[FIX,v2,2/2] powerpc,numa: Fix memory_hotplug_max()
diff mbox

Message ID 1463060055-24158-3-git-send-email-bharata@linux.vnet.ibm.com
State Accepted
Headers show

Commit Message

Bharata B Rao May 12, 2016, 1:34 p.m. UTC
memory_hotplug_max() uses hot_add_drconf_memory_max() to get maxmimum
addressable memory by referring to ibm,dyanamic-memory property. There
are three problems with the current approach:

1 hot_add_drconf_memory_max() assumes that ibm,dynamic-memory includes
  all the LMBs of the guest, but that is not true for PowerKVM which
  populates only DR LMBs (LMBs that can be hotplugged/removed) in that
  property.
2 hot_add_drconf_memory_max() multiplies lmb-size with lmb-count to arrive
  at the max possible address. Since ibm,dynamic-memory doesn't include
  RMA LMBs, the address thus obtained will be less than the actual max
  address. For example, if max possible memory size is 32G, with lmb-size
  of 256MB there can be 127 LMBs in ibm,dynamic-memory (1 LMB for RMA
  which won't be present here).  hot_add_drconf_memory_max() would then
  return the max addressable memory as 127 * 256MB = 31.75GB, the max
  address should have been 32G which is what ibm,lrdr-capacity shows.
3 In PowerKVM, there can be a gap between the end of boot time RAM and
  beginning of hotplug RAM area. So just multiplying lmb-count with
  lmb-size will not provide the correct max possible address for PowerKVM.

This patch fixes 1 by using ibm,lrdr-capacity property to return the max
addressable memory whenever the property is present. Then it fixes 2 & 3
by fetching the address of the last LMB in ibm,dynamic-memory property.

Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
---
 arch/powerpc/mm/numa.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

