Skip to contentSkip to conversation

cxl/region: Allow 6 & 12 way regions on 3-way HB interleaves

PATCH Superseded [email protected] 4 participants 7 messages Feb 24, 2025 – Mar 6, 2025 CXL

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 cxl/region: Allow 6 & 12 way regions on 3-way HB interleaves

From: Alison Schofield <[email protected]>

The CXL driver requires the granularity of a region and it's root decoder to be the same. This is particularly restrictive for 3-way host bridge interleaves where the only spec defined interleave configurations for creating 6-way and 12-way regions on a 3-way HB interleave require mixed granularities.

CXL 3.1 Specification 9.13.1.1:
Legal Interleaving Configurations: 12-way, 6-way, and 3-way

Adding support for these new interleaves touches these areas:

1) User created regions employing "cxl create-region" fail when the ndctl tool gets a mixed granularity request. That is addressed in a patch to the ndctl tool.

2) User created regions employing sysfs directly fail at the sysfs store of the non-matching region granularity. That restriction is lifted here. Note that the driver immediately allows any reasonable 3-way HB granularity, but the region config may still fail later if the ways/gran between region and root decoder are incompatible.

3) Auto created regions, in which the drivers role is to reflect the BIOS created region and provide RAS support, basically sneak on by. The driver ignores the root decoders granularity and assumes it is the same as the regions. The impact being that endpoint devices appear out of order and DPA to HPA address translations that depend on that order are invalid. Here the driver stops making that same granularity assumption and checks for an allowable granularity.

A new helper, interleave_granularity_allow(), is used in both the user and auto creation path to allow the newly supported configs.

Another new helper, gran_multiple(), is used in sorting, target list searching, and distance calculations, supporting the new root granularity to region granularity multiples.

And, as noted in 2), there's an added check to see if the selected sysfs ways/gran make sense when considered together.

---
 drivers/cxl/core/region.c | 101 +++++++++++++++++++++++++++++++-------
 1 file changed, 82 insertions(+), 19 deletions(-)
drivers/cxl/core/region.c +82 -19
index 3cb91cf0e2dd..9d4716b7fcf8 100644
@@ -532,6 +532,31 @@ static ssize_t interleave_granularity_show(struct device *dev,
 	return rc;
 }
 
+static bool interleave_granularity_allow(struct cxl_decoder *cxld, u16 ig)
+{
+	/*
+	 * When the host-bridge is interleaved, disallow region granularity
+	 * != root granularity with the exception of 3-way HB interleaves.
+	 * Allow the CXL Spec defined 3-way HB interleaves that can only be
+	 * configured when host-bridge interleave is greater that the
+	 * region interleave. (CXL 3.1 Specification 9.13.1.1)
+	 * Allow 2+2+2 interleave where HB gran is 2 * region granularity
+	 *	 4+4+4 interleave where HB gran is 4 * region granularity
+	 *
+	 * Regions with a granularity greater than the root interleave result
+	 * in invalid DPA translations (invalid to support).
+	 */
+	if (cxld->interleave_ways > 1 && ig != cxld->interleave_granularity) {
+		if (cxld->interleave_ways != 3)
+			return false;
+
+		if (cxld->interleave_granularity % (2 * ig) &&
+		    cxld->interleave_granularity % (4 * ig))
+			return false;
+	}
+	return true;
+}
+
 static ssize_t interleave_granularity_store(struct device *dev,
 					    struct device_attribute *attr,
 					    const char *buf, size_t len)
@@ -551,15 +576,7 @@ static ssize_t interleave_granularity_store(struct device *dev,
 	if (rc)
 		return rc;
 
-	/*
-	 * When the host-bridge is interleaved, disallow region granularity !=
-	 * root granularity. Regions with a granularity less than the root
-	 * interleave result in needing multiple endpoints to support a single
-	 * slot in the interleave (possible to support in the future). Regions
-	 * with a granularity greater than the root interleave result in invalid
-	 * DPA translations (invalid to support).
-	 */
-	if (cxld->interleave_ways > 1 && val != cxld->interleave_granularity)
+	if (!interleave_granularity_allow(cxld, val))
 		return -EINVAL;
 
 	rc = down_write_killable(&cxl_region_rwsem);
@@ -1290,6 +1307,23 @@ static int check_interleave_cap(struct cxl_decoder *cxld, int iw, int ig)
 	return 0;
 }
 
+static int gran_multiple(struct cxl_region *cxlr)
+{
+	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(cxlr->dev.parent);
+	int root_decoder_gran = cxlrd->cxlsd.cxld.interleave_granularity;
+	int region_gran = cxlr->params.interleave_granularity;
+
+	/*
+	 * In regions built upon 3-way HB interleaves, the root decoder
+	 * granularity can be a multiple of the region granularity. The
+	 * multiple value is used in sorting and distance calculations.
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways != 3 || !root_decoder_gran)
+		return 1;
+
+	return root_decoder_gran / region_gran;
+}
+
 static int cxl_port_setup_targets(struct cxl_port *port,
 				  struct cxl_region *cxlr,
 				  struct cxl_endpoint_decoder *cxled)
@@ -1321,6 +1355,7 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 	cxlsd = to_cxl_switch_decoder(&cxld->dev);
 	if (cxl_rr->nr_targets_set) {
 		int i, distance = 1;
+		int multiple = gran_multiple(cxlr);
 		struct cxl_region_ref *cxl_rr_iter;
 
