cxl/test: mock fixups and new topology support

PATCH In review Alison Schofield 2 participants 5 messages Nov 5 – Nov 6 CXL

lore

Share & tools

lore

Alison Schofield

author maintainer

2026-11-05 03:57 UTC #0 lore raw

v3 cover cxl/test: mock fixups and new topology support

Changes in v3:
- Rebase onto 7.4-rc1

Changes in v2:
- Patch 2: Avoid duplicate look up of auto region window (RichardC)
- CoverLetter: Add link to the ndctl companion series

While working to enhance the cxl-cli cxl-create-region command to handle out-of-order target lists or no user-supplied target list, I ran into some basic issues in the existing command that need to be addressed first. Developing tests for those fixes exposed two bugs and a topology gap in the cxl_test mock module.

This cxl_test series addresses those issues:
1/3 fixes a DPA skip leak across mock decoder reset.
2/3 skips auto-region setup when no volatile window is active. 3/3 adds two x1 CFMWS windows needed to exercise root decoder selection.

A companion ndctl series[1] fixes the existing cxl-create-region issues in cxl-cli, clearing the way for out-of-order or automatically generated target lists.

These changes preserve the existing cxl_test unit test behavior while enabling new create-region test cases.

[1] https://lore.kernel.org/nvdimm/[email protected]/

Alison Schofield (3):
tools/testing/cxl: Preserve DPA skip across mock decoder reset tools/testing/cxl: Skip the auto region when no volatile window is active tools/testing/cxl: Add CFMWS windows for region selection tests

 tools/testing/cxl/test/cxl.c | 116 ++++++++++++++++++++++++++---------
 1 file changed, 87 insertions(+), 29 deletions(-)

base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4

Signature · 1 line

2.37.3

Alison Schofield

author maintainer

2026-11-05 03:57 UTC #1 lore raw

PATCH v3 1/3 tools/testing/cxl: Preserve DPA skip across mock decoder reset

Mock decoder reset clears the DPA skip before releasing the decoder's resources. This leaks skipped capacity and can cause a later DPA allocation to fail with ENOSPC.

Preserve the skip until the DPA resource is released.

Reviewed-by Richard Cheng <[email protected]>

Reviewed-by Dave Jiang <[email protected]>

Signed-off-by Alison Schofield <[email protected]>

---
 tools/testing/cxl/test/cxl.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

tools/testing/cxl/test/cxl.c +3 -6

index 3eb1051b1ee8..6819143640a9 100644
@@ -913,15 +913,13 @@ static void __cxld_registry_save(struct cxl_test_decoder *td,
 
 		cxld_copy(&td->cxled.cxld, cxld);
 		td->cxled.state = cxled->state;
-		td->cxled.skip = cxled->skip;
 
-		if (!(cxld->flags & CXL_DECODER_F_ENABLE)) {
-			td->dpa_range.start = 0;
-			td->dpa_range.end = -1;
-		} else if (cxled->dpa_res) {
+		if ((cxld->flags & CXL_DECODER_F_ENABLE) && cxled->dpa_res) {
+			td->cxled.skip = cxled->skip;
 			td->dpa_range.start = cxled->dpa_res->start;
 			td->dpa_range.end = cxled->dpa_res->end;
 		} else {
+			td->cxled.skip = 0;
 			td->dpa_range.start = 0;
 			td->dpa_range.end = -1;
 		}

@@ -1001,7 +999,6 @@ static void mock_decoder_reset(struct cxl_decoder *cxld)
 			to_cxl_endpoint_decoder(&cxld->dev);
 
 		cxled->state = CXL_DECODER_STATE_MANUAL;
-		cxled->skip = 0;
 	}
 	if (decoder_reset_preserve_registry)
 		dev_dbg(port->uport_dev, "decoder%d: skip registry update\n",

Signature · 1 line

2.37.3

Alison Schofield

author maintainer

2026-11-05 03:57 UTC #2 lore raw

PATCH v3 2/3 tools/testing/cxl: Skip the auto region when no volatile window is active

In reply to Alison Schofield

The mock auto region always uses CFMWS0. With XOR interleave arithmetic that window is inactive, so its HPA base is never assigned and the endpoint decoders are programmed at HPA 0.

No root decoder matches that range, leaving DPA allocated to endpoint decoders that are not part of a region.

Set up the auto region only when a volatile window is active.

