cxl/pci: Fix CDAT retrieval on big endian

The CDAT exposed in sysfs differs between little endian and big endian
arches:  On big endian, every 4 bytes are byte-swapped.

PCI Configuration Space is little endian (PCI r3.0 sec 6.1).  Accessors
such as pci_read_config_dword() implicitly swap bytes on big endian.
That way, the macros in include/uapi/linux/pci_regs.h work regardless of
the arch's endianness.  For an example of implicit byte-swapping, see
ppc4xx_pciex_read_config(), which calls in_le32(), which uses lwbrx
(Load Word Byte-Reverse Indexed).

DOE Read/Write Data Mailbox Registers are unlike other registers in
Configuration Space in that they contain or receive a 4 byte portion of
an opaque byte stream (a "Data Object" per PCIe r6.0 sec 7.9.24.5f).
They need to be copied to or from the request/response buffer verbatim.
So amend pci_doe_send_req() and pci_doe_recv_resp() to undo the implicit
byte-swapping.

The CXL_DOE_TABLE_ACCESS_* and PCI_DOE_DATA_OBJECT_DISC_* macros assume
implicit byte-swapping.  Byte-swap requests after constructing them with
those macros and byte-swap responses before parsing them.

Change the request and response type to __le32 to avoid sparse warnings.
Per a request from Jonathan, replace sizeof(u32) with sizeof(__le32) for
consistency.

Fixes: c97006046c79 ("cxl/port: Read CDAT table")
Tested-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Cc: stable@vger.kernel.org # v6.0+
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Link: https://lore.kernel.org/r/3051114102f41d19df3debbee123129118fc5e6d.1678543498.git.lukas@wunner.de
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

1 files changed, 6 insertions, 2 deletions

diff --git a/include/linux/pci-doe.h b/include/linux/pci-doe.h
index ed9b4df792b8..43765eaf2342 100644
--- a/include/linux/pci-doe.h
+++ b/include/linux/pci-doe.h
@@ -34,6 +34,10 @@ struct pci_doe_mb;
  * @work: Used internally by the mailbox
  * @doe_mb: Used internally by the mailbox
  *
+ * Payloads are treated as opaque byte streams which are transmitted verbatim,
+ * without byte-swapping.  If payloads contain little-endian register values,
+ * the caller is responsible for conversion with cpu_to_le32() / le32_to_cpu().
+ *
  * The payload sizes and rv are specified in bytes with the following
  * restrictions concerning the protocol.
  *
@@ -45,9 +49,9 @@ struct pci_doe_mb;
  */
 struct pci_doe_task {
 	struct pci_doe_protocol prot;
-	u32 *request_pl;
+	__le32 *request_pl;
 	size_t request_pl_sz;
-	u32 *response_pl;
+	__le32 *response_pl;
 	size_t response_pl_sz;
 	int rv;
 	void (*complete)(struct pci_doe_task *task);