 		/*
@@ -1332,12 +1367,17 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 		 * always 1 as every index targets a different host-bridge. At
 		 * each subsequent switch level those ports map every Nth region
 		 * position where N is the width of the switch == distance.
+		 *
+		 * With the introduction of mixed granularities in 3-way HB
+		 * interleaves, divide N by a multiple that represents the root
+		 * decoder to region granularity.
 		 */
 		do {
 			cxl_rr_iter = cxl_rr_load(iter, cxlr);
 			distance *= cxl_rr_iter->nr_targets;
 			iter = to_cxl_port(iter->dev.parent);
 		} while (!is_cxl_root(iter));
+		distance /= multiple;
 		distance *= cxlrd->cxlsd.cxld.interleave_ways;
 
 		for (i = 0; i < cxl_rr->nr_targets_set; i++)
@@ -1354,12 +1394,9 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 	if (is_cxl_root(parent_port)) {
 		/*
 		 * Root decoder IG is always set to value in CFMWS which
-		 * may be different than this region's IG.  We can use the
-		 * region's IG here since interleave_granularity_store()
-		 * does not allow interleaved host-bridges with
-		 * root IG != region IG.
+		 * may be different than this region's IG.
 		 */
-		parent_ig = p->interleave_granularity;
+		parent_ig = cxlrd->cxlsd.cxld.interleave_granularity;
 		parent_iw = cxlrd->cxlsd.cxld.interleave_ways;
 		/*
 		 * For purposes of address bit routing, use power-of-2 math for
@@ -1431,7 +1468,7 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 
 	if (test_bit(CXL_REGION_F_AUTO, &cxlr->flags)) {
 		if (cxld->interleave_ways != iw ||
-		    cxld->interleave_granularity != ig ||
+		    !interleave_granularity_allow(cxld, ig) ||
 		    cxld->hpa_range.start != p->res->start ||
 		    cxld->hpa_range.end != p->res->end ||
 		    ((cxld->flags & CXL_DECODER_F_ENABLE) == 0)) {
@@ -1661,11 +1698,12 @@ static int cxl_region_attach_position(struct cxl_region *cxlr,
 	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
 	struct cxl_switch_decoder *cxlsd = &cxlrd->cxlsd;
 	struct cxl_decoder *cxld = &cxlsd->cxld;
+	int multiple = gran_multiple(cxlr);
 	int iw = cxld->interleave_ways;
 	struct cxl_port *iter;
 	int rc;
 
-	if (dport != cxlrd->cxlsd.target[pos % iw]) {
+	if (dport != cxlrd->cxlsd.target[pos / multiple % iw]) {
 		dev_dbg(&cxlr->dev, "%s:%s invalid target position for %s\n",
 			dev_name(&cxlmd->dev), dev_name(&cxled->cxld.dev),
 			dev_name(&cxlrd->cxlsd.cxld.dev));
@@ -1809,7 +1847,8 @@ static int find_pos_and_ways(struct cxl_port *port, struct range *range,
  * Return: position >= 0 on success
  *	   -ENXIO on failure
  */
-static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
+static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled,
+				   int multiple)
 {
 	struct cxl_port *iter, *port = cxled_to_port(cxled);
 	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
@@ -1855,6 +1894,11 @@ static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
 		if (rc)
 			return rc;
 
+		if (multiple > 1 && is_cxl_root(next_port(iter))) {
+			pos = pos + multiple * parent_pos;
+			break;
+		}
+
 		pos = pos * parent_ways + parent_pos;
 	}
 
@@ -1869,12 +1913,13 @@ static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
 static int cxl_region_sort_targets(struct cxl_region *cxlr)
 {
 	struct cxl_region_params *p = &cxlr->params;
+	int multiple = gran_multiple(cxlr);
 	int i, rc = 0;
 
 	for (i = 0; i < p->nr_targets; i++) {
 		struct cxl_endpoint_decoder *cxled = p->targets[i];
 
-		cxled->pos = cxl_calc_interleave_pos(cxled);
+		cxled->pos = cxl_calc_interleave_pos(cxled, multiple);
 		/*
 		 * Record that sorting failed, but still continue to calc
 		 * cxled->pos so that follow-on code paths can reliably
@@ -1902,6 +1947,23 @@ static int cxl_region_attach(struct cxl_region *cxlr,
 	struct cxl_dport *dport;
 	int rc = -ENXIO;
 
+	/*
+	 * Protect against improper gran mixes on 3-way HB interleave
+	 * Expect decoder gran*ways == region gran*ways
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways == 3) {
+		if ((cxlrd->cxlsd.cxld.interleave_granularity * 3) !=
+		    (p->interleave_ways * p->interleave_granularity)) {
+			dev_dbg(&cxlr->dev,
+				"invalid config region:%d/%d decoder %d/%d\n",
+				p->interleave_ways, p->interleave_granularity,
+				cxlrd->cxlsd.cxld.interleave_ways,
+				cxlrd->cxlsd.cxld.interleave_granularity);
+
+			return rc;
+		}
+	}
+
 	rc = check_interleave_cap(&cxled->cxld, p->interleave_ways,
 				  p->interleave_granularity);
 	if (rc) {
@@ -2055,9 +2117,10 @@ static int cxl_region_attach(struct cxl_region *cxlr,
 	 */
 	for (int i = 0; i < p->nr_targets; i++) {
 		struct cxl_endpoint_decoder *cxled = p->targets[i];
+		int multiple = gran_multiple(cxlr);
 		int test_pos;
 
-		test_pos = cxl_calc_interleave_pos(cxled);
+		test_pos = cxl_calc_interleave_pos(cxled, multiple);
 		dev_dbg(&cxled->cxld.dev,
 			"Test cxl_calc_interleave_pos(): %s test_pos:%d cxled->pos:%d\n",
 			(test_pos == cxled->pos) ? "success" : "fail",

base-commit: 2bb67004903cfd35710750654669a77e7223fcd1

Signature · 1 line
2.37.3
maintainer
#1 lore raw
Quoted 3 lines · On 2/24/25 4:58 PM, [email protected] wrote:

From: Alison Schofield <[email protected]>

The CXL driver requires the granularity of a region and it's root

s/it's/its/

Quoted 3 lines

decoder to be the same. This is particularly restrictive for 3-way host bridge interleaves where the only spec defined interleave configurations for creating 6-way and 12-way regions on a 3-way HB

configuration?

