nodes(); > + } > +#endif > #ifdef CONFIG_ACPI_NUMA > if (!strncmp(opt, "noacpi", 6)) > acpi_numa = -1; > diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h > index dba35e4..ba0f82d 100644 > --- a/include/linux/nodemask.h > +++ b/include/linux/nodemask.h > @@ -371,6 +371,10 @@ static inline void __nodes_fold(nodemask_t *dstp, const nodemask_t *origp, > */ > enum node_states { > N_POSSIBLE, /* The node could become online at some point */ > +#ifdef CONFIG_NODE_HOTPLUG_EMU > + N_HIDDEN, /* The node is hidden at booting time, could be > + * onlined in run time */ > +#endif > N_ONLINE, /* The node is online */ > N_NORMAL_MEMORY, /* The node has regular memory */ > #ifdef CONFIG_HIGHMEM > @@ -470,6 +474,13 @@ static inline int num_node_state(enum node_states state) > #define node_online(node) node_state((node), N_ONLINE) > #define node_possible(node) node_state((node), N_POSSIBLE) > > +#ifdef CONFIG_NODE_HOTPLUG_EMU > +#define node_set_hidden(node) node_set_state((node), N_HIDDEN) > +#define node_clear_hidden(node) node_clear_state((node), N_HIDDEN) > +#define node_hidden(node) node_state((node), N_HIDDEN) > +extern int hotadd_hidden_nodes(int nid); > +#endif > + > #define for_each_node(node) for_each_node_state(node, N_POSSIBLE) > #define for_each_online_node(node) for_each_node_state(node, N_ONLINE) > > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |