Skip to contentSkip to conversation

Collection of DOE material

PATCH Superseded Lukas Wunner 7 participants 62 messages Feb 10, 2023 – Apr 3, 2023

Subsystems (3)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover Collection of DOE material

Collection of DOE material, v3:

Migrate to synchronous API, create mailboxes in PCI core instead of in drivers, relax restrictions on request/response size.

One major change since v2 is the behavior on device removal:
Previously ongoing DOE exchanges were canceled before driver unbinding, so that a driver doesn't wait for an ongoing DOE exchange to time out. That's the right thing to do for surprise removal, but not for orderly removal via sysfs or Attention Button, so canceling is now confined to the former.

Another major change is a new patch at the end of the series to rightsize CXL CDAT response allocation (per Jonathan's request).

A number of issues in CXL CDAT retrieval caught my eye. Some of them look like potential vulnerabilities.
I'm adding 4 patches at the beginning of the series to fix them.

Ira kindly tested the series again and verified that the CDAT exposed in sysfs remains identical.

Changes v2 -> v3:

* [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian [PATCH v3 02/16] cxl/pci: Handle truncated CDAT header [PATCH v3 03/16] cxl/pci: Handle truncated CDAT entries [PATCH v3 04/16] cxl/pci: Handle excessive CDAT length
* Newly added patch in v3

* [PATCH v3 05/16] PCI/DOE: Silence WARN splat with CONFIG_DEBUG_OBJECTS=y
* Explain in commit message what the long-term fix implemented by a subsequent commit is (Jonathan)

* [PATCH v3 10/16] PCI/DOE: Deduplicate mailbox flushing
* Newly added patch in v3

* [PATCH v3 11/16] PCI/DOE: Allow mailbox creation without devres management
* Call pci_doe_flush_mb() from pci_doe_destroy_mb() to simplify error paths (Jonathan)
* Explain in commit message why xa_destroy() is reordered in error path of the new pci_doe_create_mb() (Jonathan)

* [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration
* Don't cancel ongoing DOE exchanges in pci_stop_dev() so that

    drivers may perform DOE in their ->remove() hooks

* Instead cancel ongoing DOE exchanges on surprise removal in pci_dev_set_disconnected()
* Emit error message in pci_doe_init() if mailbox creation fails (Ira)
* Explain in commit message that pci_find_doe_mailbox() can later be amended with pci_find_next_doe_mailbox() (Jonathan)

* [PATCH v3 15/16] PCI/DOE: Relax restrictions on request and response size
* Fix byte order of last dword on big endian arches (Ira)
* Explain in commit message and kerneldoc that arbitrary-sized payloads are not stipulated by the spec, but merely for convenience of the caller (Bjorn, Jonathan)
* Add code comment that "remainder" in pci_doe_recv_resp() signifies the number of data bytes in the last payload dword (Ira)

* [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation
* Newly added patch in v3 on popular request (Jonathan)

Link to v2:

https://lore.kernel.org/all/[email protected]/

Lukas Wunner (16):
cxl/pci: Fix CDAT retrieval on big endian cxl/pci: Handle truncated CDAT header
cxl/pci: Handle truncated CDAT entries
cxl/pci: Handle excessive CDAT length
PCI/DOE: Silence WARN splat with CONFIG_DEBUG_OBJECTS=y PCI/DOE: Fix memory leak with CONFIG_DEBUG_OBJECTS=y PCI/DOE: Provide synchronous API and use it internally cxl/pci: Use synchronous API for DOE
PCI/DOE: Make asynchronous API private
PCI/DOE: Deduplicate mailbox flushing
PCI/DOE: Allow mailbox creation without devres management PCI/DOE: Create mailboxes on device enumeration cxl/pci: Use CDAT DOE mailbox created by PCI core PCI/DOE: Make mailbox creation API private PCI/DOE: Relax restrictions on request and response size cxl/pci: Rightsize CDAT response allocation

 .clang-format           |   1 -
 drivers/cxl/core/pci.c  | 125 ++++++---------
 drivers/cxl/cxl.h       |   3 +-
 drivers/cxl/cxlmem.h    |   3 -
 drivers/cxl/cxlpci.h    |  14 ++
 drivers/cxl/pci.c       |  49 ------
 drivers/cxl/port.c      |   2 +-
 drivers/pci/doe.c       | 340 ++++++++++++++++++++++++++++++----------
 drivers/pci/pci.h       |  12 ++
 drivers/pci/probe.c     |   1 +
 drivers/pci/remove.c    |   1 +
 include/linux/pci-doe.h |  62 +-------
 include/linux/pci.h     |   3 +
 13 files changed, 345 insertions(+), 271 deletions(-)
Signature · 1 line
2.39.1
#1 lore raw

PATCH v3 1/16 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.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c  | 12 ++++++------
 drivers/pci/doe.c       | 13 ++++++++-----
 include/linux/pci-doe.h |  8 ++++++--
 3 files changed, 20 insertions(+), 13 deletions(-)
drivers/cxl/core/pci.c +6 -6
index 57764e9cd19d..d3cf1d9d67d4 100644
@@ -480,7 +480,7 @@ static struct pci_doe_mb *find_cdat_doe(struct device *uport)
 	return NULL;
 }
 
-#define CDAT_DOE_REQ(entry_handle)					\
+#define CDAT_DOE_REQ(entry_handle) cpu_to_le32				\
 	(FIELD_PREP(CXL_DOE_TABLE_ACCESS_REQ_CODE,			\
 		    CXL_DOE_TABLE_ACCESS_REQ_CODE_READ) |		\
 	 FIELD_PREP(CXL_DOE_TABLE_ACCESS_TABLE_TYPE,			\
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
 }
 
 struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];
 	struct completion c;
 	struct pci_doe_task task;
 };
@@ -531,7 +531,7 @@ static int cxl_cdat_get_length(struct device *dev,
 	if (t.task.rv < sizeof(u32))
 		return -EIO;
 
-	*length = t.response_pl[1];
+	*length = le32_to_cpu(t.response_pl[1]);
 	dev_dbg(dev, "CDAT length %zu\n", *length);
 
 	return 0;
@@ -548,7 +548,7 @@ static int cxl_cdat_read_table(struct device *dev,
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
 		size_t entry_dw;
-		u32 *entry;
+		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);
@@ -563,7 +563,7 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 t.response_pl[0]);
+					 le32_to_cpu(t.response_pl[0]));
 		entry = t.response_pl + 1;
 		entry_dw = t.task.rv / sizeof(u32);
 		/* Skip Header */
drivers/pci/doe.c +8 -5
index 66d9ab288646..69efa9a250b9 100644
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 					  length));
 	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
 		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));
 
 	pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
 
@@ -198,8 +198,8 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 	payload_length = min(length, task->response_pl_sz / sizeof(u32));
 	/* Read the rest of the response payload */
 	for (i = 0; i < payload_length; i++) {
-		pci_read_config_dword(pdev, offset + PCI_DOE_READ,
-				      &task->response_pl[i]);
+		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
+		task->response_pl[i] = cpu_to_le32(val);
 		/* Prior to the last ack, ensure Data Object Ready */
 		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
 			return -EIO;
@@ -322,15 +322,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	struct pci_doe_task task = {
 		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
 		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = &request_pl,
+		.request_pl = (__le32 *)&request_pl,
 		.request_pl_sz = sizeof(request_pl),
-		.response_pl = &response_pl,
+		.response_pl = (__le32 *)&response_pl,
 		.response_pl_sz = sizeof(response_pl),
 		.complete = pci_doe_task_complete,
 		.private = &c,
 	};
 	int rc;
 
+	cpu_to_le32s(&request_pl);
+
 	rc = pci_doe_submit_task(doe_mb, &task);
 	if (rc < 0)
 		return rc;
@@ -340,6 +342,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	if (task.rv != sizeof(response_pl))
 		return -EIO;
 
+	le32_to_cpus(&response_pl);
 	*vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
 	*protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
 			      response_pl);
include/linux/pci-doe.h +6 -2
index ed9b4df792b8..43765eaf2342 100644
@@ -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);
Signature · 1 line
2.39.1
Reviewed-by on v3 1/16
#2 lore raw

RE: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 21 lines · Lukas Wunner

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.

Quoted 6 lines

Change the request and response type to __le32 to avoid sparse warnings.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

Good catch, a question below, but either way:

Quoted 65 lines

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c  | 12 ++++++------
 drivers/pci/doe.c       | 13 ++++++++-----
 include/linux/pci-doe.h |  8 ++++++--
 3 files changed, 20 insertions(+), 13 deletions(-)
drivers/cxl/core/pci.c +6 -6
index 57764e9cd19d..d3cf1d9d67d4 100644
@@ -480,7 +480,7 @@ static struct pci_doe_mb *find_cdat_doe(struct device *uport)
 	return NULL;
 }
 
-#define CDAT_DOE_REQ(entry_handle)					\
+#define CDAT_DOE_REQ(entry_handle) cpu_to_le32				\
 	(FIELD_PREP(CXL_DOE_TABLE_ACCESS_REQ_CODE,			\
 		    CXL_DOE_TABLE_ACCESS_REQ_CODE_READ) |		\
 	 FIELD_PREP(CXL_DOE_TABLE_ACCESS_TABLE_TYPE,			\
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
 }
 
 struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];
 	struct completion c;
 	struct pci_doe_task task;
 };
@@ -531,7 +531,7 @@ static int cxl_cdat_get_length(struct device *dev,
 	if (t.task.rv < sizeof(u32))
 		return -EIO;
 
-	*length = t.response_pl[1];
+	*length = le32_to_cpu(t.response_pl[1]);
 	dev_dbg(dev, "CDAT length %zu\n", *length);
 
 	return 0;
@@ -548,7 +548,7 @@ static int cxl_cdat_read_table(struct device *dev,
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
 		size_t entry_dw;
-		u32 *entry;
+		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);
@@ -563,7 +563,7 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 t.response_pl[0]);
+					 le32_to_cpu(t.response_pl[0]));
 		entry = t.response_pl + 1;
 		entry_dw = t.task.rv / sizeof(u32);
 		/* Skip Header */
drivers/pci/doe.c +0 -0
index 66d9ab288646..69efa9a250b9 100644

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 146

Quoted 5 lines
drivers/pci/doe.c +1 -1
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 					  length));
 	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
 		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));

What do you think about defining:

int pci_doe_write(const struct pci_dev *dev, int where, __le32 val); int pci_doe_read(const struct pci_dev *dev, int where, __le32 *val);

...local to this file to make it extra clear that DOE transfers are passing raw byte-streams and not register values as pci_{write,read}_config_dword() expect.

#3 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 14 lines · Dan Williams
Quoted 13 lines · Lukas Wunner

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.

[...]

Quoted 11 lines

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 146

Quoted 8 lines
drivers/pci/doe.c +1 -1
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 					  length));
 	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
 		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));

What do you think about defining:

Quoted 6 lines

int pci_doe_write(const struct pci_dev *dev, int where, __le32 val); int pci_doe_read(const struct pci_dev *dev, int where, __le32 *val);

...local to this file to make it extra clear that DOE transfers are passing raw byte-streams and not register values as pci_{write,read}_config_dword() expect.

I've done a prototype (see below), but it doesn't strike me as an improvement:

There are only two occurrences for each of those read and write accessors, so the diffstat isn't pretty (27 insertions, 12 deletions).

Moreover, the request header constructed in pci_doe_send_req() is constructed in native endianness and written using the standard pci_write_config_dword() accessor. Same for the response header parsed in pci_doe_recv_resp(). Thus, the functions do not consistently use the new custom accessors, but rather use a mix of the standard accessors and custom accessors. So clarity may not improve all that much.

Let me know if you feel otherwise!

The patch below goes on top of the series. I'm using a variation of your suggested approach in that I've named the custom accessors pci_doe_{write,read}_data() (for consistency with the existing pci_doe_write_ctrl()).

Thanks,

Lukas

-- >8 --

drivers/pci/doe.c +27 -12
index 8499cf9..6b0148e 100644
@@ -106,6 +106,24 @@ static void pci_doe_write_ctrl(struct pci_doe_mb *doe_mb, u32 val)
 	pci_write_config_dword(pdev, offset + PCI_DOE_CTRL, val);
 }
 
+static void pci_doe_write_data(struct pci_doe_mb *doe_mb, __le32 lav)
+{
+	struct pci_dev *pdev = doe_mb->pdev;
+	int offset = doe_mb->cap_offset;
+
+	pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, le32_to_cpu(lav));
+}
+
+static void pci_doe_read_data(struct pci_doe_mb *doe_mb, __le32 *lav)
+{
+	struct pci_dev *pdev = doe_mb->pdev;
+	int offset = doe_mb->cap_offset;
+	u32 val;
+
+	pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
+	*lav = cpu_to_le32(val);
+}
+
 static int pci_doe_abort(struct pci_doe_mb *doe_mb)
 {
 	struct pci_dev *pdev = doe_mb->pdev;
@@ -144,6 +162,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 	struct pci_dev *pdev = doe_mb->pdev;
 	int offset = doe_mb->cap_offset;
 	size_t length, remainder;
+	__le32 lav;
 	u32 val;
 	int i;
 
@@ -177,16 +196,14 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 
 	/* Write payload */
 	for (i = 0; i < task->request_pl_sz / sizeof(__le32); i++)
-		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       le32_to_cpu(task->request_pl[i]));
+		pci_doe_write_data(doe_mb, task->request_pl[i]);
 
 	/* Write last payload dword */
 	remainder = task->request_pl_sz % sizeof(__le32);
 	if (remainder) {
-		val = 0;
-		memcpy(&val, &task->request_pl[i], remainder);
-		le32_to_cpus(&val);
-		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, val);
+		lav = 0;
+		memcpy(&lav, &task->request_pl[i], remainder);
+		pci_doe_write_data(doe_mb, lav);
 	}
 
 	pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
@@ -211,6 +228,7 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 	size_t length, payload_length, remainder, received;
 	struct pci_dev *pdev = doe_mb->pdev;
 	int offset = doe_mb->cap_offset;
+	__le32 lav;
 	int i = 0;
 	u32 val;
 
@@ -256,16 +274,13 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 	if (payload_length) {
 		/* Read all payload dwords except the last */
 		for (; i < payload_length - 1; i++) {
-			pci_read_config_dword(pdev, offset + PCI_DOE_READ,
-					      &val);
-			task->response_pl[i] = cpu_to_le32(val);
+			pci_doe_read_data(doe_mb, &task->response_pl[i]);
 			pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
 		}
 
 		/* Read last payload dword */
-		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
-		cpu_to_le32s(&val);
-		memcpy(&task->response_pl[i], &val, remainder);
+		pci_doe_read_data(doe_mb, &lav);
+		memcpy(&task->response_pl[i], &lav, remainder);
 		/* Prior to the last ack, ensure Data Object Ready */
 		if (!pci_doe_data_obj_ready(doe_mb))
 			return -EIO;
#4 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

In reply to Lukas Wunner
Quoted 17 lines · On Fri, 10 Feb 2023 21:25:01 +0100 Lukas Wunner <[email protected]> wrote:

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.

Quoted 5 lines

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.

I'll start by saying I hate endian issues. Endless headaches.

The type swaps in here are confusing me but after arguing myself both ways this morning, I think you have it right.

In brief this patch set ensures that internal handling of the payloads is in le32 chunks by unwinding the implicit conversions in the pci config accessors for BE platforms. Thus the data in the payloads is always in the same order. Good start. Hence the binary read back of CDAT is little endian (including all fields within it).

Should we reflect that in the type of cdat->table or at least he u32 *data pointer in cxl_cdat_read_table()

A few other trivial simplifications inline.