Quoted 1 line

interleave require mixed granularities.

requries?

Quoted 2 lines

CXL 3.1 Specification 9.13.1.1:

May as well go to 3.2

Quoted 6 lines

Legal Interleaving Configurations: 12-way, 6-way, and 3-way

Adding support for these new interleaves touches these areas:

1) User created regions employing "cxl create-region" fail when the ndctl tool gets a mixed granularity request. That is addressed in

s/ndctl/CXL CLI/

Quoted 37 lines

a patch to the ndctl tool.

2) User created regions employing sysfs directly fail at the sysfs store of the non-matching region granularity. That restriction is lifted here. Note that the driver immediately allows any reasonable 3-way HB granularity, but the region config may still fail later if the ways/gran between region and root decoder are incompatible.

3) Auto created regions, in which the drivers role is to reflect the BIOS created region and provide RAS support, basically sneak on by. The driver ignores the root decoders granularity and assumes it is the same as the regions. The impact being that endpoint devices appear out of order and DPA to HPA address translations that depend on that order are invalid. Here the driver stops making that same granularity assumption and checks for an allowable granularity.

A new helper, interleave_granularity_allow(), is used in both the user and auto creation path to allow the newly supported configs.

Another new helper, gran_multiple(), is used in sorting, target list searching, and distance calculations, supporting the new root granularity to region granularity multiples.

And, as noted in 2), there's an added check to see if the selected sysfs ways/gran make sense when considered together.

---
 drivers/cxl/core/region.c | 101 +++++++++++++++++++++++++++++++-------
 1 file changed, 82 insertions(+), 19 deletions(-)
drivers/cxl/core/region.c +0 -0
index 3cb91cf0e2dd..9d4716b7fcf8 100644
@@ -532,6 +532,31 @@ static ssize_t interleave_granularity_show(struct device *dev,
 	return rc;

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 542

Quoted 10 lines
drivers/cxl/core/region.c +8 -0
 }
 
+static bool interleave_granularity_allow(struct cxl_decoder *cxld, u16 ig)
+{
+	/*
+	 * When the host-bridge is interleaved, disallow region granularity
+	 * != root granularity with the exception of 3-way HB interleaves.
+	 * Allow the CXL Spec defined 3-way HB interleaves that can only be
+	 * configured when host-bridge interleave is greater that the
+	 * region interleave. (CXL 3.1 Specification 9.13.1.1)

spec 3.2

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 555

Quoted 13 lines
drivers/cxl/core/region.c +13 -0
+	 * Allow 2+2+2 interleave where HB gran is 2 * region granularity
+	 *	 4+4+4 interleave where HB gran is 4 * region granularity
+	 *
+	 * Regions with a granularity greater than the root interleave result
+	 * in invalid DPA translations (invalid to support).
+	 */
+	if (cxld->interleave_ways > 1 && ig != cxld->interleave_granularity) {
+		if (cxld->interleave_ways != 3)
+			return false;
+
+		if (cxld->interleave_granularity % (2 * ig) &&
+		    cxld->interleave_granularity % (4 * ig))
+			return false;

Can you please explain how the math works here? I'm not understanding the code here vs the comments above.

DJ

Quoted 187 lines
drivers/cxl/core/region.c +61 -19
+	}
+	return true;
+}
+
 static ssize_t interleave_granularity_store(struct device *dev,
 					    struct device_attribute *attr,
 					    const char *buf, size_t len)
@@ -551,15 +576,7 @@ static ssize_t interleave_granularity_store(struct device *dev,
 	if (rc)
 		return rc;
 
-	/*
-	 * When the host-bridge is interleaved, disallow region granularity !=
-	 * root granularity. Regions with a granularity less than the root
-	 * interleave result in needing multiple endpoints to support a single
-	 * slot in the interleave (possible to support in the future). Regions
-	 * with a granularity greater than the root interleave result in invalid
-	 * DPA translations (invalid to support).
-	 */
-	if (cxld->interleave_ways > 1 && val != cxld->interleave_granularity)
+	if (!interleave_granularity_allow(cxld, val))
 		return -EINVAL;
 
 	rc = down_write_killable(&cxl_region_rwsem);
@@ -1290,6 +1307,23 @@ static int check_interleave_cap(struct cxl_decoder *cxld, int iw, int ig)
 	return 0;
 }
 
+static int gran_multiple(struct cxl_region *cxlr)
+{
+	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(cxlr->dev.parent);
+	int root_decoder_gran = cxlrd->cxlsd.cxld.interleave_granularity;
+	int region_gran = cxlr->params.interleave_granularity;
+
+	/*
+	 * In regions built upon 3-way HB interleaves, the root decoder
+	 * granularity can be a multiple of the region granularity. The
+	 * multiple value is used in sorting and distance calculations.
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways != 3 || !root_decoder_gran)
+		return 1;
+
+	return root_decoder_gran / region_gran;
+}
+
 static int cxl_port_setup_targets(struct cxl_port *port,
 				  struct cxl_region *cxlr,
 				  struct cxl_endpoint_decoder *cxled)
@@ -1321,6 +1355,7 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 	cxlsd = to_cxl_switch_decoder(&cxld->dev);
 	if (cxl_rr->nr_targets_set) {
 		int i, distance = 1;
+		int multiple = gran_multiple(cxlr);
 		struct cxl_region_ref *cxl_rr_iter;
 
 		/*
@@ -1332,12 +1367,17 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 		 * always 1 as every index targets a different host-bridge. At
 		 * each subsequent switch level those ports map every Nth region
 		 * position where N is the width of the switch == distance.
+		 *
+		 * With the introduction of mixed granularities in 3-way HB
+		 * interleaves, divide N by a multiple that represents the root
+		 * decoder to region granularity.
 		 */
 		do {
 			cxl_rr_iter = cxl_rr_load(iter, cxlr);
 			distance *= cxl_rr_iter->nr_targets;
 			iter = to_cxl_port(iter->dev.parent);
 		} while (!is_cxl_root(iter));
+		distance /= multiple;
 		distance *= cxlrd->cxlsd.cxld.interleave_ways;
 
 		for (i = 0; i < cxl_rr->nr_targets_set; i++)
@@ -1354,12 +1394,9 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 	if (is_cxl_root(parent_port)) {
 		/*
 		 * Root decoder IG is always set to value in CFMWS which
-		 * may be different than this region's IG.  We can use the
-		 * region's IG here since interleave_granularity_store()
-		 * does not allow interleaved host-bridges with
-		 * root IG != region IG.
+		 * may be different than this region's IG.
 		 */
-		parent_ig = p->interleave_granularity;
+		parent_ig = cxlrd->cxlsd.cxld.interleave_granularity;
 		parent_iw = cxlrd->cxlsd.cxld.interleave_ways;
 		/*
 		 * For purposes of address bit routing, use power-of-2 math for
@@ -1431,7 +1468,7 @@ static int cxl_port_setup_targets(struct cxl_port *port,
 
 	if (test_bit(CXL_REGION_F_AUTO, &cxlr->flags)) {
 		if (cxld->interleave_ways != iw ||
-		    cxld->interleave_granularity != ig ||
+		    !interleave_granularity_allow(cxld, ig) ||
 		    cxld->hpa_range.start != p->res->start ||
 		    cxld->hpa_range.end != p->res->end ||
 		    ((cxld->flags & CXL_DECODER_F_ENABLE) == 0)) {
@@ -1661,11 +1698,12 @@ static int cxl_region_attach_position(struct cxl_region *cxlr,
 	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
 	struct cxl_switch_decoder *cxlsd = &cxlrd->cxlsd;
 	struct cxl_decoder *cxld = &cxlsd->cxld;
+	int multiple = gran_multiple(cxlr);
 	int iw = cxld->interleave_ways;
 	struct cxl_port *iter;
 	int rc;
 
-	if (dport != cxlrd->cxlsd.target[pos % iw]) {
+	if (dport != cxlrd->cxlsd.target[pos / multiple % iw]) {
 		dev_dbg(&cxlr->dev, "%s:%s invalid target position for %s\n",
 			dev_name(&cxlmd->dev), dev_name(&cxled->cxld.dev),
 			dev_name(&cxlrd->cxlsd.cxld.dev));
@@ -1809,7 +1847,8 @@ static int find_pos_and_ways(struct cxl_port *port, struct range *range,
  * Return: position >= 0 on success
  *	   -ENXIO on failure
  */
-static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
+static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled,
+				   int multiple)
 {
 	struct cxl_port *iter, *port = cxled_to_port(cxled);
 	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
@@ -1855,6 +1894,11 @@ static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
 		if (rc)
 			return rc;
 
+		if (multiple > 1 && is_cxl_root(next_port(iter))) {
+			pos = pos + multiple * parent_pos;
+			break;
+		}
+
 		pos = pos * parent_ways + parent_pos;
 	}
 
@@ -1869,12 +1913,13 @@ static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled)
 static int cxl_region_sort_targets(struct cxl_region *cxlr)
 {
 	struct cxl_region_params *p = &cxlr->params;
+	int multiple = gran_multiple(cxlr);
 	int i, rc = 0;
 
 	for (i = 0; i < p->nr_targets; i++) {
 		struct cxl_endpoint_decoder *cxled = p->targets[i];
 
-		cxled->pos = cxl_calc_interleave_pos(cxled);
+		cxled->pos = cxl_calc_interleave_pos(cxled, multiple);
 		/*
 		 * Record that sorting failed, but still continue to calc
 		 * cxled->pos so that follow-on code paths can reliably
@@ -1902,6 +1947,23 @@ static int cxl_region_attach(struct cxl_region *cxlr,
 	struct cxl_dport *dport;
 	int rc = -ENXIO;
 
+	/*
+	 * Protect against improper gran mixes on 3-way HB interleave
+	 * Expect decoder gran*ways == region gran*ways
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways == 3) {
+		if ((cxlrd->cxlsd.cxld.interleave_granularity * 3) !=
+		    (p->interleave_ways * p->interleave_granularity)) {
+			dev_dbg(&cxlr->dev,
+				"invalid config region:%d/%d decoder %d/%d\n",
+				p->interleave_ways, p->interleave_granularity,
+				cxlrd->cxlsd.cxld.interleave_ways,
+				cxlrd->cxlsd.cxld.interleave_granularity);
+
+			return rc;
+		}
+	}
+
 	rc = check_interleave_cap(&cxled->cxld, p->interleave_ways,
 				  p->interleave_granularity);
 	if (rc) {
@@ -2055,9 +2117,10 @@ static int cxl_region_attach(struct cxl_region *cxlr,
 	 */
 	for (int i = 0; i < p->nr_targets; i++) {
 		struct cxl_endpoint_decoder *cxled = p->targets[i];
+		int multiple = gran_multiple(cxlr);
 		int test_pos;
 
-		test_pos = cxl_calc_interleave_pos(cxled);
+		test_pos = cxl_calc_interleave_pos(cxled, multiple);
 		dev_dbg(&cxled->cxld.dev,
 			"Test cxl_calc_interleave_pos(): %s test_pos:%d cxled->pos:%d\n",
 			(test_pos == cxled->pos) ? "success" : "fail",

base-commit: 2bb67004903cfd35710750654669a77e7223fcd1

author maintainer
#2 lore raw
Quoted 8 lines · Dave Jiang
Quoted 3 lines · On 2/24/25 4:58 PM, [email protected] wrote:

From: Alison Schofield <[email protected]>

The CXL driver requires the granularity of a region and it's root

s/it's/its/

got it

Quoted 7 lines
Quoted 4 lines

decoder to be the same. This is particularly restrictive for 3-way host bridge interleaves where the only spec defined interleave configurations for creating 6-way and 12-way regions on a 3-way HB

configuration?

Quoted 1 line

interleave require mixed granularities.

requries?

considering that

Quoted 6 lines
Quoted 3 lines

CXL 3.1 Specification 9.13.1.1:

May as well go to 3.2

got it

Quoted 8 lines
Quoted 6 lines

Legal Interleaving Configurations: 12-way, 6-way, and 3-way

Adding support for these new interleaves touches these areas:

1) User created regions employing "cxl create-region" fail when the ndctl tool gets a mixed granularity request. That is addressed in

s/ndctl/CXL CLI/

got it

Quoted 2 lines
Quoted 2 lines

a patch to the ndctl tool.

snip

Quoted 11 lines

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 542

Quoted 9 lines
drivers/cxl/core/region.c +8 -0
+static bool interleave_granularity_allow(struct cxl_decoder *cxld, u16 ig)
+{
+	/*
+	 * When the host-bridge is interleaved, disallow region granularity
+	 * != root granularity with the exception of 3-way HB interleaves.
+	 * Allow the CXL Spec defined 3-way HB interleaves that can only be
+	 * configured when host-bridge interleave is greater that the
+	 * region interleave. (CXL 3.1 Specification 9.13.1.1)

spec 3.2

Quoted 15 lines

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 555

Quoted 13 lines
drivers/cxl/core/region.c +13 -0
+	 * Allow 2+2+2 interleave where HB gran is 2 * region granularity
+	 *	 4+4+4 interleave where HB gran is 4 * region granularity
+	 *
+	 * Regions with a granularity greater than the root interleave result
+	 * in invalid DPA translations (invalid to support).
+	 */
+	if (cxld->interleave_ways > 1 && ig != cxld->interleave_granularity) {
+		if (cxld->interleave_ways != 3)
+			return false;
+
+		if (cxld->interleave_granularity % (2 * ig) &&
+		    cxld->interleave_granularity % (4 * ig))
+			return false;

Can you please explain how the math works here? I'm not understanding the code here vs the comments above.

I'm guessing you are talking about the last chunk only. We get there after discovering a 3-way HB interleave. (HB interleave and root decoder interleave are the same thing.) Once we know it's a 3-way, there are only 2 possible values and those are defined in the spec as:
2+2+2 interleave where HB gran is 2 * region granularity 4+4+4 interleave where HB gran is 4 * region granularity

And now that I've gotten to this part of the explanation I think the code is allowing multiples of 2*ig and 4*ig, when it should be a stricter check for exactly 2*ig or 4*ig.

updating that to:

	if (cxld->interleave_granularity != 2 * ig &&
	    cxld->interleave_granularity != 4 * ig)
		return false;

Thanks for that Dave!

Quoted 3 lines

DJ

snip

maintainer
#3 lore raw
Quoted 83 lines · Alison Schofield
Quoted 8 lines · Dave Jiang
Quoted 3 lines · On 2/24/25 4:58 PM, [email protected] wrote:

From: Alison Schofield <[email protected]>

The CXL driver requires the granularity of a region and it's root

s/it's/its/

got it

Quoted 7 lines
Quoted 4 lines

decoder to be the same. This is particularly restrictive for 3-way host bridge interleaves where the only spec defined interleave configurations for creating 6-way and 12-way regions on a 3-way HB

configuration?

Quoted 1 line

interleave require mixed granularities.

requries?

considering that

Quoted 6 lines
Quoted 3 lines

CXL 3.1 Specification 9.13.1.1:

May as well go to 3.2

got it

Quoted 8 lines
Quoted 6 lines

Legal Interleaving Configurations: 12-way, 6-way, and 3-way

Adding support for these new interleaves touches these areas:

1) User created regions employing "cxl create-region" fail when the ndctl tool gets a mixed granularity request. That is addressed in

s/ndctl/CXL CLI/

got it

Quoted 2 lines
Quoted 2 lines

a patch to the ndctl tool.

snip

Quoted 11 lines

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 542

Quoted 9 lines
drivers/cxl/core/region.c +8 -0
+static bool interleave_granularity_allow(struct cxl_decoder *cxld, u16 ig)
+{
+	/*
+	 * When the host-bridge is interleaved, disallow region granularity
+	 * != root granularity with the exception of 3-way HB interleaves.
+	 * Allow the CXL Spec defined 3-way HB interleaves that can only be
+	 * configured when host-bridge interleave is greater that the
+	 * region interleave. (CXL 3.1 Specification 9.13.1.1)

spec 3.2

Quoted 15 lines

drivers/cxl/core/region.c static ssize_t interleave_granularity_show(struct device *dev, · line 555

Quoted 13 lines
drivers/cxl/core/region.c +13 -0
+	 * Allow 2+2+2 interleave where HB gran is 2 * region granularity
+	 *	 4+4+4 interleave where HB gran is 4 * region granularity
+	 *
+	 * Regions with a granularity greater than the root interleave result
+	 * in invalid DPA translations (invalid to support).
+	 */
+	if (cxld->interleave_ways > 1 && ig != cxld->interleave_granularity) {
+		if (cxld->interleave_ways != 3)
+			return false;
+
+		if (cxld->interleave_granularity % (2 * ig) &&
+		    cxld->interleave_granularity % (4 * ig))
+			return false;

Can you please explain how the math works here? I'm not understanding the code here vs the comments above.

I'm guessing you are talking about the last chunk only. We get there after discovering a 3-way HB interleave. (HB interleave and root decoder interleave are the same thing.) Once we know it's a 3-way, there are only 2 possible values and those are defined in the spec as:
2+2+2 interleave where HB gran is 2 * region granularity 4+4+4 interleave where HB gran is 4 * region granularity

Quoted 9 lines

And now that I've gotten to this part of the explanation I think the code is allowing multiples of 2*ig and 4*ig, when it should be a stricter check for exactly 2*ig or 4*ig.

updating that to:

	if (cxld->interleave_granularity != 2 * ig &&
	    cxld->interleave_granularity != 4 * ig)
		return false;

Ok the new code make sense to me now. :)

Quoted 6 lines

Thanks for that Dave!

Quoted 3 lines

DJ

snip

#4 lore raw
In reply to [email protected]
kernel test robot noticed the following build warnings:

Hi,

kernel test robot noticed the following build warnings:

[auto build test WARNING on 2bb67004903cfd35710750654669a77e7223fcd1]

url:    https://github.com/intel-lab-lkp/linux/commits/alison-schofield-intel-com/cxl-region-Allow-6-12-way-regions-on-3-way-HB-interleaves/20250225-080107
base:   2bb67004903cfd35710750654669a77e7223fcd1
patch link:    https://lore.kernel.org/r/20250224235817.2259508-1-alison.schofield%40intel.com
patch subject: [PATCH] cxl/region: Allow 6 & 12 way regions on 3-way HB interleaves
config: riscv-randconfig-001-20250225 (https://download.01.org/0day-ci/archive/20250225/[email protected]/config)
compiler: clang version 15.0.7 (https://github.com/llvm/llvm-project 8dfdcc7b7bf66834a761bd8de445840ef68e4d1a)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250225/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All warnings (new ones prefixed by >>):

>> drivers/cxl/core/region.c:1852: warning: Function parameter or struct member 'multiple' not described in 'cxl_calc_interleave_pos'