Comments

David Gibson May 26, 2016, 12:50 a.m. UTC | #1
On Thu, May 12, 2016 at 07:04:15PM +0530, Bharata B Rao wrote:
> memory_hotplug_max() uses hot_add_drconf_memory_max() to get maxmimum
> addressable memory by referring to ibm,dyanamic-memory property. There
> are three problems with the current approach:
> 
> 1 hot_add_drconf_memory_max() assumes that ibm,dynamic-memory includes
>   all the LMBs of the guest, but that is not true for PowerKVM which
>   populates only DR LMBs (LMBs that can be hotplugged/removed) in that
>   property.
> 2 hot_add_drconf_memory_max() multiplies lmb-size with lmb-count to arrive
>   at the max possible address. Since ibm,dynamic-memory doesn't include
>   RMA LMBs, the address thus obtained will be less than the actual max
>   address. For example, if max possible memory size is 32G, with lmb-size
>   of 256MB there can be 127 LMBs in ibm,dynamic-memory (1 LMB for RMA
>   which won't be present here).  hot_add_drconf_memory_max() would then
>   return the max addressable memory as 127 * 256MB = 31.75GB, the max
>   address should have been 32G which is what ibm,lrdr-capacity shows.
> 3 In PowerKVM, there can be a gap between the end of boot time RAM and
>   beginning of hotplug RAM area. So just multiplying lmb-count with
>   lmb-size will not provide the correct max possible address for PowerKVM.
> 
> This patch fixes 1 by using ibm,lrdr-capacity property to return the max
> addressable memory whenever the property is present. Then it fixes 2 & 3
> by fetching the address of the last LMB in ibm,dynamic-memory property.
> 
> Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
>  arch/powerpc/mm/numa.c | 18 +++++++++++++++++-
>  1 file changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
> index 4a87ccb..f8b1da7 100644
> --- a/arch/powerpc/mm/numa.c
> +++ b/arch/powerpc/mm/numa.c
> @@ -1164,17 +1164,33 @@ int hot_add_scn_to_nid(unsigned long scn_addr)
>  static u64 hot_add_drconf_memory_max(void)
>  {
>  	struct device_node *memory = NULL;
> +	struct device_node *dn = NULL;
>  	unsigned int drconf_cell_cnt = 0;
>  	u64 lmb_size = 0;
>  	const __be32 *dm = NULL;
> +	const __be64 *lrdr = NULL;
> +	struct of_drconf_cell drmem;
> +
> +	dn = of_find_node_by_path("/rtas");
> +	if (dn) {
> +		lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL);
> +		of_node_put(dn);
> +		if (lrdr)
> +			return be64_to_cpup(lrdr);
> +	}
>  
>  	memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
>  	if (memory) {
>  		drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
>  		lmb_size = of_get_lmb_size(memory);
> +
> +		/* Advance to the last cell, each cell has 6 32 bit integers */
> +		dm += (drconf_cell_cnt - 1) * 6;
> +		read_drconf_cell(&drmem, &dm);
>  		of_node_put(memory);
> +		return drmem.base_addr + lmb_size;
>  	}
> -	return lmb_size * drconf_cell_cnt;
> +	return 0;
>  }
>  
>  /*
Michael Ellerman June 21, 2016, 12:40 a.m. UTC | #2
On Thu, 2016-12-05 at 13:34:15 UTC, Bharata B Rao wrote:
> memory_hotplug_max() uses hot_add_drconf_memory_max() to get maxmimum
> addressable memory by referring to ibm,dyanamic-memory property. There
> are three problems with the current approach:
> 
> 1 hot_add_drconf_memory_max() assumes that ibm,dynamic-memory includes
>   all the LMBs of the guest, but that is not true for PowerKVM which
>   populates only DR LMBs (LMBs that can be hotplugged/removed) in that
>   property.
> 2 hot_add_drconf_memory_max() multiplies lmb-size with lmb-count to arrive
>   at the max possible address. Since ibm,dynamic-memory doesn't include
>   RMA LMBs, the address thus obtained will be less than the actual max
>   address. For example, if max possible memory size is 32G, with lmb-size
>   of 256MB there can be 127 LMBs in ibm,dynamic-memory (1 LMB for RMA
>   which won't be present here).  hot_add_drconf_memory_max() would then
>   return the max addressable memory as 127 * 256MB = 31.75GB, the max
>   address should have been 32G which is what ibm,lrdr-capacity shows.
> 3 In PowerKVM, there can be a gap between the end of boot time RAM and
>   beginning of hotplug RAM area. So just multiplying lmb-count with
>   lmb-size will not provide the correct max possible address for PowerKVM.
> 
> This patch fixes 1 by using ibm,lrdr-capacity property to return the max
> addressable memory whenever the property is present. Then it fixes 2 & 3
> by fetching the address of the last LMB in ibm,dynamic-memory property.
> 
> Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/45b64ee64970dee9392229302e

cheers

Patch
diff mbox

diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 4a87ccb..f8b1da7 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -1164,17 +1164,33 @@  int hot_add_scn_to_nid(unsigned long scn_addr)
 static u64 hot_add_drconf_memory_max(void)
 {
 	struct device_node *memory = NULL;
+	struct device_node *dn = NULL;
 	unsigned int drconf_cell_cnt = 0;
 	u64 lmb_size = 0;
 	const __be32 *dm = NULL;
+	const __be64 *lrdr = NULL;
+	struct of_drconf_cell drmem;
+
+	dn = of_find_node_by_path("/rtas");
+	if (dn) {
+		lrdr = of_get_property(dn, "ibm,lrdr-capacity", NULL);
+		of_node_put(dn);
+		if (lrdr)
+			return be64_to_cpup(lrdr);
+	}
 
 	memory = of_find_node_by_path("/ibm,dynamic-reconfiguration-memory");
 	if (memory) {
 		drconf_cell_cnt = of_get_drconf_memory(memory, &dm);
 		lmb_size = of_get_lmb_size(memory);
+
+		/* Advance to the last cell, each cell has 6 32 bit integers */
+		dm += (drconf_cell_cnt - 1) * 6;
+		read_drconf_cell(&drmem, &dm);
 		of_node_put(memory);
+		return drmem.base_addr + lmb_size;
 	}
-	return lmb_size * drconf_cell_cnt;
+	return 0;
 }
 
 /*