Reviewed-by Dave Jiang <[email protected]>

Signed-off-by Alison Schofield <[email protected]>

---
 tools/testing/cxl/test/cxl.c | 36 ++++++++++++++++++++++++++++--------
 1 file changed, 28 insertions(+), 8 deletions(-)

tools/testing/cxl/test/cxl.c +28 -8

index 6819143640a9..d0793961b0ee 100644
@@ -424,6 +424,18 @@ struct acpi_cedt_cxims *mock_cxims[1] = {
 	[0] = &mock_cedt.cxims0.cxims,
 };
 
+static struct acpi_cedt_cfmws *mock_auto_region_cfmws(void)
+{
+	for (int i = cfmws_start; i <= cfmws_end; i++) {
+		struct acpi_cedt_cfmws *window = mock_cfmws[i];
+
+		if (window->restrictions & ACPI_CEDT_CFMWS_RESTRICT_VOLATILE)
+			return window;
+	}
+
+	return NULL;
+}
+
 struct cxl_mock_res {
 	struct list_head list;
 	struct range range;

@@ -1124,7 +1136,8 @@ enum cxld_init_type {
 static enum cxld_init_type get_decoder_init_type(struct cxl_decoder *cxld,
 						 struct platform_device *pdev,
 						 bool hb0,
-						 struct cxl_test_decoder **td)
+						 struct cxl_test_decoder **td,
+						 struct acpi_cedt_cfmws **window)
 {
 	struct cxl_test_decoder *found_td = cxld_registry_find(cxld);

@@ -1134,6 +1147,7 @@ static enum cxld_init_type get_decoder_init_type(struct cxl_decoder *cxld,
 	}
 
 	*td = NULL;
+	*window = NULL;
 
 	/*
 	 * The first decoder on the first 2 devices on the first switch

@@ -1147,6 +1161,11 @@ static enum cxld_init_type get_decoder_init_type(struct cxl_decoder *cxld,
 	    pdev->id > 4 || cxld->id > (mock_zero_size_decoders ? 2 : 0))
 		return MOCK_DECODER_INIT_DEFAULT;
 
+	/* XOR windows are all pmem, no window to host the RAM region */
+	*window = mock_auto_region_cfmws();
+	if (!*window)
+		return MOCK_DECODER_INIT_DEFAULT;
+
 	return type2_test ? MOCK_DECODER_INIT_TYPE2_AUTO :
 			    MOCK_DECODER_INIT_TYPE3_AUTO;
 }

@@ -1168,9 +1187,9 @@ static bool mock_decoder_handle_saved(struct cxl_decoder *cxld, struct cxl_test_
 }
 
 static void mock_init_hdm_type2_cxled(struct cxl_endpoint_decoder *cxled,
-				      struct cxl_port *port)
+				      struct cxl_port *port,
+				      struct acpi_cedt_cfmws *window)
 {
-	struct acpi_cedt_cfmws *window = mock_cfmws[0];
 	struct cxl_decoder *cxld = &cxled->cxld;
 	struct cxl_switch_decoder *cxlsd;
 	struct cxl_dport *dport;

@@ -1243,9 +1262,9 @@ static void mock_init_hdm_type2_cxled(struct cxl_endpoint_decoder *cxled,
 static void mock_init_hdm_type3_cxled(struct cxl_endpoint_decoder *cxled,
 				      struct cxl_port *port,
 				      struct platform_device *pdev,
-				      bool hb0)
+				      bool hb0,
+				      struct acpi_cedt_cfmws *window)
 {
-	struct acpi_cedt_cfmws *window = mock_cfmws[0];
 	struct cxl_decoder *cxld = &cxled->cxld;
 	struct cxl_switch_decoder *cxlsd;
 	struct cxl_dport *dport;

@@ -1396,6 +1415,7 @@ static bool mock_init_hdm_decoder(struct cxl_decoder *cxld)
 {
 	struct cxl_endpoint_decoder *cxled = NULL;
 	struct platform_device *pdev = NULL;
+	struct acpi_cedt_cfmws *window;
 	struct cxl_test_decoder *td;
 	struct cxl_memdev *cxlmd;
 	struct cxl_port *port;

@@ -1424,7 +1444,7 @@ static bool mock_init_hdm_decoder(struct cxl_decoder *cxld)
 		port = to_cxl_port(cxld->dev.parent);
 	}
 
-	switch (get_decoder_init_type(cxld, pdev, hb0, &td)) {
+	switch (get_decoder_init_type(cxld, pdev, hb0, &td, &window)) {
 	case MOCK_DECODER_INIT_SAVED:
 		if (WARN_ON(!td))
 			return false;