vim +1852 drivers/cxl/core/region.c

a3e00c964fb9439 Alison Schofield 2023-10-27  1834  
a3e00c964fb9439 Alison Schofield 2023-10-27  1835  /**
a3e00c964fb9439 Alison Schofield 2023-10-27  1836   * cxl_calc_interleave_pos() - calculate an endpoint position in a region
a3e00c964fb9439 Alison Schofield 2023-10-27  1837   * @cxled: endpoint decoder member of given region
a3e00c964fb9439 Alison Schofield 2023-10-27  1838   *
a3e00c964fb9439 Alison Schofield 2023-10-27  1839   * The endpoint position is calculated by traversing the topology from
a3e00c964fb9439 Alison Schofield 2023-10-27  1840   * the endpoint to the root decoder and iteratively applying this
a3e00c964fb9439 Alison Schofield 2023-10-27  1841   * calculation:
a3e00c964fb9439 Alison Schofield 2023-10-27  1842   *
a3e00c964fb9439 Alison Schofield 2023-10-27  1843   *    position = position * parent_ways + parent_pos;
a3e00c964fb9439 Alison Schofield 2023-10-27  1844   *
a3e00c964fb9439 Alison Schofield 2023-10-27  1845   * ...where @position is inferred from switch and root decoder target lists.
a3e00c964fb9439 Alison Schofield 2023-10-27  1846   *
a3e00c964fb9439 Alison Schofield 2023-10-27  1847   * Return: position >= 0 on success
a3e00c964fb9439 Alison Schofield 2023-10-27  1848   *	   -ENXIO on failure
a3e00c964fb9439 Alison Schofield 2023-10-27  1849   */
22639743643957e Alison Schofield 2025-02-24  1850  static int cxl_calc_interleave_pos(struct cxl_endpoint_decoder *cxled,
22639743643957e Alison Schofield 2025-02-24  1851  				   int multiple)
a3e00c964fb9439 Alison Schofield 2023-10-27 @1852  {
a3e00c964fb9439 Alison Schofield 2023-10-27  1853  	struct cxl_port *iter, *port = cxled_to_port(cxled);
a3e00c964fb9439 Alison Schofield 2023-10-27  1854  	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
a3e00c964fb9439 Alison Schofield 2023-10-27  1855  	struct range *range = &cxled->cxld.hpa_range;
a3e00c964fb9439 Alison Schofield 2023-10-27  1856  	int parent_ways = 0, parent_pos = 0, pos = 0;
a3e00c964fb9439 Alison Schofield 2023-10-27  1857  	int rc;
a3e00c964fb9439 Alison Schofield 2023-10-27  1858  
a3e00c964fb9439 Alison Schofield 2023-10-27  1859  	/*
a3e00c964fb9439 Alison Schofield 2023-10-27  1860  	 * Example: the expected interleave order of the 4-way region shown
a3e00c964fb9439 Alison Schofield 2023-10-27  1861  	 * below is: mem0, mem2, mem1, mem3
a3e00c964fb9439 Alison Schofield 2023-10-27  1862  	 *
a3e00c964fb9439 Alison Schofield 2023-10-27  1863  	 *		  root_port
a3e00c964fb9439 Alison Schofield 2023-10-27  1864  	 *                 /      \
a3e00c964fb9439 Alison Schofield 2023-10-27  1865  	 *      host_bridge_0    host_bridge_1
a3e00c964fb9439 Alison Schofield 2023-10-27  1866  	 *        |    |           |    |
a3e00c964fb9439 Alison Schofield 2023-10-27  1867  	 *       mem0 mem1        mem2 mem3
a3e00c964fb9439 Alison Schofield 2023-10-27  1868  	 *
a3e00c964fb9439 Alison Schofield 2023-10-27  1869  	 * In the example the calculator will iterate twice. The first iteration
a3e00c964fb9439 Alison Schofield 2023-10-27  1870  	 * uses the mem position in the host-bridge and the ways of the host-
a3e00c964fb9439 Alison Schofield 2023-10-27  1871  	 * bridge to generate the first, or local, position. The second
a3e00c964fb9439 Alison Schofield 2023-10-27  1872  	 * iteration uses the host-bridge position in the root_port and the ways
a3e00c964fb9439 Alison Schofield 2023-10-27  1873  	 * of the root_port to refine the position.
a3e00c964fb9439 Alison Schofield 2023-10-27  1874  	 *
a3e00c964fb9439 Alison Schofield 2023-10-27  1875  	 * A trace of the calculation per endpoint looks like this:
a3e00c964fb9439 Alison Schofield 2023-10-27  1876  	 * mem0: pos = 0 * 2 + 0    mem2: pos = 0 * 2 + 0
a3e00c964fb9439 Alison Schofield 2023-10-27  1877  	 *       pos = 0 * 2 + 0          pos = 0 * 2 + 1
a3e00c964fb9439 Alison Schofield 2023-10-27  1878  	 *       pos: 0                   pos: 1
a3e00c964fb9439 Alison Schofield 2023-10-27  1879  	 *
a3e00c964fb9439 Alison Schofield 2023-10-27  1880  	 * mem1: pos = 0 * 2 + 1    mem3: pos = 0 * 2 + 1
a3e00c964fb9439 Alison Schofield 2023-10-27  1881  	 *       pos = 1 * 2 + 0          pos = 1 * 2 + 1
a3e00c964fb9439 Alison Schofield 2023-10-27  1882  	 *       pos: 2                   pos = 3
a3e00c964fb9439 Alison Schofield 2023-10-27  1883  	 *
a3e00c964fb9439 Alison Schofield 2023-10-27  1884  	 * Note that while this example is simple, the method applies to more
a3e00c964fb9439 Alison Schofield 2023-10-27  1885  	 * complex topologies, including those with switches.
a3e00c964fb9439 Alison Schofield 2023-10-27  1886  	 */
a3e00c964fb9439 Alison Schofield 2023-10-27  1887  
a3e00c964fb9439 Alison Schofield 2023-10-27  1888  	/* Iterate from endpoint to root_port refining the position */
a3e00c964fb9439 Alison Schofield 2023-10-27  1889  	for (iter = port; iter; iter = next_port(iter)) {
a3e00c964fb9439 Alison Schofield 2023-10-27  1890  		if (is_cxl_root(iter))
a3e00c964fb9439 Alison Schofield 2023-10-27  1891  			break;
a3e00c964fb9439 Alison Schofield 2023-10-27  1892  
a3e00c964fb9439 Alison Schofield 2023-10-27  1893  		rc = find_pos_and_ways(iter, range, &parent_pos, &parent_ways);
a3e00c964fb9439 Alison Schofield 2023-10-27  1894  		if (rc)
a3e00c964fb9439 Alison Schofield 2023-10-27  1895  			return rc;
a3e00c964fb9439 Alison Schofield 2023-10-27  1896  
22639743643957e Alison Schofield 2025-02-24  1897  		if (multiple > 1 && is_cxl_root(next_port(iter))) {
22639743643957e Alison Schofield 2025-02-24  1898  			pos = pos + multiple * parent_pos;
22639743643957e Alison Schofield 2025-02-24  1899  			break;
22639743643957e Alison Schofield 2025-02-24  1900  		}
22639743643957e Alison Schofield 2025-02-24  1901  
a3e00c964fb9439 Alison Schofield 2023-10-27  1902  		pos = pos * parent_ways + parent_pos;
a3e00c964fb9439 Alison Schofield 2023-10-27  1903  	}
a3e00c964fb9439 Alison Schofield 2023-10-27  1904  
a3e00c964fb9439 Alison Schofield 2023-10-27  1905  	dev_dbg(&cxlmd->dev,
a3e00c964fb9439 Alison Schofield 2023-10-27  1906  		"decoder:%s parent:%s port:%s range:%#llx-%#llx pos:%d\n",
a3e00c964fb9439 Alison Schofield 2023-10-27  1907  		dev_name(&cxled->cxld.dev), dev_name(cxlmd->dev.parent),
a3e00c964fb9439 Alison Schofield 2023-10-27  1908  		dev_name(&port->dev), range->start, range->end, pos);
a3e00c964fb9439 Alison Schofield 2023-10-27  1909  
a3e00c964fb9439 Alison Schofield 2023-10-27  1910  	return pos;
a3e00c964fb9439 Alison Schofield 2023-10-27  1911  }
a3e00c964fb9439 Alison Schofield 2023-10-27  1912  
Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
reviewer
#5 lore raw
In reply to [email protected]
Quoted 106 lines · On 2/25/2025 7:58 AM, [email protected] wrote:

From: Alison Schofield <[email protected]>

The CXL driver requires the granularity of a region and it's root decoder to be the same. This is particularly restrictive for 3-way host bridge interleaves where the only spec defined interleave configurations for creating 6-way and 12-way regions on a 3-way HB interleave require mixed granularities.

CXL 3.1 Specification 9.13.1.1:
Legal Interleaving Configurations: 12-way, 6-way, and 3-way

Adding support for these new interleaves touches these areas:

1) User created regions employing "cxl create-region" fail when the ndctl tool gets a mixed granularity request. That is addressed in a patch to the ndctl tool.

2) User created regions employing sysfs directly fail at the sysfs store of the non-matching region granularity. That restriction is lifted here. Note that the driver immediately allows any reasonable 3-way HB granularity, but the region config may still fail later if the ways/gran between region and root decoder are incompatible.

