cxl/pci: Get rid of pointer arithmetic reading CDAT table

Reading the CDAT table using DOE requires a Table Access Response
Header in addition to the CDAT entry. In current implementation this
has caused offsets with sizeof(__le32) to the actual buffers. This led
to hardly readable code and even bugs. E.g., see fix of devm_kfree()
in read_cdat_data():

 commit c65efe3685 ("cxl/cdat: Free correct buffer on checksum error")

Rework code to avoid calculations with sizeof(__le32). Introduce
struct cdat_doe_rsp for this which contains the Table Access Response
Header and a variable payload size for various data structures
afterwards to access the CDAT table and its CDAT Data Structures
without recalculating buffer offsets.

Cc: Lukas Wunner <lukas@wunner.de>
Cc: Fan Ni <nifan.cxl@gmail.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Robert Richter <rrichter@amd.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Link: https://lore.kernel.org/r/20240216155844.406996-3-rrichter@amd.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

drivers/cxl/core/pci.c

@@ -544,55 +544,57 @@ static int cxl_cdat_get_length(struct device *dev,
 
 static int cxl_cdat_read_table(struct device *dev,
 			       struct pci_doe_mb *doe_mb,
-			       void *cdat_table, size_t *cdat_length)
+			       struct cdat_doe_rsp *rsp, size_t *length)
 {
-	size_t length = *cdat_length + sizeof(__le32);
+	size_t received, remaining = *length;
-	__le32 *data = cdat_table;
+	unsigned int entry_handle = 0;
-	int entry_handle = 0;
+	union cdat_data *data;
 	__le32 saved_dw = 0;
 
 	do {
 		__le32 request = CDAT_DOE_REQ(entry_handle);
-		struct cdat_entry_header *entry;
-		size_t entry_dw;
 		int rc;
 
 		rc = pci_doe(doe_mb, PCI_DVSEC_VENDOR_ID_CXL,
 			     CXL_DOE_PROTOCOL_TABLE_ACCESS,
 			     &request, sizeof(request),
-			     data, length);
+			     rsp, sizeof(*rsp) + remaining);
 		if (rc < 0) {
 			dev_err(dev, "DOE failed: %d", rc);
 			return rc;
 		}
 
-		/* 1 DW Table Access Response Header + CDAT entry */
+		if (rc < sizeof(*rsp))
-		entry = (struct cdat_entry_header *)(data + 1);
-		if ((entry_handle == 0 &&
-		     rc != sizeof(__le32) + sizeof(struct cdat_header)) ||
-		    (entry_handle > 0 &&
-		     (rc < sizeof(__le32) + sizeof(*entry) ||
-		      rc != sizeof(__le32) + le16_to_cpu(entry->length))))
 			return -EIO;
 
+		data = (union cdat_data *)rsp->data;
+		received = rc - sizeof(*rsp);
+
+		if (entry_handle == 0) {
+			if (received != sizeof(data->header))
+				return -EIO;
+		} else {
+			if (received < sizeof(data->entry) ||
+			    received != le16_to_cpu(data->entry.length))
+				return -EIO;
+		}
+
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 le32_to_cpu(data[0]));
+					 le32_to_cpu(rsp->doe_header));
-		entry_dw = rc / sizeof(__le32);
+
-		/* Skip Header */
-		entry_dw -= 1;
 		/*
 		 * Table Access Response Header overwrote the last DW of
 		 * previous entry, so restore that DW
 		 */
-		*data = saved_dw;
+		rsp->doe_header = saved_dw;
-		length -= entry_dw * sizeof(__le32);
+		remaining -= received;
-		data += entry_dw;
+		rsp = (void *)rsp + received;
-		saved_dw = *data;
+		saved_dw = rsp->doe_header;
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
 	/* Length in CDAT header may exceed concatenation of CDAT entries */
-	*cdat_length -= length - sizeof(__le32);
+	*length -= remaining;
 
 	return 0;
 }
@@ -620,8 +622,8 @@ void read_cdat_data(struct cxl_port *port)
 	struct pci_doe_mb *doe_mb;
 	struct pci_dev *pdev = NULL;
 	struct cxl_memdev *cxlmd;
-	size_t cdat_length;
+	struct cdat_doe_rsp *buf;
-	void *cdat_table, *cdat_buf;
+	size_t length;
 	int rc;
 
 	if (is_cxl_memdev(uport)) {
@@ -647,30 +649,33 @@ void read_cdat_data(struct cxl_port *port)
 
 	port->cdat_available = true;
 
-	if (cxl_cdat_get_length(dev, doe_mb, &cdat_length)) {
+	if (cxl_cdat_get_length(dev, doe_mb, &length)) {
 		dev_dbg(dev, "No CDAT length\n");
 		return;
 	}
 
-	cdat_buf = devm_kzalloc(dev, cdat_length + sizeof(__le32), GFP_KERNEL);
+	/*
-	if (!cdat_buf)
+	 * The begin of the CDAT buffer needs space for additional 4
-		return;
+	 * bytes for the DOE header. Table data starts afterwards.
+	 */
+	buf = devm_kzalloc(dev, sizeof(*buf) + length, GFP_KERNEL);
+	if (!buf)
+		goto err;
 
-	rc = cxl_cdat_read_table(dev, doe_mb, cdat_buf, &cdat_length);
+	rc = cxl_cdat_read_table(dev, doe_mb, buf, &length);
 	if (rc)
 		goto err;
 
-	cdat_table = cdat_buf + sizeof(__le32);
+	if (cdat_checksum(buf->data, length))
-	if (cdat_checksum(cdat_table, cdat_length))
 		goto err;
 
-	port->cdat.table = cdat_table;
+	port->cdat.table = buf->data;
-	port->cdat.length = cdat_length;
+	port->cdat.length = length;
-	return;
 
+	return;
 err:
 	/* Don't leave table data allocated on error */
-	devm_kfree(dev, cdat_buf);
+	devm_kfree(dev, buf);
 	dev_err(dev, "Failed to read/validate CDAT.\n");
 }
 EXPORT_SYMBOL_NS_GPL(read_cdat_data, CXL);

drivers/cxl/cxlpci.h

@@ -71,6 +71,15 @@ enum cxl_regloc_type {
 	CXL_REGLOC_RBI_TYPES
 };
 
+/*
+ * Table Access DOE, CDAT Read Entry Response
+ *
+ * Spec refs:
+ *
+ * CXL 3.1 8.1.11, Table 8-14: Read Entry Response
+ * CDAT Specification 1.03: 2 CDAT Data Structures
+ */
+
 struct cdat_header {
 	__le32 length;
 	u8 revision;
@@ -85,6 +94,21 @@ struct cdat_entry_header {
 	__le16 length;
 } __packed;
 
+/*
+ * The DOE CDAT read response contains a CDAT read entry (either the
+ * CDAT header or a structure).
+ */
+union cdat_data {
+	struct cdat_header header;
+	struct cdat_entry_header entry;
+} __packed;
+
+/* There is an additional CDAT response header of 4 bytes. */
+struct cdat_doe_rsp {
+	__le32 doe_header;
+	u8 data[];
+} __packed;
+
 /*
  * CXL v3.0 6.2.3 Table 6-4
  * The table indicates that if PCIe Flit Mode is set, then CXL is in 256B flits