@@ -1437,10 +1457,10 @@ static bool mock_init_hdm_decoder(struct cxl_decoder *cxld)
 		default_mock_decoder(cxld);
 		return false;
 	case MOCK_DECODER_INIT_TYPE3_AUTO:
-		mock_init_hdm_type3_cxled(cxled, port, pdev, hb0);
+		mock_init_hdm_type3_cxled(cxled, port, pdev, hb0, window);
 		return false;
 	case MOCK_DECODER_INIT_TYPE2_AUTO:
-		mock_init_hdm_type2_cxled(cxled, port);
+		mock_init_hdm_type2_cxled(cxled, port, window);
 		return false;
 	default:
 		return false;

Signature · 1 line

2.37.3

Alison Schofield

author maintainer

2026-11-05 03:57 UTC #3 lore raw

PATCH v3 3/3 tools/testing/cxl: Add CFMWS windows for region selection tests

In reply to Alison Schofield

Region selection needs topologies where multiple root decoders can satisfy a request and where a decoder supports both RAM and PMEM. The mock topology provides neither.

Add two x1 windows, one PMEM only and one supporting both RAM and PMEM, while preserving the existing modulo decoder identities.

Reviewed-by Jonathan Cameron <[email protected]>

Reviewed-by Dave Jiang <[email protected]>

Signed-off-by Alison Schofield <[email protected]>

---
 tools/testing/cxl/test/cxl.c | 71 ++++++++++++++++++++++++++++--------
 1 file changed, 56 insertions(+), 15 deletions(-)

tools/testing/cxl/test/cxl.c +56 ~~-15~~

index d0793961b0ee..7fae489d00e9 100644
@@ -185,12 +185,20 @@ static struct {
 	} cfmws6;
 	struct {
 		struct acpi_cedt_cfmws cfmws;
-		u32 target[2];
+		u32 target[1];
 	} cfmws7;
 	struct {
 		struct acpi_cedt_cfmws cfmws;
-		u32 target[3];
+		u32 target[1];
 	} cfmws8;
+	struct {
+		struct acpi_cedt_cfmws cfmws;
+		u32 target[2];
+	} cfmws9;
+	struct {
+		struct acpi_cedt_cfmws cfmws;
+		u32 target[3];
+	} cfmws10;
 	struct {
 		struct acpi_cedt_cxims cxims;
 		u64 xormap_list[2];

@@ -325,22 +333,20 @@ static struct {
 		},
 		.target = { 3 },
 	},
-	/* .cfmws6,7,8 use ACPI_CEDT_CFMWS_ARITHMETIC_XOR */
 	.cfmws6 = {
 		.cfmws = {
 			.header = {
 				.type = ACPI_CEDT_TYPE_CFMWS,
 				.length = sizeof(mock_cedt.cfmws6),
 			},
-			.interleave_arithmetic = ACPI_CEDT_CFMWS_ARITHMETIC_XOR,
 			.interleave_ways = 0,
 			.granularity = 4,
 			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
 					ACPI_CEDT_CFMWS_RESTRICT_PMEM,
 			.qtg_id = FAKE_QTG_ID,
-			.window_size = SZ_256M * 8UL,
+			.window_size = SZ_256M * 4UL,
 		},
-		.target = { 0, },
+		.target = { 2 },
 	},
 	.cfmws7 = {
 		.cfmws = {

@@ -348,16 +354,17 @@ static struct {
 				.type = ACPI_CEDT_TYPE_CFMWS,
 				.length = sizeof(mock_cedt.cfmws7),
 			},
-			.interleave_arithmetic = ACPI_CEDT_CFMWS_ARITHMETIC_XOR,
-			.interleave_ways = 1,
-			.granularity = 0,
+			.interleave_ways = 0,
+			.granularity = 4,
 			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
+					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE |
 					ACPI_CEDT_CFMWS_RESTRICT_PMEM,
 			.qtg_id = FAKE_QTG_ID,
-			.window_size = SZ_256M * 8UL,
+			.window_size = SZ_256M * 4UL,
 		},
-		.target = { 0, 1, },
+		.target = { 1 },
 	},
+	/* .cfmws8,9,10 use ACPI_CEDT_CFMWS_ARITHMETIC_XOR */
 	.cfmws8 = {
 		.cfmws = {
 			.header = {

@@ -365,6 +372,38 @@ static struct {
 				.length = sizeof(mock_cedt.cfmws8),
 			},
 			.interleave_arithmetic = ACPI_CEDT_CFMWS_ARITHMETIC_XOR,
+			.interleave_ways = 0,
+			.granularity = 4,
+			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
+					ACPI_CEDT_CFMWS_RESTRICT_PMEM,
+			.qtg_id = FAKE_QTG_ID,
+			.window_size = SZ_256M * 8UL,
+		},
+		.target = { 0, },
+	},
+	.cfmws9 = {
+		.cfmws = {
+			.header = {
+				.type = ACPI_CEDT_TYPE_CFMWS,
+				.length = sizeof(mock_cedt.cfmws9),
+			},
+			.interleave_arithmetic = ACPI_CEDT_CFMWS_ARITHMETIC_XOR,
+			.interleave_ways = 1,
+			.granularity = 0,
+			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
+					ACPI_CEDT_CFMWS_RESTRICT_PMEM,
+			.qtg_id = FAKE_QTG_ID,
+			.window_size = SZ_256M * 8UL,
+		},
+		.target = { 0, 1, },
+	},
+	.cfmws10 = {
+		.cfmws = {
+			.header = {
+				.type = ACPI_CEDT_TYPE_CFMWS,
+				.length = sizeof(mock_cedt.cfmws10),
+			},
+			.interleave_arithmetic = ACPI_CEDT_CFMWS_ARITHMETIC_XOR,
 			.interleave_ways = 8,
 			.granularity = 1,
 			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |

@@ -407,18 +446,20 @@ struct acpi_cedt_cfmws *mock_cfmws[] = {
 	[3] = &mock_cedt.cfmws3.cfmws,
 	[4] = &mock_cedt.cfmws4.cfmws,
 	[5] = &mock_cedt.cfmws5.cfmws,
-	/* Modulo Math above, XOR Math below */
 	[6] = &mock_cedt.cfmws6.cfmws,
 	[7] = &mock_cedt.cfmws7.cfmws,
+	/* Modulo Math above, XOR Math below */
 	[8] = &mock_cedt.cfmws8.cfmws,
+	[9] = &mock_cedt.cfmws9.cfmws,
+	[10] = &mock_cedt.cfmws10.cfmws,
 };
 
 static int cfmws_start;
 static int cfmws_end;
 #define CFMWS_MOD_ARRAY_START 0
-#define CFMWS_MOD_ARRAY_END   5
-#define CFMWS_XOR_ARRAY_START 6
-#define CFMWS_XOR_ARRAY_END   8
+#define CFMWS_MOD_ARRAY_END   7
+#define CFMWS_XOR_ARRAY_START 8
+#define CFMWS_XOR_ARRAY_END   10
 
 struct acpi_cedt_cxims *mock_cxims[1] = {
 	[0] = &mock_cedt.cxims0.cxims,

Signature · 1 line

2.37.3

Dave Jiang

maintainer

2026-11-06 16:34 UTC #4 lore raw

In reply to Alison Schofield

Quoted 34 lines · Alison Schofield

Changes in v3:
- Rebase onto 7.4-rc1
Changes in v2:
- Patch 2: Avoid duplicate look up of auto region window (RichardC)
- CoverLetter: Add link to the ndctl companion series
While working to enhance the cxl-cli cxl-create-region command to handle out-of-order target lists or no user-supplied target list, I ran into some basic issues in the existing command that need to be addressed first. Developing tests for those fixes exposed two bugs and a topology gap in the cxl_test mock module.
This cxl_test series addresses those issues:
1/3 fixes a DPA skip leak across mock decoder reset.
2/3 skips auto-region setup when no volatile window is active. 3/3 adds two x1 CFMWS windows needed to exercise root decoder selection.
A companion ndctl series[1] fixes the existing cxl-create-region issues in cxl-cli, clearing the way for out-of-order or automatically generated target lists.
These changes preserve the existing cxl_test unit test behavior while enabling new create-region test cases.
[1] https://lore.kernel.org/nvdimm/[email protected]/
Alison Schofield (3):
tools/testing/cxl: Preserve DPA skip across mock decoder reset tools/testing/cxl: Skip the auto region when no volatile window is active tools/testing/cxl: Add CFMWS windows for region selection tests

Quoted 5 lines

 tools/testing/cxl/test/cxl.c | 116 ++++++++++++++++++++++++++---------
 1 file changed, 87 insertions(+), 29 deletions(-)

base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4

Applied to cxl/next
4d9f66751dd4 tools/testing/cxl: Add CFMWS windows for region selection tests a7c6a06c19d7 tools/testing/cxl: Skip the auto region when no volatile window is active aa6558a48986 tools/testing/cxl: Preserve DPA skip across mock decoder reset