3) Auto created regions, in which the drivers role is to reflect the BIOS created region and provide RAS support, basically sneak on by. The driver ignores the root decoders granularity and assumes it is the same as the regions. The impact being that endpoint devices appear out of order and DPA to HPA address translations that depend on that order are invalid. Here the driver stops making that same granularity assumption and checks for an allowable granularity.

A new helper, interleave_granularity_allow(), is used in both the user and auto creation path to allow the newly supported configs.

Another new helper, gran_multiple(), is used in sorting, target list searching, and distance calculations, supporting the new root granularity to region granularity multiples.

And, as noted in 2), there's an added check to see if the selected sysfs ways/gran make sense when considered together.

---
 drivers/cxl/core/region.c | 101 +++++++++++++++++++++++++++++++-------
 1 file changed, 82 insertions(+), 19 deletions(-)
drivers/cxl/core/region.c +29 -9
index 3cb91cf0e2dd..9d4716b7fcf8 100644
@@ -532,6 +532,31 @@ static ssize_t interleave_granularity_show(struct device *dev,
 	return rc;
 }
 
+static bool interleave_granularity_allow(struct cxl_decoder *cxld, u16 ig)
+{
+	/*
+	 * When the host-bridge is interleaved, disallow region granularity
+	 * != root granularity with the exception of 3-way HB interleaves.
+	 * Allow the CXL Spec defined 3-way HB interleaves that can only be
+	 * configured when host-bridge interleave is greater that the
+	 * region interleave. (CXL 3.1 Specification 9.13.1.1)
+	 * Allow 2+2+2 interleave where HB gran is 2 * region granularity
+	 *	 4+4+4 interleave where HB gran is 4 * region granularity
+	 *
+	 * Regions with a granularity greater than the root interleave result
+	 * in invalid DPA translations (invalid to support).
+	 */
+	if (cxld->interleave_ways > 1 && ig != cxld->interleave_granularity) {
+		if (cxld->interleave_ways != 3)
+			return false;
+
+		if (cxld->interleave_granularity % (2 * ig) &&
+		    cxld->interleave_granularity % (4 * ig))
+			return false;
+	}
+	return true;
+}
+
 static ssize_t interleave_granularity_store(struct device *dev,
 					    struct device_attribute *attr,
 					    const char *buf, size_t len)
@@ -551,15 +576,7 @@ static ssize_t interleave_granularity_store(struct device *dev,
 	if (rc)
 		return rc;
 
-	/*
-	 * When the host-bridge is interleaved, disallow region granularity !=
-	 * root granularity. Regions with a granularity less than the root
-	 * interleave result in needing multiple endpoints to support a single
-	 * slot in the interleave (possible to support in the future). Regions
-	 * with a granularity greater than the root interleave result in invalid
-	 * DPA translations (invalid to support).
-	 */
-	if (cxld->interleave_ways > 1 && val != cxld->interleave_granularity)
+	if (!interleave_granularity_allow(cxld, val))
 		return -EINVAL;
 
 	rc = down_write_killable(&cxl_region_rwsem);
@@ -1290,6 +1307,23 @@ static int check_interleave_cap(struct cxl_decoder *cxld, int iw, int ig)
 	return 0;
 }
 
+static int gran_multiple(struct cxl_region *cxlr)
+{
+	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(cxlr->dev.parent);

drivers/cxl/core/region.c static int check_interleave_cap(struct cxl_decoder *cxld, int iw, int ig) · line 1322

Quoted 10 lines
drivers/cxl/core/region.c +10 -0
+	int root_decoder_gran = cxlrd->cxlsd.cxld.interleave_granularity;
+	int region_gran = cxlr->params.interleave_granularity;
+
+	/*
+	 * In regions built upon 3-way HB interleaves, the root decoder
+	 * granularity can be a multiple of the region granularity. The
+	 * multiple value is used in sorting and distance calculations.
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways != 3 || !root_decoder_gran)
+		return 1;

Hi Alison,

root decoder granularity is possible to be 0 here? I checked the code, the minimal value of it should be 256(CXL_DECODER_MIN_GRANULARITY), it was set in __cxl_parse_cfmws().

maybe this '!root_decoder_gran' checking is not needed?

Ming

[snip]

author maintainer
#6 lore raw
Quoted 2 lines · Li Ming
Quoted 1 line · On 2/25/2025 7:58 AM, [email protected] wrote:

From: Alison Schofield <[email protected]>

snip

Quoted 18 lines
Quoted 4 lines
drivers/cxl/core/region.c +3 -0
+static int gran_multiple(struct cxl_region *cxlr)
+{
+	struct cxl_root_decoder *cxlrd = to_cxl_root_decoder(cxlr->dev.parent);

drivers/cxl/core/region.c static int check_interleave_cap(struct cxl_decoder *cxld, int iw, int ig) · line 1322

Quoted 10 lines
drivers/cxl/core/region.c +10 -0
+	int root_decoder_gran = cxlrd->cxlsd.cxld.interleave_granularity;
+	int region_gran = cxlr->params.interleave_granularity;
+
+	/*
+	 * In regions built upon 3-way HB interleaves, the root decoder
+	 * granularity can be a multiple of the region granularity. The
+	 * multiple value is used in sorting and distance calculations.
+	 */
+	if (cxlrd->cxlsd.cxld.interleave_ways != 3 || !root_decoder_gran)
+		return 1;

Hi Alison,

Quoted 3 lines

root decoder granularity is possible to be 0 here? I checked the code, the minimal value of it should be 256(CXL_DECODER_MIN_GRANULARITY), it was set in __cxl_parse_cfmws().

maybe this '!root_decoder_gran' checking is not needed?

Thanks for looking that up. I imagine I had some divide by zero concern, which would have been wrong too.

Removed check in v2.

-- Alison

Quoted 7 lines

Ming

[snip]