Jonathan

Quoted 57 lines
Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c  | 12 ++++++------
 drivers/pci/doe.c       | 13 ++++++++-----
 include/linux/pci-doe.h |  8 ++++++--
 3 files changed, 20 insertions(+), 13 deletions(-)
drivers/cxl/core/pci.c +5 -5
index 57764e9cd19d..d3cf1d9d67d4 100644
@@ -480,7 +480,7 @@ static struct pci_doe_mb *find_cdat_doe(struct device *uport)
 	return NULL;
 }
 
-#define CDAT_DOE_REQ(entry_handle)					\
+#define CDAT_DOE_REQ(entry_handle) cpu_to_le32				\
 	(FIELD_PREP(CXL_DOE_TABLE_ACCESS_REQ_CODE,			\
 		    CXL_DOE_TABLE_ACCESS_REQ_CODE_READ) |		\
 	 FIELD_PREP(CXL_DOE_TABLE_ACCESS_TABLE_TYPE,			\
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
 }
 
 struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];
 	struct completion c;
 	struct pci_doe_task task;
 };
@@ -531,7 +531,7 @@ static int cxl_cdat_get_length(struct device *dev,
 	if (t.task.rv < sizeof(u32))
 		return -EIO;
 
-	*length = t.response_pl[1];
+	*length = le32_to_cpu(t.response_pl[1]);
 	dev_dbg(dev, "CDAT length %zu\n", *length);
 
 	return 0;
@@ -548,7 +548,7 @@ static int cxl_cdat_read_table(struct device *dev,
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
 		size_t entry_dw;
-		u32 *entry;
+		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 568

Quoted 7 lines
drivers/cxl/core/pci.c +1 -1
@@ -563,7 +563,7 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 t.response_pl[0]);
+					 le32_to_cpu(t.response_pl[0]));
 		entry = t.response_pl + 1;
 		entry_dw = t.task.rv / sizeof(u32);

Given we are manipulating it as __le32, I'd prefer to see that reflected in the sizeof() calls.

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 144

Quoted 8 lines
drivers/cxl/core/pci.c +0 -0
 		/* Skip Header */
drivers/pci/doe.c +0 -0
index 66d9ab288646..69efa9a250b9 100644
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 					  length));
 	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)

sizeof(__le32) or sizeof(task->request_pl[0])

Quoted 18 lines
drivers/pci/doe.c +3 -3
 		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));
 
 	pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
 
@@ -198,8 +198,8 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 	payload_length = min(length, task->response_pl_sz / sizeof(u32));
 	/* Read the rest of the response payload */
 	for (i = 0; i < payload_length; i++) {
-		pci_read_config_dword(pdev, offset + PCI_DOE_READ,
-				      &task->response_pl[i]);
+		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
+		task->response_pl[i] = cpu_to_le32(val);
 		/* Prior to the last ack, ensure Data Object Ready */
 		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
 			return -EIO;

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 325

Quoted 5 lines
drivers/pci/doe.c +1 -1
@@ -322,15 +322,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	struct pci_doe_task task = {
 		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
 		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = &request_pl,
+		.request_pl = (__le32 *)&request_pl,

I don't like this type cast. request_pl is local anyway, so why not change it's type and set it appropriately in the first place.

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 327

Quoted 3 lines
drivers/pci/doe.c +1 -1
 		.request_pl_sz = sizeof(request_pl),
-		.response_pl = &response_pl,
+		.response_pl = (__le32 *)&response_pl,

Likewise here.

Quoted 46 lines
drivers/pci/doe.c +3 -0
 		.response_pl_sz = sizeof(response_pl),
 		.complete = pci_doe_task_complete,
 		.private = &c,
 	};
 	int rc;
 
+	cpu_to_le32s(&request_pl);
+
 	rc = pci_doe_submit_task(doe_mb, &task);
 	if (rc < 0)
 		return rc;
@@ -340,6 +342,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	if (task.rv != sizeof(response_pl))
 		return -EIO;
 
+	le32_to_cpus(&response_pl);
 	*vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
 	*protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
 			      response_pl);
include/linux/pci-doe.h +6 -2
index ed9b4df792b8..43765eaf2342 100644
@@ -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);
#5 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 22 lines · Jonathan Cameron
Quoted 21 lines · On Fri, 10 Feb 2023 21:25:01 +0100 Lukas Wunner <[email protected]> wrote:

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.

Quoted 7 lines
Quoted 1 line

Change the request and response type to __le32 to avoid sparse warnings.

In brief this patch set ensures that internal handling of the payloads is in le32 chunks by unwinding the implicit conversions in the pci config accessors for BE platforms. Thus the data in the payloads is always in the same order. Good start. Hence the binary read back of CDAT is little endian (including all fields within it).

Correct. On big endian this means writing request dwords and reading response dwords involves 2 endianness conversions.

If the request is constructed with macros such as CXL_DOE_TABLE_ACCESS_* and PCI_DOE_DATA_OBJECT_DISC_*, then an additional (3rd) endianness conversion is necessary: The request is constructed in native big endian order, then converted to a little endian DOE payload. That payload is then converted back to big endian so that it can be written with the config space accessor, which converts to little endian.

I recognize this is crazy but I do not see a better solution. Dan suggested amending struct pci_ops with ->read_raw and ->write_raw callbacks which omit the endianness conversion. However there are 175 struct pci_ops in the kernel and each one would have to be looked at. I also do not know the Assembler instructions for every architecture to perform a non-byte-swapped write or read. It's a big task and lots of code to add. Given that DOE is likely the only user and big endian is seeing decreasing use, it's probably not worth the effort and LoC.

Quoted 10 lines

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 325

Quoted 6 lines
drivers/pci/doe.c +1 -1
@@ -322,15 +322,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	struct pci_doe_task task = {
 		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
 		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = &request_pl,
+		.request_pl = (__le32 *)&request_pl,

I don't like this type cast. request_pl is local anyway, so why not change it's type and set it appropriately in the first place.

That doesn't work I'm afraid. The request_pl is constructed with FIELD_PREP() here and the response_pl is parsed with FIELD_GET().

Those macros operate on an unsigned integer of native endianness. If I declare them __le32, I get sparse warnings and rightfully so.

The casts are the simplest, least intrusive solution. The only alternative would be to use separate variables but then I'd have to change a lot more lines and it would be more difficult to review and probably not all that more readable.

Thanks,

Lukas

Quoted 25 lines

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 327

Quoted 3 lines
drivers/pci/doe.c +1 -1
 		.request_pl_sz = sizeof(request_pl),
-		.response_pl = &response_pl,
+		.response_pl = (__le32 *)&response_pl,

Likewise here.

Quoted 19 lines
drivers/pci/doe.c +3 -0
 		.response_pl_sz = sizeof(response_pl),
 		.complete = pci_doe_task_complete,
 		.private = &c,
 	};
 	int rc;
 
+	cpu_to_le32s(&request_pl);
+
 	rc = pci_doe_submit_task(doe_mb, &task);
 	if (rc < 0)
 		return rc;
@@ -340,6 +342,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	if (task.rv != sizeof(response_pl))
 		return -EIO;
 
+	le32_to_cpus(&response_pl);
 	*vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
 	*protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
 			      response_pl);
#6 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 65 lines · On Tue, 14 Feb 2023 14:51:10 +0100 Lukas Wunner <[email protected]> wrote:
Quoted 22 lines · Jonathan Cameron
Quoted 21 lines · On Fri, 10 Feb 2023 21:25:01 +0100 Lukas Wunner <[email protected]> wrote:

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.

Quoted 7 lines
Quoted 1 line

Change the request and response type to __le32 to avoid sparse warnings.

In brief this patch set ensures that internal handling of the payloads is in le32 chunks by unwinding the implicit conversions in the pci config accessors for BE platforms. Thus the data in the payloads is always in the same order. Good start. Hence the binary read back of CDAT is little endian (including all fields within it).

Correct. On big endian this means writing request dwords and reading response dwords involves 2 endianness conversions.

If the request is constructed with macros such as CXL_DOE_TABLE_ACCESS_* and PCI_DOE_DATA_OBJECT_DISC_*, then an additional (3rd) endianness conversion is necessary: The request is constructed in native big endian order, then converted to a little endian DOE payload. That payload is then converted back to big endian so that it can be written with the config space accessor, which converts to little endian.

I recognize this is crazy but I do not see a better solution. Dan suggested amending struct pci_ops with ->read_raw and ->write_raw callbacks which omit the endianness conversion. However there are 175 struct pci_ops in the kernel and each one would have to be looked at. I also do not know the Assembler instructions for every architecture to perform a non-byte-swapped write or read. It's a big task and lots of code to add. Given that DOE is likely the only user and big endian is seeing decreasing use, it's probably not worth the effort and LoC.

Quoted 10 lines

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 325

Quoted 6 lines
drivers/pci/doe.c +1 -1
@@ -322,15 +322,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	struct pci_doe_task task = {
 		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
 		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = &request_pl,
+		.request_pl = (__le32 *)&request_pl,  

I don't like this type cast. request_pl is local anyway, so why not change it's type and set it appropriately in the first place.

That doesn't work I'm afraid. The request_pl is constructed with FIELD_PREP() here and the response_pl is parsed with FIELD_GET().

Quoted 7 lines

Those macros operate on an unsigned integer of native endianness. If I declare them __le32, I get sparse warnings and rightfully so.

The casts are the simplest, least intrusive solution. The only alternative would be to use separate variables but then I'd have to change a lot more lines and it would be more difficult to review and probably not all that more readable.

Separate variables was what I meant (less than clear)

Add a __le32 le32_req_pl then
do the endian conversion into that rather than in place conversion. Equivalent in teh other direction for the response.

That way the endian markings end up correct on all the local variables and it shouldn't require many more lines to change.

Quoted 31 lines

Thanks,

Lukas

Quoted 25 lines

drivers/pci/doe.c static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid, · line 327

Quoted 3 lines
drivers/pci/doe.c +1 -1
 		.request_pl_sz = sizeof(request_pl),
-		.response_pl = &response_pl,
+		.response_pl = (__le32 *)&response_pl,  

Likewise here.

Quoted 19 lines
drivers/pci/doe.c +3 -0
 		.response_pl_sz = sizeof(response_pl),
 		.complete = pci_doe_task_complete,
 		.private = &c,
 	};
 	int rc;
 
+	cpu_to_le32s(&request_pl);
+
 	rc = pci_doe_submit_task(doe_mb, &task);
 	if (rc < 0)
 		return rc;
@@ -340,6 +342,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	if (task.rv != sizeof(response_pl))
 		return -EIO;
 
+	le32_to_cpus(&response_pl);
 	*vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
 	*protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
 			      response_pl);  
#7 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

In reply to Lukas Wunner
Quoted 2 lines · Lukas Wunner

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

hexdump prints different byte streams on LE and BE? Does not seem right.

Quoted 47 lines

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.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

  drivers/cxl/core/pci.c  | 12 ++++++------
  drivers/pci/doe.c       | 13 ++++++++-----
  include/linux/pci-doe.h |  8 ++++++--
  3 files changed, 20 insertions(+), 13 deletions(-)
drivers/cxl/core/pci.c +1 -1
index 57764e9cd19d..d3cf1d9d67d4 100644
@@ -480,7 +480,7 @@ static struct pci_doe_mb *find_cdat_doe(struct device *uport)
  	return NULL;
  }
 
-#define CDAT_DOE_REQ(entry_handle)					\
+#define CDAT_DOE_REQ(entry_handle) cpu_to_le32				\
  	(FIELD_PREP(CXL_DOE_TABLE_ACCESS_REQ_CODE,			\
  		    CXL_DOE_TABLE_ACCESS_REQ_CODE_READ) |		\
  	 FIELD_PREP(CXL_DOE_TABLE_ACCESS_TABLE_TYPE,			\

drivers/cxl/core/pci.c static void cxl_doe_task_complete(struct pci_doe_task *task) · line 497

Quoted 7 lines
drivers/cxl/core/pci.c +2 -2
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
  }
 
  struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];

This is ok as it is a binary format of DOE message (is it?)...

Quoted 35 lines
drivers/cxl/core/pci.c +3 -3
  	struct completion c;
  	struct pci_doe_task task;
  };
@@ -531,7 +531,7 @@ static int cxl_cdat_get_length(struct device *dev,
  	if (t.task.rv < sizeof(u32))
  		return -EIO;
 
-	*length = t.response_pl[1];
+	*length = le32_to_cpu(t.response_pl[1]);
  	dev_dbg(dev, "CDAT length %zu\n", *length);
 
  	return 0;
@@ -548,7 +548,7 @@ static int cxl_cdat_read_table(struct device *dev,
  	do {
  		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
  		size_t entry_dw;
-		u32 *entry;
+		__le32 *entry;
  		int rc;
 
  		rc = pci_doe_submit_task(cdat_doe, &t.task);
@@ -563,7 +563,7 @@ static int cxl_cdat_read_table(struct device *dev,
 
  		/* Get the CXL table access header entry handle */
  		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 t.response_pl[0]);
+					 le32_to_cpu(t.response_pl[0]));
  		entry = t.response_pl + 1;
  		entry_dw = t.task.rv / sizeof(u32);
  		/* Skip Header */
drivers/pci/doe.c +0 -0
index 66d9ab288646..69efa9a250b9 100644

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 146

Quoted 5 lines
drivers/pci/doe.c +1 -1
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
  					  length));
  	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
  		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));

Does it really work on BE? My little brain explodes on all these convertions :)

char buf[] = { 1, 2, 3, 4 }
u32 *request_pl = buf;

request_pl[0] will be 0x01020304. le32_to_cpu(request_pl[0]) will be 0x04030201 And then pci_write_config_dword() will do another swap.

Did I miss something? (/me is gone bringing up a BE system).

Quoted 58 lines
drivers/pci/doe.c +7 -4
  	pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
 
@@ -198,8 +198,8 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
  	payload_length = min(length, task->response_pl_sz / sizeof(u32));
  	/* Read the rest of the response payload */
  	for (i = 0; i < payload_length; i++) {
-		pci_read_config_dword(pdev, offset + PCI_DOE_READ,
-				      &task->response_pl[i]);
+		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
+		task->response_pl[i] = cpu_to_le32(val);
  		/* Prior to the last ack, ensure Data Object Ready */
  		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
  			return -EIO;
@@ -322,15 +322,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
  	struct pci_doe_task task = {
  		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
  		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = &request_pl,
+		.request_pl = (__le32 *)&request_pl,
  		.request_pl_sz = sizeof(request_pl),
-		.response_pl = &response_pl,
+		.response_pl = (__le32 *)&response_pl,
  		.response_pl_sz = sizeof(response_pl),
  		.complete = pci_doe_task_complete,
  		.private = &c,
  	};
  	int rc;
 
+	cpu_to_le32s(&request_pl);
+
  	rc = pci_doe_submit_task(doe_mb, &task);
  	if (rc < 0)
  		return rc;
@@ -340,6 +342,7 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
  	if (task.rv != sizeof(response_pl))
  		return -EIO;
 
+	le32_to_cpus(&response_pl);
  	*vid = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_VID, response_pl);
  	*protocol = FIELD_GET(PCI_DOE_DATA_OBJECT_DISC_RSP_3_PROTOCOL,
  			      response_pl);
include/linux/pci-doe.h +4 -0
index ed9b4df792b8..43765eaf2342 100644
@@ -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.
   *

include/linux/pci-doe.h struct pci_doe_mb; · line 54

Quoted 8 lines
include/linux/pci-doe.h +2 -2
@@ -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;

This does not look right. Either:
- pci_doe() should also take __le32* or
- pci_doe() should do cpu_to_le32() for the request, and pci_doe_task_complete() - for the response.

Thanks,

Quoted 3 lines
include/linux/pci-doe.h +0 -0
  	size_t response_pl_sz;
  	int rv;
  	void (*complete)(struct pci_doe_task *task);
Signature · 1 line
Alexey
#8 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 10 lines · Alexey Kardashevskiy

drivers/cxl/core/pci.c static void cxl_doe_task_complete(struct pci_doe_task *task) · line 496

Quoted 7 lines · Lukas Wunner
drivers/cxl/core/pci.c +2 -2
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
  }
  struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];

This is ok as it is a binary format of DOE message (is it?)...

Yes, the DOE request and response is an opaque byte stream.

The above-quoted type changes are necessary to avoid sparse warnings (as noted in the commit message).

Quoted 11 lines

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 146

Quoted 8 lines
drivers/pci/doe.c +1 -1
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
  					  length));
  	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
  		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));

Does it really work on BE? My little brain explodes on all these convertions

Quoted 8 lines

char buf[] = { 1, 2, 3, 4 }
u32 *request_pl = buf;

request_pl[0] will be 0x01020304. le32_to_cpu(request_pl[0]) will be 0x04030201 And then pci_write_config_dword() will do another swap.

Did I miss something? (/me is gone bringing up a BE system).

Correct.

Quoted 7 lines
Quoted 7 lines
drivers/pci/doe.c +1 -1
@@ -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;
Quoted 8 lines

drivers/pci/doe.c struct pci_doe_mb; · line 54

Quoted 2 lines
drivers/pci/doe.c +1 -1
-	u32 *response_pl;
+	__le32 *response_pl;

This does not look right. Either:
- pci_doe() should also take __le32* or
- pci_doe() should do cpu_to_le32() for the request, and pci_doe_task_complete() - for the response.

Again, the type changes are necessary to avoid sparse warnings.

pci_doe() takes a void * because patch [15/16] eliminates the need for the request and response size to be a multiple of dwords. Passing __le32 * to pci_doe() would then no longer be correct.

Thanks,

Lukas

#9 lore raw

Re: [PATCH v3 01/16] cxl/pci: Fix CDAT retrieval on big endian

Quoted 32 lines · Lukas Wunner
Quoted 10 lines · Alexey Kardashevskiy

drivers/cxl/core/pci.c static void cxl_doe_task_complete(struct pci_doe_task *task) · line 496

Quoted 7 lines · Lukas Wunner
drivers/cxl/core/pci.c +2 -2
@@ -493,8 +493,8 @@ static void cxl_doe_task_complete(struct pci_doe_task *task)
   }
   struct cdat_doe_task {
-	u32 request_pl;
-	u32 response_pl[32];
+	__le32 request_pl;
+	__le32 response_pl[32];

This is ok as it is a binary format of DOE message (is it?)...

Yes, the DOE request and response is an opaque byte stream.

The above-quoted type changes are necessary to avoid sparse warnings (as noted in the commit message).

Quoted 14 lines

drivers/pci/doe.c static int pci_doe_send_req(struct pci_doe_mb *doe_mb, · line 146

Quoted 8 lines
drivers/pci/doe.c +1 -1
@@ -143,7 +143,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
   					  length));
   	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
   		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
-				       task->request_pl[i]);
+				       le32_to_cpu(task->request_pl[i]));

Does it really work on BE? My little brain explodes on all these convertions

char buf[] = { 1, 2, 3, 4 }
u32 *request_pl = buf;

Quoted 7 lines
Quoted 5 lines

request_pl[0] will be 0x01020304. le32_to_cpu(request_pl[0]) will be 0x04030201 And then pci_write_config_dword() will do another swap.

Did I miss something? (/me is gone bringing up a BE system).

Correct.

Uff, ok, your code works correct, sorry for the noise.

Quoted 17 lines
Quoted 16 lines

drivers/pci/doe.c struct pci_doe_mb; · line 54

Quoted 10 lines
drivers/pci/doe.c +2 -2
@@ -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;

This does not look right. Either:
- pci_doe() should also take __le32* or
- pci_doe() should do cpu_to_le32() for the request, and pci_doe_task_complete() - for the response.

Quoted 1 line

Again, the type changes are necessary to avoid sparse warnings.

I'd shut it up like this:

===
for (i = 0; i < task->request_pl_sz / sizeof(__le32); i++)
pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
- le32_to_cpu(task->request_pl[i]));
+ le32_to_cpu(*(__le32 *)&task->request_pl[i]));
===

+ may be a comment that LE is a natural order. Dunno.

Changing types to __le32 without explicit conversions or comments just confuses. Thanks,

Quoted 3 lines

pci_doe() takes a void * because patch [15/16] eliminates the need for the request and response size to be a multiple of dwords. Passing __le32 * to pci_doe() would then no longer be correct.

Signature · 1 line
Alexey
#10 lore raw

PATCH v3 2/16 cxl/pci: Handle truncated CDAT header

In reply to Lukas Wunner

cxl_cdat_get_length() only checks whether the DOE response size is sufficient for the Table Access response header (1 dword), but not the succeeding CDAT header (1 dword length plus other fields).

It thus returns whatever uninitialized memory happens to be on the stack if a truncated DOE response with only 1 dword was received. Fix it.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cxl/core/pci.c +1 -1
index d3cf1d9d67d4..11a85b3a9a0b 100644
@@ -528,7 +528,7 @@ static int cxl_cdat_get_length(struct device *dev,
 		return rc;
 	}
 	wait_for_completion(&t.c);
-	if (t.task.rv < sizeof(u32))
+	if (t.task.rv < 2 * sizeof(u32))
 		return -EIO;
 
 	*length = le32_to_cpu(t.response_pl[1]);
Signature · 1 line
2.39.1
Reviewed-by on v3 2/16
#11 lore raw

RE: [PATCH v3 02/16] cxl/pci: Handle truncated CDAT header

Quoted 24 lines · Lukas Wunner

cxl_cdat_get_length() only checks whether the DOE response size is sufficient for the Table Access response header (1 dword), but not the succeeding CDAT header (1 dword length plus other fields).

It thus returns whatever uninitialized memory happens to be on the stack if a truncated DOE response with only 1 dword was received. Fix it.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cxl/core/pci.c +0 -0
index d3cf1d9d67d4..11a85b3a9a0b 100644

drivers/cxl/core/pci.c static int cxl_cdat_get_length(struct device *dev, · line 532

Quoted 6 lines
drivers/cxl/core/pci.c +1 -1
@@ -528,7 +528,7 @@ static int cxl_cdat_get_length(struct device *dev,
 		return rc;
 	}
 	wait_for_completion(&t.c);
-	if (t.task.rv < sizeof(u32))
+	if (t.task.rv < 2 * sizeof(u32))
 		return -EIO;

Looks good, I wonder since this is standard for all data objects whether the check should be pushed into the core?

For now this is easier to backport, but a follow-on could push it down a level.

#12 lore raw

Re: [PATCH v3 02/16] cxl/pci: Handle truncated CDAT header

Quoted 7 lines · Dan Williams
Quoted 6 lines · Lukas Wunner

cxl_cdat_get_length() only checks whether the DOE response size is sufficient for the Table Access response header (1 dword), but not the succeeding CDAT header (1 dword length plus other fields).

It thus returns whatever uninitialized memory happens to be on the stack if a truncated DOE response with only 1 dword was received. Fix it.

[...]

Quoted 4 lines
Quoted 4 lines
drivers/cxl/core/pci.c +0 -0
@@ -528,7 +528,7 @@ static int cxl_cdat_get_length(struct device *dev,
 		return rc;
Quoted 8 lines

drivers/cxl/core/pci.c static int cxl_cdat_get_length(struct device *dev, · line 532

Quoted 5 lines
drivers/cxl/core/pci.c +1 -1
 	}
 	wait_for_completion(&t.c);
-	if (t.task.rv < sizeof(u32))
+	if (t.task.rv < 2 * sizeof(u32))
 		return -EIO;

Looks good, I wonder since this is standard for all data objects whether the check should be pushed into the core?

No, "t.task.rv" contains the payload length in bytes of the response received via DOE. It doesn't include the DOE header.

I think it is legal to receive an empty response via DOE, so I cannot push a length check down into the core.

In this case, the payload contains one dword for the Table Access Response header (CXL r3.0 sec 8.1.11.1), followed by 3 dwords for the CDAT header:

https://uefi.org/sites/default/files/resources/Coherent%20Device%20Attribute%20Table_1.01.pdf

The CDAT header contains the length of the entire CDAT in the first dword, hence the above-quoted check only verifies that at least two dwords were received. It's harmless if the remainder of the CDAT header is truncated.

Thanks,

Lukas

Reviewed-by on v3 2/16
#13 lore raw

Re: [PATCH v3 02/16] cxl/pci: Handle truncated CDAT header

In reply to Lukas Wunner
Quoted 4 lines · On Fri, 10 Feb 2023 21:25:02 +0100 Lukas Wunner <[email protected]> wrote:

cxl_cdat_get_length() only checks whether the DOE response size is sufficient for the Table Access response header (1 dword), but not the succeeding CDAT header (1 dword length plus other fields).

Quoted 8 lines

It thus returns whatever uninitialized memory happens to be on the stack if a truncated DOE response with only 1 dword was received. Fix it.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

Oops + thanks for fixing.

Quoted 20 lines

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cxl/core/pci.c +1 -1
index d3cf1d9d67d4..11a85b3a9a0b 100644
@@ -528,7 +528,7 @@ static int cxl_cdat_get_length(struct device *dev,
 		return rc;
 	}
 	wait_for_completion(&t.c);
-	if (t.task.rv < sizeof(u32))
+	if (t.task.rv < 2 * sizeof(u32))
 		return -EIO;
 
 	*length = le32_to_cpu(t.response_pl[1]);
Reviewed-by on v3 2/16
#14 lore raw

Re: [PATCH v3 02/16] cxl/pci: Handle truncated CDAT header

In reply to Lukas Wunner
Quoted 11 lines · Lukas Wunner

cxl_cdat_get_length() only checks whether the DOE response size is sufficient for the Table Access response header (1 dword), but not the succeeding CDAT header (1 dword length plus other fields).

It thus returns whatever uninitialized memory happens to be on the stack if a truncated DOE response with only 1 dword was received. Fix it.

Fixes c97006046c79 ("cxl/port: Read CDAT table")
#15 lore raw

PATCH v3 3/16 cxl/pci: Handle truncated CDAT entries

In reply to Lukas Wunner

If truncated CDAT entries are received from a device, the concatenation of those entries constitutes a corrupt CDAT, yet is happily exposed to user space.

Avoid by verifying response lengths and erroring out if truncation is detected.

The last CDAT entry may still be truncated despite the checks introduced herein if the length in the CDAT header is too small. However, that is easily detectable by user space because it reaches EOF prematurely. A subsequent commit which rightsizes the CDAT response allocation closes that remaining loophole.

The two lines introduced here which exceed 80 chars are shortened to less than 80 chars by a subsequent commit which migrates to a synchronous DOE API and replaces "t.task.rv" by "rc".

The existing acpi_cdat_header and acpi_table_cdat struct definitions provided by ACPICA cannot be used because they do not employ __le16 or __le32 types. I believe that cannot be changed because those types are Linux-specific and ACPI is specified for little endian platforms only, hence doesn't care about endianness. So duplicate the structs.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 13 +++++++++----
 drivers/cxl/cxlpci.h   | 14 ++++++++++++++
 2 files changed, 23 insertions(+), 4 deletions(-)
drivers/cxl/core/pci.c +9 -4
index 11a85b3a9a0b..a3fb6bd68d17 100644
@@ -547,8 +547,8 @@ static int cxl_cdat_read_table(struct device *dev,
 
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
+		struct cdat_entry_header *entry;
 		size_t entry_dw;
-		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);
@@ -557,14 +557,19 @@ static int cxl_cdat_read_table(struct device *dev,
 			return rc;
 		}
 		wait_for_completion(&t.c);
-		/* 1 DW header + 1 DW data min */
-		if (t.task.rv < (2 * sizeof(u32)))
+
+		/* 1 DW Table Access Response Header + CDAT entry */
+		entry = (struct cdat_entry_header *)(t.response_pl + 1);
+		if ((entry_handle == 0 &&
+		     t.task.rv != sizeof(u32) + sizeof(struct cdat_header)) ||
+		    (entry_handle > 0 &&
+		     (t.task.rv < sizeof(u32) + sizeof(struct cdat_entry_header) ||
+		      t.task.rv != sizeof(u32) + le16_to_cpu(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(t.response_pl[0]));
-		entry = t.response_pl + 1;
 		entry_dw = t.task.rv / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
drivers/cxl/cxlpci.h +14 -0
index 920909791bb9..104ad2b72516 100644
@@ -62,6 +62,20 @@ enum cxl_regloc_type {
 	CXL_REGLOC_RBI_TYPES
 };
 
+struct cdat_header {
+	__le32 length;
+	u8 revision;
+	u8 checksum;
+	u8 reserved[6];
+	__le32 sequence;
+} __packed;
+
+struct cdat_entry_header {
+	u8 type;
+	u8 reserved;
+	__le16 length;
+} __packed;
+
 int devm_cxl_port_enumerate_dports(struct cxl_port *port);
 struct cxl_dev_state;
 int cxl_hdm_decode_init(struct cxl_dev_state *cxlds, struct cxl_hdm *cxlhdm);
Signature · 1 line
2.39.1
Reviewed-by on v3 3/16
#16 lore raw

RE: [PATCH v3 03/16] cxl/pci: Handle truncated CDAT entries

Quoted 50 lines · Lukas Wunner

If truncated CDAT entries are received from a device, the concatenation of those entries constitutes a corrupt CDAT, yet is happily exposed to user space.

Avoid by verifying response lengths and erroring out if truncation is detected.

The last CDAT entry may still be truncated despite the checks introduced herein if the length in the CDAT header is too small. However, that is easily detectable by user space because it reaches EOF prematurely. A subsequent commit which rightsizes the CDAT response allocation closes that remaining loophole.

The two lines introduced here which exceed 80 chars are shortened to less than 80 chars by a subsequent commit which migrates to a synchronous DOE API and replaces "t.task.rv" by "rc".

The existing acpi_cdat_header and acpi_table_cdat struct definitions provided by ACPICA cannot be used because they do not employ __le16 or __le32 types. I believe that cannot be changed because those types are Linux-specific and ACPI is specified for little endian platforms only, hence doesn't care about endianness. So duplicate the structs.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 13 +++++++++----
 drivers/cxl/cxlpci.h   | 14 ++++++++++++++
 2 files changed, 23 insertions(+), 4 deletions(-)
drivers/cxl/core/pci.c +1 -1
index 11a85b3a9a0b..a3fb6bd68d17 100644
@@ -547,8 +547,8 @@ static int cxl_cdat_read_table(struct device *dev,
 
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
+		struct cdat_entry_header *entry;
 		size_t entry_dw;
-		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 561

Quoted 5 lines
drivers/cxl/core/pci.c +0 -2
@@ -557,14 +557,19 @@ static int cxl_cdat_read_table(struct device *dev,
 			return rc;
 		}
 		wait_for_completion(&t.c);
-		/* 1 DW header + 1 DW data min */
-		if (t.task.rv < (2 * sizeof(u32)))

Ah, I guess that's why the previous check can not be pushed down further, its obviated by this more comprehensive check.

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 568

Quoted 9 lines
drivers/cxl/core/pci.c +8 -0
+
+		/* 1 DW Table Access Response Header + CDAT entry */
+		entry = (struct cdat_entry_header *)(t.response_pl + 1);
+		if ((entry_handle == 0 &&
+		     t.task.rv != sizeof(u32) + sizeof(struct cdat_header)) ||
+		    (entry_handle > 0 &&
+		     (t.task.rv < sizeof(u32) + sizeof(struct cdat_entry_header) ||
+		      t.task.rv != sizeof(u32) + le16_to_cpu(entry->length))))
 			return -EIO;

Looks correct for catching that the response is large enough to communicate the length and that the expected length was retrieved.

#17 lore raw

Re: [PATCH v3 03/16] cxl/pci: Handle truncated CDAT entries

Quoted 7 lines · Dan Williams
Quoted 6 lines · Lukas Wunner

If truncated CDAT entries are received from a device, the concatenation of those entries constitutes a corrupt CDAT, yet is happily exposed to user space.

Avoid by verifying response lengths and erroring out if truncation is detected.

[...]

Quoted 9 lines

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 561

Quoted 6 lines
drivers/cxl/core/pci.c +0 -2
@@ -557,14 +557,19 @@ static int cxl_cdat_read_table(struct device *dev,
 			return rc;
 		}
 		wait_for_completion(&t.c);
-		/* 1 DW header + 1 DW data min */
-		if (t.task.rv < (2 * sizeof(u32)))

Ah, I guess that's why the previous check can not be pushed down further, its obviated by this more comprehensive check.

The check in the previous patch ("cxl/pci: Handle truncated CDAT header") fixes response size verification for cxl_cdat_get_length().

The check here however is in a different function, cxl_cdat_read_table(). Previously the function only checked whether the DOE response contained two dwords. That's one dword for the Table Access Response header (CXL r3.0 sec 8.1.11.1) plus one dword for the common header shared by all CDAT structures.

What can happen is we receive a truncated entry with, say, just two dwords, and we copy that to the cached CDAT exposed in sysfs. Obviously that's a corrupt CDAT. The consumer of the CDAT doesn't know that the data was truncated and will try to parse the entry even though the data may actually be a portion of the next entry.

The improved verification here ensures that the received amount of data corresponds to the length in the entry header.

It is up to the parsing function of each entry to verify whether that length is correct for that specific entry. E.g. a DSMAS entry should contain 24 bytes, so the parsing function needs to verify that the length in the entry header is actually 24. (See the e-mail I sent to Dave a few minutes ago.)

Thanks,

Lukas

Quoted 8 lines
Quoted 8 lines
drivers/cxl/core/pci.c +7 -0
+		/* 1 DW Table Access Response Header + CDAT entry */
+		entry = (struct cdat_entry_header *)(t.response_pl + 1);
+		if ((entry_handle == 0 &&
+		     t.task.rv != sizeof(u32) + sizeof(struct cdat_header)) ||
+		    (entry_handle > 0 &&
+		     (t.task.rv < sizeof(u32) + sizeof(struct cdat_entry_header) ||
+		      t.task.rv != sizeof(u32) + le16_to_cpu(entry->length))))
 			return -EIO;
Reviewed-by on v3 3/16
#18 lore raw

Re: [PATCH v3 03/16] cxl/pci: Handle truncated CDAT entries

In reply to Lukas Wunner
Quoted 23 lines · On Fri, 10 Feb 2023 21:25:03 +0100 Lukas Wunner <[email protected]> wrote:

If truncated CDAT entries are received from a device, the concatenation of those entries constitutes a corrupt CDAT, yet is happily exposed to user space.

Avoid by verifying response lengths and erroring out if truncation is detected.

The last CDAT entry may still be truncated despite the checks introduced herein if the length in the CDAT header is too small. However, that is easily detectable by user space because it reaches EOF prematurely. A subsequent commit which rightsizes the CDAT response allocation closes that remaining loophole.

The two lines introduced here which exceed 80 chars are shortened to less than 80 chars by a subsequent commit which migrates to a synchronous DOE API and replaces "t.task.rv" by "rc".

The existing acpi_cdat_header and acpi_table_cdat struct definitions provided by ACPICA cannot be used because they do not employ __le16 or __le32 types. I believe that cannot be changed because those types are Linux-specific and ACPI is specified for little endian platforms only, hence doesn't care about endianness. So duplicate the structs.

Quoted 4 lines
Fixes c97006046c79 ("cxl/port: Read CDAT table")

One trivial suggestion.

Glad to see this tightened up.

Quoted 25 lines

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 13 +++++++++----
 drivers/cxl/cxlpci.h   | 14 ++++++++++++++
 2 files changed, 23 insertions(+), 4 deletions(-)
drivers/cxl/core/pci.c +1 -1
index 11a85b3a9a0b..a3fb6bd68d17 100644
@@ -547,8 +547,8 @@ static int cxl_cdat_read_table(struct device *dev,
 
 	do {
 		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
+		struct cdat_entry_header *entry;
 		size_t entry_dw;
-		__le32 *entry;
 		int rc;
 
 		rc = pci_doe_submit_task(cdat_doe, &t.task);
@@ -557,14 +557,19 @@ static int cxl_cdat_read_table(struct device *dev,
 			return rc;
 		}

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 566

Quoted 10 lines
drivers/cxl/core/pci.c +7 -2
 		wait_for_completion(&t.c);
-		/* 1 DW header + 1 DW data min */
-		if (t.task.rv < (2 * sizeof(u32)))
+
+		/* 1 DW Table Access Response Header + CDAT entry */
+		entry = (struct cdat_entry_header *)(t.response_pl + 1);
+		if ((entry_handle == 0 &&
+		     t.task.rv != sizeof(u32) + sizeof(struct cdat_header)) ||
+		    (entry_handle > 0 &&
+		     (t.task.rv < sizeof(u32) + sizeof(struct cdat_entry_header) ||

Slight preference for sizeof(*entry)
so that it is clear that we are checking we can do the next check without getting garbage.

Quoted 35 lines
drivers/cxl/core/pci.c +1 -1
+		      t.task.rv != sizeof(u32) + le16_to_cpu(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(t.response_pl[0]));
-		entry = t.response_pl + 1;
 		entry_dw = t.task.rv / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
drivers/cxl/cxlpci.h +14 -0
index 920909791bb9..104ad2b72516 100644
@@ -62,6 +62,20 @@ enum cxl_regloc_type {
 	CXL_REGLOC_RBI_TYPES
 };
 
+struct cdat_header {
+	__le32 length;
+	u8 revision;
+	u8 checksum;
+	u8 reserved[6];
+	__le32 sequence;
+} __packed;
+
+struct cdat_entry_header {
+	u8 type;
+	u8 reserved;
+	__le16 length;
+} __packed;
+
 int devm_cxl_port_enumerate_dports(struct cxl_port *port);
 struct cxl_dev_state;
 int cxl_hdm_decode_init(struct cxl_dev_state *cxlds, struct cxl_hdm *cxlhdm);
#19 lore raw

PATCH v3 4/16 cxl/pci: Handle excessive CDAT length

In reply to Lukas Wunner

If the length in the CDAT header is larger than the concatenation of the header and all table entries, then the CDAT exposed to user space contains trailing null bytes.

Not every consumer may be able to handle that. Per Postel's robustness principle, "be liberal in what you accept" and silently reduce the cached length to avoid exposing those null bytes.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/cxl/core/pci.c +3 -0
index a3fb6bd68d17..c37c41d7acb6 100644
@@ -582,6 +582,9 @@ static int cxl_cdat_read_table(struct device *dev,
 		}
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
+	/* Length in CDAT header may exceed concatenation of CDAT entries */
+	cdat->length -= length;
+
 	return 0;
 }
 
Signature · 1 line
2.39.1
Reviewed-by on v3 4/16
#20 lore raw

RE: [PATCH v3 04/16] cxl/pci: Handle excessive CDAT length

Quoted 24 lines · Lukas Wunner

If the length in the CDAT header is larger than the concatenation of the header and all table entries, then the CDAT exposed to user space contains trailing null bytes.

Not every consumer may be able to handle that. Per Postel's robustness principle, "be liberal in what you accept" and silently reduce the cached length to avoid exposing those null bytes.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/cxl/core/pci.c +0 -0
index a3fb6bd68d17..c37c41d7acb6 100644

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 587

Quoted 6 lines
drivers/cxl/core/pci.c +3 -0
@@ -582,6 +582,9 @@ static int cxl_cdat_read_table(struct device *dev,
 		}
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
+	/* Length in CDAT header may exceed concatenation of CDAT entries */
+	cdat->length -= length;
+

Looks good,

Reviewed-by on v3 4/16
#21 lore raw

Re: [PATCH v3 04/16] cxl/pci: Handle excessive CDAT length

In reply to Lukas Wunner
Quoted 4 lines · On Fri, 10 Feb 2023 21:25:04 +0100 Lukas Wunner <[email protected]> wrote:

If the length in the CDAT header is larger than the concatenation of the header and all table entries, then the CDAT exposed to user space contains trailing null bytes.

Quoted 8 lines

Not every consumer may be able to handle that. Per Postel's robustness principle, "be liberal in what you accept" and silently reduce the cached length to avoid exposing those null bytes.

Fixes c97006046c79 ("cxl/port: Read CDAT table")

Fair enough. I'd argue that we are papering over broken hardware if we hit these conditions, so given we aren't aware of any (I hope) not sure this is stable material. Argument in favor of stable being that if we do get broken hardware we don't want an ABI change when we paper over the garbage... hmm.

Quoted 21 lines

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/cxl/core/pci.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/cxl/core/pci.c +3 -0
index a3fb6bd68d17..c37c41d7acb6 100644
@@ -582,6 +582,9 @@ static int cxl_cdat_read_table(struct device *dev,
 		}
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
+	/* Length in CDAT header may exceed concatenation of CDAT entries */
+	cdat->length -= length;
+
 	return 0;
 }
 
#22 lore raw

Re: [PATCH v3 04/16] cxl/pci: Handle excessive CDAT length

Quoted 8 lines · Jonathan Cameron
Quoted 7 lines · On Fri, 10 Feb 2023 21:25:04 +0100 Lukas Wunner <[email protected]> wrote:

If the length in the CDAT header is larger than the concatenation of the header and all table entries, then the CDAT exposed to user space contains trailing null bytes.

Not every consumer may be able to handle that. Per Postel's robustness principle, "be liberal in what you accept" and silently reduce the cached length to avoid exposing those null bytes.

[...]

Quoted 5 lines

Fair enough. I'd argue that we are papering over broken hardware if we hit these conditions, so given we aren't aware of any (I hope) not sure this is stable material. Argument in favor of stable being that if we do get broken hardware we don't want an ABI change when we paper over the garbage... hmm.

Type 0 is assigned for DSMAS structures. So user space might believe there's an additional DSMAS in the CDAT. It *could* detect that the length is bogus (it is 0 but should be 24), but what if it doesn't check that? It seems way too dangerous to leave this loophole open, hence the stable designation.

Thanks,

Lukas

Quoted 12 lines
Quoted 12 lines
drivers/cxl/core/pci.c +3 -0
@@ -582,6 +582,9 @@ static int cxl_cdat_read_table(struct device *dev,
 		}
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
+	/* Length in CDAT header may exceed concatenation of CDAT entries */
+	cdat->length -= length;
+
 	return 0;
 }
 
#23 lore raw

Re: [PATCH v3 04/16] cxl/pci: Handle excessive CDAT length

Quoted 16 lines · On Thu, 16 Feb 2023 11:26:16 +0100 Lukas Wunner <[email protected]> wrote:
Quoted 8 lines · Jonathan Cameron
Quoted 7 lines · On Fri, 10 Feb 2023 21:25:04 +0100 Lukas Wunner <[email protected]> wrote:

If the length in the CDAT header is larger than the concatenation of the header and all table entries, then the CDAT exposed to user space contains trailing null bytes.

Not every consumer may be able to handle that. Per Postel's robustness principle, "be liberal in what you accept" and silently reduce the cached length to avoid exposing those null bytes.

[...]

Quoted 5 lines

Fair enough. I'd argue that we are papering over broken hardware if we hit these conditions, so given we aren't aware of any (I hope) not sure this is stable material. Argument in favor of stable being that if we do get broken hardware we don't want an ABI change when we paper over the garbage... hmm.

Quoted 5 lines

Type 0 is assigned for DSMAS structures. So user space might believe there's an additional DSMAS in the CDAT. It *could* detect that the length is bogus (it is 0 but should be 24), but what if it doesn't check that? It seems way too dangerous to leave this loophole open, hence the stable designation.

Ok

Quoted 17 lines

Thanks,

Lukas

Quoted 12 lines
Quoted 12 lines
drivers/cxl/core/pci.c +3 -0
@@ -582,6 +582,9 @@ static int cxl_cdat_read_table(struct device *dev,
 		}
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
+	/* Length in CDAT header may exceed concatenation of CDAT entries */
+	cdat->length -= length;
+
 	return 0;
 }
 
#24 lore raw

PATCH v3 5/16 PCI/DOE: Silence WARN splat with CONFIG_DEBUG_OBJECTS=y

In reply to Lukas Wunner

Gregory Price reports a WARN splat with CONFIG_DEBUG_OBJECTS=y upon CXL probing because pci_doe_submit_task() invokes INIT_WORK() instead of INIT_WORK_ONSTACK() for a work_struct that was allocated on the stack.

All callers of pci_doe_submit_task() allocate the work_struct on the stack, so replace INIT_WORK() with INIT_WORK_ONSTACK() as a backportable short-term fix.

The long-term fix implemented by a subsequent commit is to move to a synchronous API which allocates the work_struct internally in the DOE library.

Stacktrace for posterity:

WARNING: CPU: 0 PID: 23 at lib/debugobjects.c:545 __debug_object_init.cold+0x18/0x183
CPU: 0 PID: 23 Comm: kworker/u2:1 Not tainted 6.1.0-0.rc1.20221019gitaae703b02f92.17.fc38.x86_64 #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
Call Trace:
 pci_doe_submit_task+0x5d/0xd0
 pci_doe_discovery+0xb4/0x100
 pcim_doe_create_mb+0x219/0x290
 cxl_pci_probe+0x192/0x430
 local_pci_probe+0x41/0x80
 pci_device_probe+0xb3/0x220
 really_probe+0xde/0x380
 __driver_probe_device+0x78/0x170
 driver_probe_device+0x1f/0x90
 __driver_attach_async_helper+0x5c/0xe0
 async_run_entry_fn+0x30/0x130
 process_one_work+0x294/0x5b0

---
Changes v2 -> v3:
* Explain in commit message what the long-term fix implemented by a subsequent commit is (Jonathan)

 drivers/pci/doe.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/pci/doe.c +3 -1
index 69efa9a250b9..c20ca62a8c9d 100644
@@ -523,6 +523,8 @@ EXPORT_SYMBOL_GPL(pci_doe_supports_prot);
  * task->complete will be called when the state machine is done processing this
  * task.
  *
+ * @task must be allocated on the stack.
+ *
  * Excess data will be discarded.
  *
  * RETURNS: 0 when task has been successfully queued, -ERRNO on error
@@ -544,7 +546,7 @@ int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
 		return -EIO;
 
 	task->doe_mb = doe_mb;
-	INIT_WORK(&task->work, doe_statemachine_work);
+	INIT_WORK_ONSTACK(&task->work, doe_statemachine_work);
 	queue_work(doe_mb->work_queue, &task->work);
 	return 0;
 }
Signature · 1 line
2.39.1
#25 lore raw

PATCH v3 6/16 PCI/DOE: Fix memory leak with CONFIG_DEBUG_OBJECTS=y

In reply to Lukas Wunner

After a pci_doe_task completes, its work_struct needs to be destroyed to avoid a memory leak with CONFIG_DEBUG_OBJECTS=y.

Fixes 9d24322e887b ("PCI/DOE: Add DOE mailbox support functions")
---
 drivers/pci/doe.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pci/doe.c +1 -0
index c20ca62a8c9d..6cf0600a38aa 100644
@@ -224,6 +224,7 @@ static void signal_task_complete(struct pci_doe_task *task, int rv)
 {
 	task->rv = rv;
 	task->complete(task);
+	destroy_work_on_stack(&task->work);
 }
 
 static void signal_task_abort(struct pci_doe_task *task, int rv)
Signature · 1 line
2.39.1
Reviewed-by on v3 6/16
#26 lore raw

RE: [PATCH v3 06/16] PCI/DOE: Fix memory leak with CONFIG_DEBUG_OBJECTS=y

Quoted 29 lines · Lukas Wunner

After a pci_doe_task completes, its work_struct needs to be destroyed to avoid a memory leak with CONFIG_DEBUG_OBJECTS=y.

Fixes 9d24322e887b ("PCI/DOE: Add DOE mailbox support functions")
---
 drivers/pci/doe.c | 1 +
 1 file changed, 1 insertion(+)
drivers/pci/doe.c +1 -0
index c20ca62a8c9d..6cf0600a38aa 100644
@@ -224,6 +224,7 @@ static void signal_task_complete(struct pci_doe_task *task, int rv)
 {
 	task->rv = rv;
 	task->complete(task);
+	destroy_work_on_stack(&task->work);
 }
 
 static void signal_task_abort(struct pci_doe_task *task, int rv)
Signature · 1 line
2.39.1

For 5 and 6, carried over from the v2:

1 through 6 seem suitable to go in as fixes at any time.

maintainer Reviewed-by on v3 6/16
#27 lore raw

Re: [PATCH v3 06/16] PCI/DOE: Fix memory leak with CONFIG_DEBUG_OBJECTS=y

In reply to Lukas Wunner
Quoted 9 lines · Lukas Wunner

After a pci_doe_task completes, its work_struct needs to be destroyed to avoid a memory leak with CONFIG_DEBUG_OBJECTS=y.

Fixes 9d24322e887b ("PCI/DOE: Add DOE mailbox support functions")
#28 lore raw

PATCH v3 7/16 PCI/DOE: Provide synchronous API and use it internally

In reply to Lukas Wunner

The DOE API only allows asynchronous exchanges and forces callers to provide a completion callback. Yet all existing callers only perform synchronous exchanges. Upcoming commits for CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31) likewise require only synchronous DOE exchanges.

Provide a synchronous pci_doe() API call which builds on the internal asynchronous machinery.

Convert the internal pci_doe_discovery() to the new call.

The new API allows submission of const-declared requests, necessitating the addition of a const qualifier in struct pci_doe_task.

---
 drivers/pci/doe.c       | 69 ++++++++++++++++++++++++++++++++---------
 include/linux/pci-doe.h |  6 +++-
 2 files changed, 59 insertions(+), 16 deletions(-)
drivers/pci/doe.c +54 -15
index 6cf0600a38aa..d2edae8a32ac 100644
@@ -319,28 +319,17 @@ static int pci_doe_discovery(struct pci_doe_mb *doe_mb, u8 *index, u16 *vid,
 	u32 request_pl = FIELD_PREP(PCI_DOE_DATA_OBJECT_DISC_REQ_3_INDEX,
 				    *index);
 	u32 response_pl;
-	DECLARE_COMPLETION_ONSTACK(c);
-	struct pci_doe_task task = {
-		.prot.vid = PCI_VENDOR_ID_PCI_SIG,
-		.prot.type = PCI_DOE_PROTOCOL_DISCOVERY,
-		.request_pl = (__le32 *)&request_pl,
-		.request_pl_sz = sizeof(request_pl),
-		.response_pl = (__le32 *)&response_pl,
-		.response_pl_sz = sizeof(response_pl),
-		.complete = pci_doe_task_complete,
-		.private = &c,
-	};
 	int rc;
 
 	cpu_to_le32s(&request_pl);
 
-	rc = pci_doe_submit_task(doe_mb, &task);
+	rc = pci_doe(doe_mb, PCI_VENDOR_ID_PCI_SIG, PCI_DOE_PROTOCOL_DISCOVERY,
+		     &request_pl, sizeof(request_pl),
+		     &response_pl, sizeof(response_pl));
 	if (rc < 0)
 		return rc;
 
-	wait_for_completion(&c);
-
-	if (task.rv != sizeof(response_pl))
+	if (rc != sizeof(response_pl))
 		return -EIO;
 
 	le32_to_cpus(&response_pl);
@@ -552,3 +541,53 @@ int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
 	return 0;
 }
 EXPORT_SYMBOL_GPL(pci_doe_submit_task);
+
+/**
+ * pci_doe() - Perform Data Object Exchange
+ *
+ * @doe_mb: DOE Mailbox
+ * @vendor: Vendor ID
+ * @type: Data Object Type
+ * @request: Request payload
+ * @request_sz: Size of request payload (bytes)
+ * @response: Response payload
+ * @response_sz: Size of response payload (bytes)
+ *
+ * Submit @request to @doe_mb and store the @response.
+ * The DOE exchange is performed synchronously and may therefore sleep.
+ *
+ * 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().
+ *
+ * RETURNS: Length of received response or negative errno.
+ * Received data in excess of @response_sz is discarded.
+ * The length may be smaller than @response_sz and the caller
+ * is responsible for checking that.
+ */
+int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
+	    const void *request, size_t request_sz,
+	    void *response, size_t response_sz)
+{
+	DECLARE_COMPLETION_ONSTACK(c);
+	struct pci_doe_task task = {
+		.prot.vid = vendor,
+		.prot.type = type,
+		.request_pl = request,
+		.request_pl_sz = request_sz,
+		.response_pl = response,
+		.response_pl_sz = response_sz,
+		.complete = pci_doe_task_complete,
+		.private = &c,
+	};
+	int rc;
+
+	rc = pci_doe_submit_task(doe_mb, &task);
+	if (rc)
+		return rc;
+
+	wait_for_completion(&c);
+
+	return task.rv;
+}
+EXPORT_SYMBOL_GPL(pci_doe);
include/linux/pci-doe.h +5 -1
index 43765eaf2342..5dcd54f892e5 100644
@@ -49,7 +49,7 @@ struct pci_doe_mb;
  */
 struct pci_doe_task {
 	struct pci_doe_protocol prot;
-	__le32 *request_pl;
+	const __le32 *request_pl;
 	size_t request_pl_sz;
 	__le32 *response_pl;
 	size_t response_pl_sz;
@@ -78,4 +78,8 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset);
 bool pci_doe_supports_prot(struct pci_doe_mb *doe_mb, u16 vid, u8 type);
 int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task);
 
+int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
+	    const void *request, size_t request_sz,
+	    void *response, size_t response_sz);
+
 #endif
Signature · 1 line
2.39.1
Reviewed-by on v3 7/16
#29 lore raw

Re: [PATCH v3 07/16] PCI/DOE: Provide synchronous API and use it internally

Quoted 11 lines · Lukas Wunner

The DOE API only allows asynchronous exchanges and forces callers to provide a completion callback. Yet all existing callers only perform synchronous exchanges. Upcoming commits for CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31) likewise require only synchronous DOE exchanges.

Provide a synchronous pci_doe() API call which builds on the internal asynchronous machinery.

Convert the internal pci_doe_discovery() to the new call.

Quoted 8 lines

The new API allows submission of const-declared requests, necessitating the addition of a const qualifier in struct pci_doe_task.

maintainer Reviewed-by on v3 7/16
#30 lore raw

Re: [PATCH v3 07/16] PCI/DOE: Provide synchronous API and use it internally

In reply to Lukas Wunner
Quoted 11 lines · Lukas Wunner

The DOE API only allows asynchronous exchanges and forces callers to provide a completion callback. Yet all existing callers only perform synchronous exchanges. Upcoming commits for CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31) likewise require only synchronous DOE exchanges.

Provide a synchronous pci_doe() API call which builds on the internal asynchronous machinery.

Convert the internal pci_doe_discovery() to the new call.

Quoted 8 lines

The new API allows submission of const-declared requests, necessitating the addition of a const qualifier in struct pci_doe_task.

#31 lore raw

PATCH v3 8/16 cxl/pci: Use synchronous API for DOE

In reply to Lukas Wunner

A synchronous API for DOE has just been introduced. Convert CXL CDAT retrieval over to it.

---
 drivers/cxl/core/pci.c | 66 ++++++++++++++----------------------------
 1 file changed, 22 insertions(+), 44 deletions(-)
drivers/cxl/core/pci.c +22 -44
index c37c41d7acb6..09a9fa67d12a 100644
@@ -487,51 +487,26 @@ static struct pci_doe_mb *find_cdat_doe(struct device *uport)
 		    CXL_DOE_TABLE_ACCESS_TABLE_TYPE_CDATA) |		\
 	 FIELD_PREP(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE, (entry_handle)))
 
-static void cxl_doe_task_complete(struct pci_doe_task *task)
-{
-	complete(task->private);
-}
-
-struct cdat_doe_task {
-	__le32 request_pl;
-	__le32 response_pl[32];
-	struct completion c;
-	struct pci_doe_task task;
-};
-
-#define DECLARE_CDAT_DOE_TASK(req, cdt)                       \
-struct cdat_doe_task cdt = {                                  \
-	.c = COMPLETION_INITIALIZER_ONSTACK(cdt.c),           \
-	.request_pl = req,				      \
-	.task = {                                             \
-		.prot.vid = PCI_DVSEC_VENDOR_ID_CXL,        \
-		.prot.type = CXL_DOE_PROTOCOL_TABLE_ACCESS, \
-		.request_pl = &cdt.request_pl,                \
-		.request_pl_sz = sizeof(cdt.request_pl),      \
-		.response_pl = cdt.response_pl,               \
-		.response_pl_sz = sizeof(cdt.response_pl),    \
-		.complete = cxl_doe_task_complete,            \
-		.private = &cdt.c,                            \
-	}                                                     \
-}
-
 static int cxl_cdat_get_length(struct device *dev,
 			       struct pci_doe_mb *cdat_doe,
 			       size_t *length)
 {
-	DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(0), t);
+	__le32 request = CDAT_DOE_REQ(0);
+	__le32 response[32];
 	int rc;
 
-	rc = pci_doe_submit_task(cdat_doe, &t.task);
+	rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
+		     CXL_DOE_PROTOCOL_TABLE_ACCESS,
+		     &request, sizeof(request),
+		     &response, sizeof(response));
 	if (rc < 0) {
-		dev_err(dev, "DOE submit failed: %d", rc);
+		dev_err(dev, "DOE failed: %d", rc);
 		return rc;
 	}
-	wait_for_completion(&t.c);
-	if (t.task.rv < 2 * sizeof(u32))
+	if (rc < 2 * sizeof(u32))
 		return -EIO;
 
-	*length = le32_to_cpu(t.response_pl[1]);
+	*length = le32_to_cpu(response[1]);
 	dev_dbg(dev, "CDAT length %zu\n", *length);
 
 	return 0;
@@ -546,31 +521,34 @@ static int cxl_cdat_read_table(struct device *dev,
 	int entry_handle = 0;
 
 	do {
-		DECLARE_CDAT_DOE_TASK(CDAT_DOE_REQ(entry_handle), t);
+		__le32 request = CDAT_DOE_REQ(entry_handle);
 		struct cdat_entry_header *entry;
+		__le32 response[32];
 		size_t entry_dw;
 		int rc;
 
-		rc = pci_doe_submit_task(cdat_doe, &t.task);
+		rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
+			     CXL_DOE_PROTOCOL_TABLE_ACCESS,
+			     &request, sizeof(request),
+			     &response, sizeof(response));
 		if (rc < 0) {
-			dev_err(dev, "DOE submit failed: %d", rc);
+			dev_err(dev, "DOE failed: %d", rc);
 			return rc;
 		}
-		wait_for_completion(&t.c);
 
 		/* 1 DW Table Access Response Header + CDAT entry */
-		entry = (struct cdat_entry_header *)(t.response_pl + 1);
+		entry = (struct cdat_entry_header *)(response + 1);
 		if ((entry_handle == 0 &&
-		     t.task.rv != sizeof(u32) + sizeof(struct cdat_header)) ||
+		     rc != sizeof(u32) + sizeof(struct cdat_header)) ||
 		    (entry_handle > 0 &&
-		     (t.task.rv < sizeof(u32) + sizeof(struct cdat_entry_header) ||
-		      t.task.rv != sizeof(u32) + le16_to_cpu(entry->length))))
+		     (rc < sizeof(u32) + sizeof(struct cdat_entry_header) ||
+		      rc != sizeof(u32) + le16_to_cpu(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(t.response_pl[0]));
-		entry_dw = t.task.rv / sizeof(u32);
+					 le32_to_cpu(response[0]));
+		entry_dw = rc / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
 		entry_dw = min(length / sizeof(u32), entry_dw);
Signature · 1 line
2.39.1
#32 lore raw

PATCH v3 9/16 PCI/DOE: Make asynchronous API private

In reply to Lukas Wunner

A synchronous API for DOE has just been introduced. CXL (the only in-tree DOE user so far) was converted to use it instead of the asynchronous API.

Consequently, pci_doe_submit_task() as well as the pci_doe_task struct are only used internally, so make them private.

---
 drivers/pci/doe.c       | 45 ++++++++++++++++++++++++++++++++++++--
 include/linux/pci-doe.h | 48 -----------------------------------------
 2 files changed, 43 insertions(+), 50 deletions(-)
drivers/pci/doe.c +43 -2
index d2edae8a32ac..afb53bc1b4aa 100644
@@ -56,6 +56,47 @@ struct pci_doe_mb {
 	unsigned long flags;
 };
 
+struct pci_doe_protocol {
+	u16 vid;
+	u8 type;
+};
+
+/**
+ * struct pci_doe_task - represents a single query/response
+ *
+ * @prot: DOE Protocol
+ * @request_pl: The request payload
+ * @request_pl_sz: Size of the request payload (bytes)
+ * @response_pl: The response payload
+ * @response_pl_sz: Size of the response payload (bytes)
+ * @rv: Return value.  Length of received response or error (bytes)
+ * @complete: Called when task is complete
+ * @private: Private data for the consumer
+ * @work: Used internally by the mailbox
+ * @doe_mb: Used internally by the mailbox
+ *
+ * The payload sizes and rv are specified in bytes with the following
+ * restrictions concerning the protocol.
+ *
+ *	1) The request_pl_sz must be a multiple of double words (4 bytes)
+ *	2) The response_pl_sz must be >= a single double word (4 bytes)
+ *	3) rv is returned as bytes but it will be a multiple of double words
+ */
+struct pci_doe_task {
+	struct pci_doe_protocol prot;
+	const __le32 *request_pl;
+	size_t request_pl_sz;
+	__le32 *response_pl;
+	size_t response_pl_sz;
+	int rv;
+	void (*complete)(struct pci_doe_task *task);
+	void *private;
+
+	/* initialized by pci_doe_submit_task() */
+	struct work_struct work;
+	struct pci_doe_mb *doe_mb;
+};
+
 static int pci_doe_wait(struct pci_doe_mb *doe_mb, unsigned long timeout)
 {
 	if (wait_event_timeout(doe_mb->wq,
@@ -519,7 +560,8 @@ EXPORT_SYMBOL_GPL(pci_doe_supports_prot);
  *
  * RETURNS: 0 when task has been successfully queued, -ERRNO on error
  */
-int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
+static int pci_doe_submit_task(struct pci_doe_mb *doe_mb,
+			       struct pci_doe_task *task)
 {
 	if (!pci_doe_supports_prot(doe_mb, task->prot.vid, task->prot.type))
 		return -EINVAL;
@@ -540,7 +582,6 @@ int pci_doe_submit_task(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
 	queue_work(doe_mb->work_queue, &task->work);
 	return 0;
 }
-EXPORT_SYMBOL_GPL(pci_doe_submit_task);
 
 /**
  * pci_doe() - Perform Data Object Exchange
include/linux/pci-doe.h +0 -48

62 lines not on this page · raw message

Signature · 1 line
2.39.1
Reviewed-by on v3 9/16
#33 lore raw

Re: [PATCH v3 09/16] PCI/DOE: Make asynchronous API private

Quoted 11 lines · Lukas Wunner

A synchronous API for DOE has just been introduced. CXL (the only in-tree DOE user so far) was converted to use it instead of the asynchronous API.

Consequently, pci_doe_submit_task() as well as the pci_doe_task struct are only used internally, so make them private.

#34 lore raw

PATCH v3 10/16 PCI/DOE: Deduplicate mailbox flushing

In reply to Lukas Wunner

When a DOE mailbox is torn down, its workqueue is flushed once in pci_doe_flush_mb() through a call to flush_workqueue() and subsequently flushed once more in pci_doe_destroy_workqueue() through a call to destroy_workqueue().

Deduplicate by dropping flush_workqueue() from pci_doe_flush_mb().

Rename pci_doe_flush_mb() to pci_doe_cancel_tasks() to more aptly describe what it now does.

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/pci/doe.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/pci/doe.c +3 -6
index afb53bc1b4aa..291cd7a46a39 100644
@@ -429,7 +429,7 @@ static void pci_doe_destroy_workqueue(void *mb)
 	destroy_workqueue(doe_mb->work_queue);
 }
 
-static void pci_doe_flush_mb(void *mb)
+static void pci_doe_cancel_tasks(void *mb)
 {
 	struct pci_doe_mb *doe_mb = mb;
 
@@ -439,9 +439,6 @@ static void pci_doe_flush_mb(void *mb)
 	/* Cancel an in progress work item, if necessary */
 	set_bit(PCI_DOE_FLAG_CANCEL, &doe_mb->flags);
 	wake_up(&doe_mb->wq);
-
-	/* Flush all work items */
-	flush_workqueue(doe_mb->work_queue);
 }
 
 /**
@@ -498,9 +495,9 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
 
 	/*
 	 * The state machine and the mailbox should be in sync now;
-	 * Set up mailbox flush prior to using the mailbox to query protocols.
+	 * Set up cancel tasks prior to using the mailbox to query protocols.
 	 */
-	rc = devm_add_action_or_reset(dev, pci_doe_flush_mb, doe_mb);
+	rc = devm_add_action_or_reset(dev, pci_doe_cancel_tasks, doe_mb);
 	if (rc)
 		return ERR_PTR(rc);
 
Signature · 1 line
2.39.1
Reviewed-by on v3 10/16
#35 lore raw

Re: [PATCH v3 10/16] PCI/DOE: Deduplicate mailbox flushing

Quoted 5 lines · On Fri, 10 Feb 2023 21:25:10 +0100 Lukas Wunner <[email protected]> wrote:

When a DOE mailbox is torn down, its workqueue is flushed once in pci_doe_flush_mb() through a call to flush_workqueue() and subsequently flushed once more in pci_doe_destroy_workqueue() through a call to destroy_workqueue().

Quoted 7 lines

Deduplicate by dropping flush_workqueue() from pci_doe_flush_mb().

Rename pci_doe_flush_mb() to pci_doe_cancel_tasks() to more aptly describe what it now does.

LGTM

Quoted 42 lines

---
Changes v2 -> v3:
* Newly added patch in v3

 drivers/pci/doe.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/pci/doe.c +3 -6
index afb53bc1b4aa..291cd7a46a39 100644
@@ -429,7 +429,7 @@ static void pci_doe_destroy_workqueue(void *mb)
 	destroy_workqueue(doe_mb->work_queue);
 }
 
-static void pci_doe_flush_mb(void *mb)
+static void pci_doe_cancel_tasks(void *mb)
 {
 	struct pci_doe_mb *doe_mb = mb;
 
@@ -439,9 +439,6 @@ static void pci_doe_flush_mb(void *mb)
 	/* Cancel an in progress work item, if necessary */
 	set_bit(PCI_DOE_FLAG_CANCEL, &doe_mb->flags);
 	wake_up(&doe_mb->wq);
-
-	/* Flush all work items */
-	flush_workqueue(doe_mb->work_queue);
 }
 
 /**
@@ -498,9 +495,9 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
 
 	/*
 	 * The state machine and the mailbox should be in sync now;
-	 * Set up mailbox flush prior to using the mailbox to query protocols.
+	 * Set up cancel tasks prior to using the mailbox to query protocols.
 	 */
-	rc = devm_add_action_or_reset(dev, pci_doe_flush_mb, doe_mb);
+	rc = devm_add_action_or_reset(dev, pci_doe_cancel_tasks, doe_mb);
 	if (rc)
 		return ERR_PTR(rc);
 
Reviewed-by on v3 10/16
#36 lore raw

Re: [PATCH v3 10/16] PCI/DOE: Deduplicate mailbox flushing

In reply to Lukas Wunner
Quoted 5 lines · Lukas Wunner

When a DOE mailbox is torn down, its workqueue is flushed once in pci_doe_flush_mb() through a call to flush_workqueue() and subsequently flushed once more in pci_doe_destroy_workqueue() through a call to destroy_workqueue().

Quoted 7 lines

Deduplicate by dropping flush_workqueue() from pci_doe_flush_mb().

Rename pci_doe_flush_mb() to pci_doe_cancel_tasks() to more aptly describe what it now does.

#37 lore raw

PATCH v3 11/16 PCI/DOE: Allow mailbox creation without devres management

In reply to Lukas Wunner

DOE mailbox creation is currently only possible through a devres-managed API. The lifetime of mailboxes thus ends with driver unbinding.

An upcoming commit will create DOE mailboxes upon device enumeration by the PCI core. Their lifetime shall not be limited by a driver.

Therefore rework pcim_doe_create_mb() into the non-devres-managed pci_doe_create_mb(). Add pci_doe_destroy_mb() for mailbox destruction on device removal.

Provide a devres-managed wrapper under the existing pcim_doe_create_mb() name.

The error path of pcim_doe_create_mb() previously called xa_destroy() if alloc_ordered_workqueue() failed. That's unnecessary because the xarray is still empty at that point. It doesn't need to be destroyed until it's been populated by pci_doe_cache_protocols(). Arrange the error path of the new pci_doe_create_mb() accordingly.

pci_doe_cancel_tasks() is no longer used as callback for devm_add_action(), so refactor it to accept a struct pci_doe_mb pointer instead of a generic void pointer.

---
Changes v2 -> v3:
* Call pci_doe_flush_mb() from pci_doe_destroy_mb() to simplify error paths (Jonathan)
* Explain in commit message why xa_destroy() is reordered in error path of the new pci_doe_create_mb() (Jonathan)

 drivers/pci/doe.c | 103 +++++++++++++++++++++++++++++-----------------
 1 file changed, 66 insertions(+), 37 deletions(-)
drivers/pci/doe.c +66 -37
index 291cd7a46a39..2bc202b64b6a 100644
@@ -37,7 +37,7 @@
  *
  * This state is used to manage a single DOE mailbox capability.  All fields
  * should be considered opaque to the consumers and the structure passed into
- * the helpers below after being created by devm_pci_doe_create()
+ * the helpers below after being created by pci_doe_create_mb().
  *
  * @pdev: PCI device this mailbox belongs to
  * @cap_offset: Capability offset
@@ -415,24 +415,8 @@ static int pci_doe_cache_protocols(struct pci_doe_mb *doe_mb)
 	return 0;
 }
 
-static void pci_doe_xa_destroy(void *mb)
+static void pci_doe_cancel_tasks(struct pci_doe_mb *doe_mb)
 {
-	struct pci_doe_mb *doe_mb = mb;
-
-	xa_destroy(&doe_mb->prots);
-}
-
-static void pci_doe_destroy_workqueue(void *mb)
-{
-	struct pci_doe_mb *doe_mb = mb;
-
-	destroy_workqueue(doe_mb->work_queue);
-}
-
-static void pci_doe_cancel_tasks(void *mb)
-{
-	struct pci_doe_mb *doe_mb = mb;
-
 	/* Stop all pending work items from starting */
 	set_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags);
 
@@ -442,7 +426,7 @@ static void pci_doe_cancel_tasks(void *mb)
 }
 
 /**
- * pcim_doe_create_mb() - Create a DOE mailbox object
+ * pci_doe_create_mb() - Create a DOE mailbox object
  *
  * @pdev: PCI device to create the DOE mailbox for
  * @cap_offset: Offset of the DOE mailbox
@@ -453,24 +437,20 @@ static void pci_doe_cancel_tasks(void *mb)
  * RETURNS: created mailbox object on success
  *	    ERR_PTR(-errno) on failure
  */
-struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
+static struct pci_doe_mb *pci_doe_create_mb(struct pci_dev *pdev,
+					    u16 cap_offset)
 {
 	struct pci_doe_mb *doe_mb;
-	struct device *dev = &pdev->dev;
 	int rc;
 
-	doe_mb = devm_kzalloc(dev, sizeof(*doe_mb), GFP_KERNEL);
+	doe_mb = kzalloc(sizeof(*doe_mb), GFP_KERNEL);
 	if (!doe_mb)
 		return ERR_PTR(-ENOMEM);
 
 	doe_mb->pdev = pdev;
 	doe_mb->cap_offset = cap_offset;
 	init_waitqueue_head(&doe_mb->wq);
-
 	xa_init(&doe_mb->prots);
-	rc = devm_add_action(dev, pci_doe_xa_destroy, doe_mb);
-	if (rc)
-		return ERR_PTR(rc);
 
 	doe_mb->work_queue = alloc_ordered_workqueue("%s %s DOE [%x]", 0,
 						dev_driver_string(&pdev->dev),
@@ -479,36 +459,85 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
 	if (!doe_mb->work_queue) {
 		pci_err(pdev, "[%x] failed to allocate work queue\n",
 			doe_mb->cap_offset);
-		return ERR_PTR(-ENOMEM);
+		rc = -ENOMEM;
+		goto err_free;
 	}
-	rc = devm_add_action_or_reset(dev, pci_doe_destroy_workqueue, doe_mb);
-	if (rc)
-		return ERR_PTR(rc);
 
 	/* Reset the mailbox by issuing an abort */
 	rc = pci_doe_abort(doe_mb);
 	if (rc) {
 		pci_err(pdev, "[%x] failed to reset mailbox with abort command : %d\n",
 			doe_mb->cap_offset, rc);
-		return ERR_PTR(rc);
+		goto err_destroy_wq;
 	}
 
 	/*
 	 * The state machine and the mailbox should be in sync now;
-	 * Set up cancel tasks prior to using the mailbox to query protocols.
+	 * Use the mailbox to query protocols.
 	 */
-	rc = devm_add_action_or_reset(dev, pci_doe_cancel_tasks, doe_mb);
-	if (rc)
-		return ERR_PTR(rc);
-
 	rc = pci_doe_cache_protocols(doe_mb);
 	if (rc) {
 		pci_err(pdev, "[%x] failed to cache protocols : %d\n",
 			doe_mb->cap_offset, rc);
-		return ERR_PTR(rc);
+		goto err_cancel;
 	}
 
 	return doe_mb;
+
+err_cancel:
+	pci_doe_cancel_tasks(doe_mb);
+	xa_destroy(&doe_mb->prots);
+err_destroy_wq:
+	destroy_workqueue(doe_mb->work_queue);
+err_free:
+	kfree(doe_mb);
+	return ERR_PTR(rc);
+}
+
+/**
+ * pci_doe_destroy_mb() - Destroy a DOE mailbox object
+ *
+ * @ptr: Pointer to DOE mailbox
+ *
+ * Destroy all internal data structures created for the DOE mailbox.
+ */
+static void pci_doe_destroy_mb(void *ptr)
+{
+	struct pci_doe_mb *doe_mb = ptr;
+
+	pci_doe_cancel_tasks(doe_mb);
+	xa_destroy(&doe_mb->prots);
+	destroy_workqueue(doe_mb->work_queue);
+	kfree(doe_mb);
+}
+
+/**
+ * pcim_doe_create_mb() - Create a DOE mailbox object
+ *
+ * @pdev: PCI device to create the DOE mailbox for
+ * @cap_offset: Offset of the DOE mailbox
+ *
+ * Create a single mailbox object to manage the mailbox protocol at the
+ * cap_offset specified.  The mailbox will automatically be destroyed on
+ * driver unbinding from @pdev.
+ *
+ * RETURNS: created mailbox object on success
+ *	    ERR_PTR(-errno) on failure
+ */
+struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
+{
+	struct pci_doe_mb *doe_mb;
+	int rc;
+
+	doe_mb = pci_doe_create_mb(pdev, cap_offset);
+	if (IS_ERR(doe_mb))
+		return doe_mb;
+
+	rc = devm_add_action_or_reset(&pdev->dev, pci_doe_destroy_mb, doe_mb);
+	if (rc)
+		return ERR_PTR(rc);
+
+	return doe_mb;
 }
 EXPORT_SYMBOL_GPL(pcim_doe_create_mb);
 
Signature · 1 line
2.39.1
Reviewed-by on v3 11/16
#38 lore raw

Re: [PATCH v3 11/16] PCI/DOE: Allow mailbox creation without devres management

Quoted 19 lines · On Fri, 10 Feb 2023 21:25:11 +0100 Lukas Wunner <[email protected]> wrote:

DOE mailbox creation is currently only possible through a devres-managed API. The lifetime of mailboxes thus ends with driver unbinding.

An upcoming commit will create DOE mailboxes upon device enumeration by the PCI core. Their lifetime shall not be limited by a driver.

Therefore rework pcim_doe_create_mb() into the non-devres-managed pci_doe_create_mb(). Add pci_doe_destroy_mb() for mailbox destruction on device removal.

Provide a devres-managed wrapper under the existing pcim_doe_create_mb() name.

The error path of pcim_doe_create_mb() previously called xa_destroy() if alloc_ordered_workqueue() failed. That's unnecessary because the xarray is still empty at that point. It doesn't need to be destroyed until it's been populated by pci_doe_cache_protocols(). Arrange the error path of the new pci_doe_create_mb() accordingly.

Quoted 6 lines

pci_doe_cancel_tasks() is no longer used as callback for devm_add_action(), so refactor it to accept a struct pci_doe_mb pointer instead of a generic void pointer.

Reviewed-by on v3 11/16
#39 lore raw

Re: [PATCH v3 11/16] PCI/DOE: Allow mailbox creation without devres management

In reply to Lukas Wunner
Quoted 19 lines · Lukas Wunner

DOE mailbox creation is currently only possible through a devres-managed API. The lifetime of mailboxes thus ends with driver unbinding.

An upcoming commit will create DOE mailboxes upon device enumeration by the PCI core. Their lifetime shall not be limited by a driver.

Therefore rework pcim_doe_create_mb() into the non-devres-managed pci_doe_create_mb(). Add pci_doe_destroy_mb() for mailbox destruction on device removal.

Provide a devres-managed wrapper under the existing pcim_doe_create_mb() name.

The error path of pcim_doe_create_mb() previously called xa_destroy() if alloc_ordered_workqueue() failed. That's unnecessary because the xarray is still empty at that point. It doesn't need to be destroyed until it's been populated by pci_doe_cache_protocols(). Arrange the error path of the new pci_doe_create_mb() accordingly.

Quoted 6 lines

pci_doe_cancel_tasks() is no longer used as callback for devm_add_action(), so refactor it to accept a struct pci_doe_mb pointer instead of a generic void pointer.

#40 lore raw

PATCH v3 12/16 PCI/DOE: Create mailboxes on device enumeration

In reply to Lukas Wunner

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance. For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Overcome this limitation by creating mailboxes in the PCI core on device enumeration.

Provide a pci_find_doe_mailbox() API call to allow drivers to get a pci_doe_mb for a given (pci_dev, vendor, protocol) triple. This API is modeled after pci_find_capability() and can later be amended with a pci_find_next_doe_mailbox() call to iterate over all mailboxes of a given pci_dev which support a specific protocol.

On removal, destroy the mailboxes in pci_destroy_dev(), after the driver is unbound. This allows drivers to use DOE in their ->remove() hook.

On surprise removal, cancel ongoing DOE exchanges and prevent new ones from being scheduled. Thereby ensure that a hot-removed device doesn't needlessly wait for a running exchange to time out.

---
Changes v2 -> v3:
* Don't cancel ongoing DOE exchanges in pci_stop_dev() so that drivers may perform DOE in their ->remove() hooks
* Instead cancel ongoing DOE exchanges on surprise removal in pci_dev_set_disconnected()
* Emit error message in pci_doe_init() if mailbox creation fails (Ira)
* Explain in commit message that pci_find_doe_mailbox() can later be amended with pci_find_next_doe_mailbox() (Jonathan)

 drivers/pci/doe.c       | 73 +++++++++++++++++++++++++++++++++++++++++
 drivers/pci/pci.h       | 12 +++++++
 drivers/pci/probe.c     |  1 +
 drivers/pci/remove.c    |  1 +
 include/linux/pci-doe.h |  2 ++
 include/linux/pci.h     |  3 ++
 6 files changed, 92 insertions(+)
drivers/pci/doe.c +73 -0
index 2bc202b64b6a..bf32875d27da 100644
@@ -20,6 +20,8 @@
 #include <linux/pci-doe.h>
 #include <linux/workqueue.h>
 
+#include "pci.h"
+
 #define PCI_DOE_PROTOCOL_DISCOVERY 0
 
 /* Timeout of 1 second from 6.30.2 Operation, PCI Spec r6.0 */
@@ -658,3 +660,74 @@ int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
 	return task.rv;
 }
 EXPORT_SYMBOL_GPL(pci_doe);
+
+/**
+ * pci_find_doe_mailbox() - Find Data Object Exchange mailbox
+ *
+ * @pdev: PCI device
+ * @vendor: Vendor ID
+ * @type: Data Object Type
+ *
+ * Find first DOE mailbox of a PCI device which supports the given protocol.
+ *
+ * RETURNS: Pointer to the DOE mailbox or NULL if none was found.
+ */
+struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
+					u8 type)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		if (pci_doe_supports_prot(doe_mb, vendor, type))
+			return doe_mb;
+
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(pci_find_doe_mailbox);
+
+void pci_doe_init(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	u16 offset = 0;
+	int rc;
+
+	xa_init(&pdev->doe_mbs);
+
+	while ((offset = pci_find_next_ext_capability(pdev, offset,
+						      PCI_EXT_CAP_ID_DOE))) {
+		doe_mb = pci_doe_create_mb(pdev, offset);
+		if (IS_ERR(doe_mb)) {
+			pci_err(pdev, "[%x] failed to create mailbox: %ld\n",
+				offset, PTR_ERR(doe_mb));
+			continue;
+		}
+
+		rc = xa_insert(&pdev->doe_mbs, offset, doe_mb, GFP_KERNEL);
+		if (rc) {
+			pci_err(pdev, "[%x] failed to insert mailbox: %d\n",
+				offset, rc);
+			pci_doe_destroy_mb(doe_mb);
+		}
+	}
+}
+
+void pci_doe_destroy(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		pci_doe_destroy_mb(doe_mb);
+
+	xa_destroy(&pdev->doe_mbs);
+}
+
+void pci_doe_disconnected(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		pci_doe_cancel_tasks(doe_mb);
+}
drivers/pci/pci.h +12 -0
index 8f5d4bd5b410..065ca9743ec1 100644
@@ -318,6 +318,16 @@ struct pci_sriov {
 	bool		drivers_autoprobe; /* Auto probing of VFs by driver */
 };
 
+#ifdef CONFIG_PCI_DOE
+void pci_doe_init(struct pci_dev *pdev);
+void pci_doe_destroy(struct pci_dev *pdev);
+void pci_doe_disconnected(struct pci_dev *pdev);
+#else
+static inline void pci_doe_init(struct pci_dev *pdev) { }
+static inline void pci_doe_destroy(struct pci_dev *pdev) { }
+static inline void pci_doe_disconnected(struct pci_dev *pdev) { }
+#endif
+
 /**
  * pci_dev_set_io_state - Set the new error state if possible.
  *
@@ -372,6 +382,8 @@ static inline int pci_dev_set_disconnected(struct pci_dev *dev, void *unused)
 	pci_dev_set_io_state(dev, pci_channel_io_perm_failure);
 	device_unlock(&dev->dev);
 
+	pci_doe_disconnected(dev);
+
 	return 0;
 }
 
drivers/pci/probe.c +1 -0
index 1779582fb500..65e60ee50489 100644
@@ -2476,6 +2476,7 @@ static void pci_init_capabilities(struct pci_dev *dev)
 	pci_aer_init(dev);		/* Advanced Error Reporting */
 	pci_dpc_init(dev);		/* Downstream Port Containment */
 	pci_rcec_init(dev);		/* Root Complex Event Collector */
+	pci_doe_init(dev);		/* Data Object Exchange */
 
 	pcie_report_downtraining(dev);
 	pci_init_reset_methods(dev);
drivers/pci/remove.c +1 -0
index 0145aef1b930..f25acf50879f 100644
@@ -39,6 +39,7 @@ static void pci_destroy_dev(struct pci_dev *dev)
 	list_del(&dev->bus_list);
 	up_write(&pci_bus_sem);
 
+	pci_doe_destroy(dev);
 	pcie_aspm_exit_link_state(dev);
 	pci_bridge_d3_update(dev);
 	pci_free_resources(dev);
include/linux/pci-doe.h +2 -0
index 7f16749c6aa3..d6192ee0ac07 100644
@@ -29,6 +29,8 @@ struct pci_doe_mb;
 
 struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset);
 bool pci_doe_supports_prot(struct pci_doe_mb *doe_mb, u16 vid, u8 type);
+struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
+					u8 type);
 
 int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
 	    const void *request, size_t request_sz,
include/linux/pci.h +3 -0
index 3c51cac3890b..b19c2965e384 100644
@@ -511,6 +511,9 @@ struct pci_dev {
 #endif
 #ifdef CONFIG_PCI_P2PDMA
 	struct pci_p2pdma __rcu *p2pdma;
+#endif
+#ifdef CONFIG_PCI_DOE
+	struct xarray	doe_mbs;	/* Data Object Exchange mailboxes */
 #endif
 	u16		acs_cap;	/* ACS Capability offset */
 	phys_addr_t	rom;		/* Physical address if not from BAR */
Signature · 1 line
2.39.1
Reviewed-by on v3 12/16
#41 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

Quoted 17 lines · Lukas Wunner

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance. For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Overcome this limitation by creating mailboxes in the PCI core on device enumeration.

Provide a pci_find_doe_mailbox() API call to allow drivers to get a pci_doe_mb for a given (pci_dev, vendor, protocol) triple. This API is modeled after pci_find_capability() and can later be amended with a pci_find_next_doe_mailbox() call to iterate over all mailboxes of a given pci_dev which support a specific protocol.

On removal, destroy the mailboxes in pci_destroy_dev(), after the driver is unbound. This allows drivers to use DOE in their ->remove() hook.

Quoted 8 lines

On surprise removal, cancel ongoing DOE exchanges and prevent new ones from being scheduled. Thereby ensure that a hot-removed device doesn't needlessly wait for a running exchange to time out.

#42 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

In reply to Lukas Wunner
Quoted 2 lines · Lukas Wunner

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance.

Sorry for my ignorance but why/how/when would a device have multiple drivers bound? Or it only one driver at the time but we also want DOE MBs to survive switching to another (different or newer) driver?

Quoted 2 lines

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

And we want this sharing why? Any example will do. Thanks,

Quoted 209 lines

Overcome this limitation by creating mailboxes in the PCI core on device enumeration.

Provide a pci_find_doe_mailbox() API call to allow drivers to get a pci_doe_mb for a given (pci_dev, vendor, protocol) triple.

This API is
modeled after pci_find_capability() and can later be amended with a pci_find_next_doe_mailbox() call to iterate over all mailboxes of a given pci_dev which support a specific protocol.

On removal, destroy the mailboxes in pci_destroy_dev(), after the driver is unbound. This allows drivers to use DOE in their ->remove() hook.

On surprise removal, cancel ongoing DOE exchanges and prevent new ones from being scheduled. Thereby ensure that a hot-removed device doesn't needlessly wait for a running exchange to time out.

---
Changes v2 -> v3:
* Don't cancel ongoing DOE exchanges in pci_stop_dev() so that

    drivers may perform DOE in their ->remove() hooks

* Instead cancel ongoing DOE exchanges on surprise removal in pci_dev_set_disconnected()
* Emit error message in pci_doe_init() if mailbox creation fails (Ira)
* Explain in commit message that pci_find_doe_mailbox() can later be amended with pci_find_next_doe_mailbox() (Jonathan)

  drivers/pci/doe.c       | 73 +++++++++++++++++++++++++++++++++++++++++
  drivers/pci/pci.h       | 12 +++++++
  drivers/pci/probe.c     |  1 +
  drivers/pci/remove.c    |  1 +
  include/linux/pci-doe.h |  2 ++
  include/linux/pci.h     |  3 ++
  6 files changed, 92 insertions(+)
drivers/pci/doe.c +73 -0
index 2bc202b64b6a..bf32875d27da 100644
@@ -20,6 +20,8 @@
  #include <linux/pci-doe.h>
  #include <linux/workqueue.h>
 
+#include "pci.h"
+
  #define PCI_DOE_PROTOCOL_DISCOVERY 0
 
  /* Timeout of 1 second from 6.30.2 Operation, PCI Spec r6.0 */
@@ -658,3 +660,74 @@ int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
  	return task.rv;
  }
  EXPORT_SYMBOL_GPL(pci_doe);
+
+/**
+ * pci_find_doe_mailbox() - Find Data Object Exchange mailbox
+ *
+ * @pdev: PCI device
+ * @vendor: Vendor ID
+ * @type: Data Object Type
+ *
+ * Find first DOE mailbox of a PCI device which supports the given protocol.
+ *
+ * RETURNS: Pointer to the DOE mailbox or NULL if none was found.
+ */
+struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
+					u8 type)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		if (pci_doe_supports_prot(doe_mb, vendor, type))
+			return doe_mb;
+
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(pci_find_doe_mailbox);
+
+void pci_doe_init(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	u16 offset = 0;
+	int rc;
+
+	xa_init(&pdev->doe_mbs);
+
+	while ((offset = pci_find_next_ext_capability(pdev, offset,
+						      PCI_EXT_CAP_ID_DOE))) {
+		doe_mb = pci_doe_create_mb(pdev, offset);
+		if (IS_ERR(doe_mb)) {
+			pci_err(pdev, "[%x] failed to create mailbox: %ld\n",
+				offset, PTR_ERR(doe_mb));
+			continue;
+		}
+
+		rc = xa_insert(&pdev->doe_mbs, offset, doe_mb, GFP_KERNEL);
+		if (rc) {
+			pci_err(pdev, "[%x] failed to insert mailbox: %d\n",
+				offset, rc);
+			pci_doe_destroy_mb(doe_mb);
+		}
+	}
+}
+
+void pci_doe_destroy(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		pci_doe_destroy_mb(doe_mb);
+
+	xa_destroy(&pdev->doe_mbs);
+}
+
+void pci_doe_disconnected(struct pci_dev *pdev)
+{
+	struct pci_doe_mb *doe_mb;
+	unsigned long index;
+
+	xa_for_each(&pdev->doe_mbs, index, doe_mb)
+		pci_doe_cancel_tasks(doe_mb);
+}
drivers/pci/pci.h +12 -0
index 8f5d4bd5b410..065ca9743ec1 100644
@@ -318,6 +318,16 @@ struct pci_sriov {
  	bool		drivers_autoprobe; /* Auto probing of VFs by driver */
  };
 
+#ifdef CONFIG_PCI_DOE
+void pci_doe_init(struct pci_dev *pdev);
+void pci_doe_destroy(struct pci_dev *pdev);
+void pci_doe_disconnected(struct pci_dev *pdev);
+#else
+static inline void pci_doe_init(struct pci_dev *pdev) { }
+static inline void pci_doe_destroy(struct pci_dev *pdev) { }
+static inline void pci_doe_disconnected(struct pci_dev *pdev) { }
+#endif
+
  /**
   * pci_dev_set_io_state - Set the new error state if possible.
   *
@@ -372,6 +382,8 @@ static inline int pci_dev_set_disconnected(struct pci_dev *dev, void *unused)
  	pci_dev_set_io_state(dev, pci_channel_io_perm_failure);
  	device_unlock(&dev->dev);
 
+	pci_doe_disconnected(dev);
+
  	return 0;
  }
 
drivers/pci/probe.c +1 -0
index 1779582fb500..65e60ee50489 100644
@@ -2476,6 +2476,7 @@ static void pci_init_capabilities(struct pci_dev *dev)
  	pci_aer_init(dev);		/* Advanced Error Reporting */
  	pci_dpc_init(dev);		/* Downstream Port Containment */
  	pci_rcec_init(dev);		/* Root Complex Event Collector */
+	pci_doe_init(dev);		/* Data Object Exchange */
 
  	pcie_report_downtraining(dev);
  	pci_init_reset_methods(dev);
drivers/pci/remove.c +1 -0
index 0145aef1b930..f25acf50879f 100644
@@ -39,6 +39,7 @@ static void pci_destroy_dev(struct pci_dev *dev)
  	list_del(&dev->bus_list);
  	up_write(&pci_bus_sem);
 
+	pci_doe_destroy(dev);
  	pcie_aspm_exit_link_state(dev);
  	pci_bridge_d3_update(dev);
  	pci_free_resources(dev);
include/linux/pci-doe.h +2 -0
index 7f16749c6aa3..d6192ee0ac07 100644
@@ -29,6 +29,8 @@ struct pci_doe_mb;
 
  struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset);
  bool pci_doe_supports_prot(struct pci_doe_mb *doe_mb, u16 vid, u8 type);
+struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
+					u8 type);
 
  int pci_doe(struct pci_doe_mb *doe_mb, u16 vendor, u8 type,
  	    const void *request, size_t request_sz,
include/linux/pci.h +3 -0
index 3c51cac3890b..b19c2965e384 100644
@@ -511,6 +511,9 @@ struct pci_dev {
  #endif
  #ifdef CONFIG_PCI_P2PDMA
  	struct pci_p2pdma __rcu *p2pdma;
+#endif
+#ifdef CONFIG_PCI_DOE
+	struct xarray	doe_mbs;	/* Data Object Exchange mailboxes */
  #endif
  	u16		acs_cap;	/* ACS Capability offset */
  	phys_addr_t	rom;		/* Physical address if not from BAR */
Signature · 1 line
Alexey
#43 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

Quoted 9 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance.

Sorry for my ignorance but why/how/when would a device have multiple drivers bound? Or it only one driver at the time but we also want DOE MBs to survive switching to another (different or newer) driver?

Quoted 4 lines
Quoted 2 lines

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

And we want this sharing why? Any example will do. Thanks,

I understood "sharing" in this context to mean that the PCI core can use the DOE instance and the device can reuse the same *through* the PCI core. So for example an SPDM session can be established by the core and used later by the driver.

#44 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

In reply to Alexey Kardashevskiy
Quoted 5 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

And we want this sharing why? Any example will do. Thanks,

The PCI core is going to perform CMA/SPDM authentication when a device gets enumerated (PCIe r6.0 sec 6.31). That's the main motivation to lift DOE mailbox creation into the PCI core. It's not mentioned here explicitly because I want the patch to stand on its own. CMA/SPDM support will be submitted separately.

A driver that later on gets bound to the device should be allowed to talk to it via DOE as well, possibly even sharing the same DOE mailboxes used by the PCI core.

Patches for CMA/SPDM are under development on this branch:

https://github.com/l1k/linux/commits/doe

Quoted 6 lines
Quoted 2 lines

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance.

Sorry for my ignorance but why/how/when would a device have multiple drivers bound? Or it only one driver at the time but we also want DOE MBs to survive switching to another (different or newer) driver?

Conceivably, a driver may have the need to talk to multiple devices via DOE, even ones it's not bound to. (E.g. devices in its ancestry or children.)

Thanks,

Lukas

#45 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

Quoted 5 lines · Lukas Wunner
Quoted 4 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Quoted 7 lines
Quoted 1 line

And we want this sharing why? Any example will do. Thanks,

The PCI core is going to perform CMA/SPDM authentication when a device gets enumerated (PCIe r6.0 sec 6.31). That's the main motivation to lift DOE mailbox creation into the PCI core. It's not mentioned here explicitly because I want the patch to stand on its own. CMA/SPDM support will be submitted separately.

I was going the opposite direction with avoiding adding this into the PCI core as 1) the pci_dev struct is already 2K and 2) it is a niche feature and 3) I wanted this CMA/SPDM session setup to be platform specific as on our platform the SPDM support requires some devices to be probed before we can any SPDM.

Quoted 7 lines

A driver that later on gets bound to the device should be allowed to talk to it via DOE as well, possibly even sharing the same DOE mailboxes used by the PCI core.

Patches for CMA/SPDM are under development on this branch:

https://github.com/l1k/linux/commits/doe

yes, thanks! Lots of reading :)

Quoted 10 lines
Quoted 6 lines
Quoted 2 lines

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance.

Sorry for my ignorance but why/how/when would a device have multiple drivers bound? Or it only one driver at the time but we also want DOE MBs to survive switching to another (different or newer) driver?

Conceivably, a driver may have the need to talk to multiple devices via DOE, even ones it's not bound to. (E.g. devices in its ancestry or children.)

Ah ok. Well, a parent device could look for the DOE MB in a child using devres_find(), this requirement alone does not require moving things to the PCI core and potentially allows it to be a module which could be a better way as distros could have it always enabled but it would not waste any memory on my laptop when not loaded. Thanks,

Signature · 1 line
Alexey
#46 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

Quoted 14 lines · On Tue, 28 Feb 2023 18:24:41 +1100 Alexey Kardashevskiy <[email protected]> wrote:
Quoted 5 lines · Lukas Wunner
Quoted 4 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Quoted 7 lines
Quoted 1 line

And we want this sharing why? Any example will do. Thanks,

The PCI core is going to perform CMA/SPDM authentication when a device gets enumerated (PCIe r6.0 sec 6.31). That's the main motivation to lift DOE mailbox creation into the PCI core. It's not mentioned here explicitly because I want the patch to stand on its own. CMA/SPDM support will be submitted separately.

Quoted 5 lines

I was going the opposite direction with avoiding adding this into the PCI core as 1) the pci_dev struct is already 2K and 2) it is a niche feature and 3) I wanted this CMA/SPDM session setup to be platform specific as on our platform the SPDM support requires some devices to be probed before we can any SPDM.

Is that happening over a DOE mailbox, or a different transport? If it's a different transport then that should be fine, though we'll need to have a slightly different security model and any part of early driver load will need to be carefully hardened against a "malicious" device if it is doing anything non trivial. If it's just "turning on the lights" then shouldn't be a problem.

Will be interesting to see how niche DOE ends up. My guess it it's worth a struct xarray, but we could take it out of line if that saves enough to bother.

Quoted 24 lines
Quoted 9 lines

A driver that later on gets bound to the device should be allowed to talk to it via DOE as well, possibly even sharing the same DOE mailboxes used by the PCI core.

Patches for CMA/SPDM are under development on this branch:

https://github.com/l1k/linux/commits/doe

yes, thanks! Lots of reading :)

Quoted 10 lines
Quoted 6 lines
Quoted 2 lines

Currently a DOE instance cannot be shared by multiple drivers because each driver creates its own pci_doe_mb struct for a given DOE instance.

Sorry for my ignorance but why/how/when would a device have multiple drivers bound? Or it only one driver at the time but we also want DOE MBs to survive switching to another (different or newer) driver?

Conceivably, a driver may have the need to talk to multiple devices via DOE, even ones it's not bound to. (E.g. devices in its ancestry or children.)

Quoted 5 lines

Ah ok. Well, a parent device could look for the DOE MB in a child using devres_find(), this requirement alone does not require moving things to the PCI core and potentially allows it to be a module which could be a better way as distros could have it always enabled but it would not waste any memory on my laptop when not loaded. Thanks,

The DOE mailboxes have an "exciting" level of flexibility and discovering supported protocols requires use of the DOE itself. So we need a single entity to take control over concurrent access to each DOE instance.

Given the mix of protocols, I'd expect some of them to be potentially accessed by a parent, and others to be accessed by driver attached to the child.

Whether it needs to chat to it's parent isn't totally clear to me yet, as depends a bit on what entities end up getting created for management of encryption etc + what other usecases we see in medium term.

Jonathan

#47 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

In reply to Alexey Kardashevskiy
Quoted 14 lines · Alexey Kardashevskiy
Quoted 5 lines · Lukas Wunner
Quoted 4 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Quoted 7 lines
Quoted 1 line

And we want this sharing why? Any example will do. Thanks,

The PCI core is going to perform CMA/SPDM authentication when a device gets enumerated (PCIe r6.0 sec 6.31). That's the main motivation to lift DOE mailbox creation into the PCI core. It's not mentioned here explicitly because I want the patch to stand on its own. CMA/SPDM support will be submitted separately.

Quoted 5 lines

I was going the opposite direction with avoiding adding this into the PCI core as 1) the pci_dev struct is already 2K and 2) it is a niche feature and 3) I wanted this CMA/SPDM session setup to be platform specific as on our platform the SPDM support requires some devices to be probed before we can any SPDM.

We had an open discussion at Plumbers with stakeholders from various companies and the consensus was that initially we'll upstream a CMA/SPDM implementation which authenticates devices on enumeration and exposes the result to user space via sysfs. Nothing more, nothing less:

https://lpc.events/event/16/contributions/1304/
https://lpc.events/event/16/contributions/1304/attachments/1029/1974/LPC2022-SPDM-BoF-v4.pdf

Thereby we seek to minimize friction in the upstreaming process because we avoid depending on PCIe features layered on top of CMA/SPDM (such as IDE) and we can postpone controversial discussions about the consequences of failed authentication (such as forbidding driver binding or firewalling the device off via ACS).

The patches under development follow the approach we agreed on back then.

I honestly think that the size of struct pci_dev is a non-issue because even the lowest-end IoT devices come with gigabytes of memory these days and as long as we do not exceed PAGE_SIZE (which is the limit when we'd have to switch from kmalloc() to vmalloc() I believe), there's no problem.

The claim that DOE and CMA/SPDM is going to be a niche feature is at least debatable. It seems rather likely to me that we'll see adoption particularly in the mobile segment as iOS/Android/Chrome promulgators will see an opportunity to harden their products further.

It's intriguing that you need some devices to be probed before you can perform authentication. What do you need this for?

Do you perhaps need to load firmware onto the device before you can authenticate it? Doesn't that defeat the whole point of authentication?

Or do you mean that certain *other* devices need to probe before a device can be authenticated? What are these other devices?

What happens if the device is suspended to D3cold or experiences a Conventional Reset? Do you need to do anything special before you can re-authenticate the device?

On the one hand, performing authentication in the PCI core could be considered a case of midlayer fallacy. On the other hand, I think we do want to afford CMA/SPDM and the features layered on top of it (IDE, TDISP) regardless whether a driver is bound: That way we can protect the PCI core's communication with the device and with ACS we can protect other PCI devices in the system from peer-to-peer communication originating from a malicious device which failed to authenticate.

I would like to make the implementation under development work for your use case as well, but to do that I need a better understanding what your requirements are.

Thanks,

Lukas

#48 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

Quoted 35 lines · Lukas Wunner
Quoted 14 lines · Alexey Kardashevskiy
Quoted 5 lines · Lukas Wunner
Quoted 4 lines · Alexey Kardashevskiy
Quoted 2 lines · Lukas Wunner

For the same reason a DOE instance cannot be shared between the PCI core and a driver.

Quoted 7 lines
Quoted 1 line

And we want this sharing why? Any example will do. Thanks,

The PCI core is going to perform CMA/SPDM authentication when a device gets enumerated (PCIe r6.0 sec 6.31). That's the main motivation to lift DOE mailbox creation into the PCI core. It's not mentioned here explicitly because I want the patch to stand on its own. CMA/SPDM support will be submitted separately.

Quoted 5 lines

I was going the opposite direction with avoiding adding this into the PCI core as 1) the pci_dev struct is already 2K and 2) it is a niche feature and 3) I wanted this CMA/SPDM session setup to be platform specific as on our platform the SPDM support requires some devices to be probed before we can any SPDM.

We had an open discussion at Plumbers with stakeholders from various companies and the consensus was that initially we'll upstream a CMA/SPDM implementation which authenticates devices on enumeration and exposes the result to user space via sysfs. Nothing more, nothing less:

https://lpc.events/event/16/contributions/1304/
https://lpc.events/event/16/contributions/1304/attachments/1029/1974/LPC2022-SPDM-BoF-v4.pdf

Thereby we seek to minimize friction in the upstreaming process because we avoid depending on PCIe features layered on top of CMA/SPDM (such as IDE) and we can postpone controversial discussions about the consequences of failed authentication (such as forbidding driver binding or firewalling the device off via ACS).

Quoted 7 lines

The patches under development follow the approach we agreed on back then.

I honestly think that the size of struct pci_dev is a non-issue because even the lowest-end IoT devices come with gigabytes of memory these days and as long as we do not exceed PAGE_SIZE (which is the limit when we'd have to switch from kmalloc() to vmalloc() I believe), there's no problem.

Well, not sizeof(pci_dev) as much, it is more about DOE/SPDM code being compiled into vmlinux - it is hard for me to imagine this running on my laptop soon but distros dislike multiple kernel configs. Not a huge deal but still.

Quoted 4 lines

The claim that DOE and CMA/SPDM is going to be a niche feature is at least debatable. It seems rather likely to me that we'll see adoption particularly in the mobile segment as iOS/Android/Chrome promulgators will see an opportunity to harden their products further.

These guys all use custom configs (as was also suggested in that threat thread - "Linux guest kernel threat model for Confidential Computing").

Quoted 7 lines

It's intriguing that you need some devices to be probed before you can perform authentication. What do you need this for?

Do you perhaps need to load firmware onto the device before you can authenticate it? Doesn't that defeat the whole point of authentication?

Quoted 1 line

Or do you mean that certain *other* devices need to probe before a device

can be authenticated? What are these other devices?

The AMD CCP device (which represents PSP == a secure processor == TCB for SEV VMs) is a PCI device and when we add TDISP/etc to the TCB, we will need the host to speak to PSP to set this up. It does not matter for host-only IDE support though.

Quoted 8 lines

What happens if the device is suspended to D3cold or experiences a Conventional Reset? Do you need to do anything special before you can re-authenticate the device?

On the one hand, performing authentication in the PCI core could be considered a case of midlayer fallacy. On the other hand, I think we do want to afford CMA/SPDM and the features layered on top of it (IDE, TDISP) regardless whether a driver is bound:

https://cdrdv2-public.intel.com/742542/software-enabling-for-tdx-tee-io-fixed.pdf suggests that TDX is going to have to reimplement SPDM/IDE/TDISP again so the common base here is DOE only, is not it?

And what is the value of TDISP without a device driver, how can the PCI core decide if the device is alright to use? I am sure there is value but my imagination fails me.

Quoted 9 lines

That way we can
protect the PCI core's communication with the device and with ACS we can protect other PCI devices in the system from peer-to-peer communication originating from a malicious device which failed to authenticate.

I would like to make the implementation under development work for your use case as well, but to do that I need a better understanding what your requirements are.

While I am figuring this out, is it a bug? :)
https://github.com/l1k/linux/commit/c1b242d74d55d361934bdcff1bb4f881922b10ae#r102531215

Thanks,

Quoted 3 lines

Thanks,

Lukas

Signature · 1 line
Alexey
#49 lore raw

Re: [PATCH v3 12/16] PCI/DOE: Create mailboxes on device enumeration

In reply to Lukas Wunner
Quoted 3 lines · Lukas Wunner

I would like to make the implementation under development work for your use case as well, but to do that I need a better understanding what your requirements are.

Soooo here is what we are up to:

https://www.amd.com/content/dam/amd/en/documents/developer/sev-tio-whitepaper.pdf

tl;dr: it is secure VFIO == TDISP on AMD SEV-SNP so it is going to use ASP (== PSP == the secure processor) for SPDM/IDE.

Signature · 1 line
Alexey
#50 lore raw

PATCH v3 13/16 cxl/pci: Use CDAT DOE mailbox created by PCI core

In reply to Lukas Wunner

The PCI core has just been amended to create a pci_doe_mb struct for every DOE instance on device enumeration.

Drop creation of a (duplicate) CDAT DOE mailbox on cxl probing in favor of the one already created by the PCI core.

---
 drivers/cxl/core/pci.c | 27 +++++------------------
 drivers/cxl/cxlmem.h   |  3 ---
 drivers/cxl/pci.c      | 49 ------------------------------------------
 3 files changed, 5 insertions(+), 74 deletions(-)
drivers/cxl/core/pci.c +5 -22
index 09a9fa67d12a..1b954783b516 100644
@@ -459,27 +459,6 @@ EXPORT_SYMBOL_NS_GPL(cxl_hdm_decode_init, CXL);
 #define CXL_DOE_TABLE_ACCESS_LAST_ENTRY		0xffff
 #define CXL_DOE_PROTOCOL_TABLE_ACCESS 2
 
-static struct pci_doe_mb *find_cdat_doe(struct device *uport)
-{
-	struct cxl_memdev *cxlmd;
-	struct cxl_dev_state *cxlds;
-	unsigned long index;
-	void *entry;
-
-	cxlmd = to_cxl_memdev(uport);
-	cxlds = cxlmd->cxlds;
-
-	xa_for_each(&cxlds->doe_mbs, index, entry) {
-		struct pci_doe_mb *cur = entry;
-
-		if (pci_doe_supports_prot(cur, PCI_DVSEC_VENDOR_ID_CXL,
-					  CXL_DOE_PROTOCOL_TABLE_ACCESS))
-			return cur;
-	}
-
-	return NULL;
-}
-
 #define CDAT_DOE_REQ(entry_handle) cpu_to_le32				\
 	(FIELD_PREP(CXL_DOE_TABLE_ACCESS_REQ_CODE,			\
 		    CXL_DOE_TABLE_ACCESS_REQ_CODE_READ) |		\
@@ -577,10 +556,14 @@ void read_cdat_data(struct cxl_port *port)
 	struct pci_doe_mb *cdat_doe;
 	struct device *dev = &port->dev;
 	struct device *uport = port->uport;
+	struct cxl_memdev *cxlmd = to_cxl_memdev(uport);
+	struct cxl_dev_state *cxlds = cxlmd->cxlds;
+	struct pci_dev *pdev = to_pci_dev(cxlds->dev);
 	size_t cdat_length;
 	int rc;
 
-	cdat_doe = find_cdat_doe(uport);
+	cdat_doe = pci_find_doe_mailbox(pdev, PCI_DVSEC_VENDOR_ID_CXL,
+					CXL_DOE_PROTOCOL_TABLE_ACCESS);
 	if (!cdat_doe) {
 		dev_dbg(dev, "No CDAT mailbox\n");
 		return;
drivers/cxl/cxlmem.h +0 -3
index ab138004f644..e1a1b23cf56c 100644
@@ -227,7 +227,6 @@ struct cxl_endpoint_dvsec_info {
  * @component_reg_phys: register base of component registers
  * @info: Cached DVSEC information about the device.
  * @serial: PCIe Device Serial Number
- * @doe_mbs: PCI DOE mailbox array
  * @mbox_send: @dev specific transport for transmitting mailbox commands
  *
  * See section 8.2.9.5.2 Capacity Configuration and Label Storage for
@@ -264,8 +263,6 @@ struct cxl_dev_state {
 	resource_size_t component_reg_phys;
 	u64 serial;
 
-	struct xarray doe_mbs;
-
 	int (*mbox_send)(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd);
 };
 
drivers/cxl/pci.c +0 -49

67 lines not on this page · raw message

Signature · 1 line
2.39.1
#51 lore raw

PATCH v3 14/16 PCI/DOE: Make mailbox creation API private

In reply to Lukas Wunner

The PCI core has just been amended to create a pci_doe_mb struct for every DOE instance on device enumeration. CXL (the only in-tree DOE user so far) has been migrated to use those mailboxes instead of creating its own.

That leaves pcim_doe_create_mb() and pci_doe_for_each_off() without any callers, so drop them.

pci_doe_supports_prot() is now only used internally, so declare it static.

pci_doe_destroy_mb() is no longer used as callback for devm_add_action(), so refactor it to accept a struct pci_doe_mb pointer instead of a generic void pointer.

Because pci_doe_create_mb() is only called on device enumeration, i.e. before driver binding, the workqueue name never contains a driver name. So replace dev_driver_string() with dev_bus_name() when generating the workqueue name.

---
 .clang-format           |  1 -
 drivers/pci/doe.c       | 41 ++++-------------------------------------
 include/linux/pci-doe.h | 14 --------------
 3 files changed, 4 insertions(+), 52 deletions(-)
.clang-format +0 -1
index b62836419ea3..cb1c17c7fcc9 100644
@@ -520,7 +520,6 @@ ForEachMacros:
   - 'of_property_for_each_string'
   - 'of_property_for_each_u32'
   - 'pci_bus_for_each_resource'
-  - 'pci_doe_for_each_off'
   - 'pcl_for_each_chunk'
   - 'pcl_for_each_segment'
   - 'pcm_for_each_format'
drivers/pci/doe.c +4 -37

75 lines not on this page · raw message

include/linux/pci-doe.h +0 -14
index d6192ee0ac07..1f14aed4354b 100644
@@ -15,20 +15,6 @@
 
 struct pci_doe_mb;
 
-/**
- * pci_doe_for_each_off - Iterate each DOE capability
- * @pdev: struct pci_dev to iterate
- * @off: u16 of config space offset of each mailbox capability found
- */
-#define pci_doe_for_each_off(pdev, off) \
-	for (off = pci_find_next_ext_capability(pdev, off, \
-					PCI_EXT_CAP_ID_DOE); \
-		off > 0; \
-		off = pci_find_next_ext_capability(pdev, off, \
-					PCI_EXT_CAP_ID_DOE))
-
-struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset);
-bool pci_doe_supports_prot(struct pci_doe_mb *doe_mb, u16 vid, u8 type);
 struct pci_doe_mb *pci_find_doe_mailbox(struct pci_dev *pdev, u16 vendor,
 					u8 type);
 
Signature · 1 line
2.39.1
Reviewed-by on v3 14/16
#52 lore raw

Re: [PATCH v3 14/16] PCI/DOE: Make mailbox creation API private

Quoted 20 lines · Lukas Wunner

The PCI core has just been amended to create a pci_doe_mb struct for every DOE instance on device enumeration. CXL (the only in-tree DOE user so far) has been migrated to use those mailboxes instead of creating its own.

That leaves pcim_doe_create_mb() and pci_doe_for_each_off() without any callers, so drop them.

pci_doe_supports_prot() is now only used internally, so declare it static.

pci_doe_destroy_mb() is no longer used as callback for devm_add_action(), so refactor it to accept a struct pci_doe_mb pointer instead of a generic void pointer.

Because pci_doe_create_mb() is only called on device enumeration, i.e. before driver binding, the workqueue name never contains a driver name. So replace dev_driver_string() with dev_bus_name() when generating the workqueue name.

Quoted 4 lines
#53 lore raw

PATCH v3 15/16 PCI/DOE: Relax restrictions on request and response size

In reply to Lukas Wunner

An upcoming user of DOE is CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31).

It builds on SPDM (Security Protocol and Data Model):
https://www.dmtf.org/dsp/DSP0274

SPDM message sizes are not always a multiple of dwords. To transport them over DOE without using bounce buffers, allow sending requests and receiving responses whose final dword is only partially populated.

To be clear, PCIe r6.0 sec 6.30.1 specifies the Data Object Header 2 "Length" in dwords and pci_doe_send_req() and pci_doe_recv_resp() read/write dwords. So from a spec point of view, DOE is still specified in dwords and allowing non-dword request/response buffers is merely for the convenience of callers.

---
Changes v2 -> v3:
* Fix byte order of last dword on big endian arches (Ira)
* Explain in commit message and kerneldoc that arbitrary-sized payloads are not stipulated by the spec, but merely for convenience of the caller (Bjorn, Jonathan)
* Add code comment that "remainder" in pci_doe_recv_resp() signifies the number of data bytes in the last payload dword (Ira)

 drivers/pci/doe.c | 74 +++++++++++++++++++++++++++++++----------------
 1 file changed, 49 insertions(+), 25 deletions(-)
drivers/pci/doe.c +49 -25
index 15e5f9df1bba..14252f2fa955 100644
@@ -76,13 +76,6 @@ struct pci_doe_protocol {
  * @private: Private data for the consumer
  * @work: Used internally by the mailbox
  * @doe_mb: Used internally by the mailbox
- *
- * The payload sizes and rv are specified in bytes with the following
- * restrictions concerning the protocol.
- *
- *	1) The request_pl_sz must be a multiple of double words (4 bytes)
- *	2) The response_pl_sz must be >= a single double word (4 bytes)
- *	3) rv is returned as bytes but it will be a multiple of double words
  */
 struct pci_doe_task {
 	struct pci_doe_protocol prot;
@@ -153,7 +146,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 {
 	struct pci_dev *pdev = doe_mb->pdev;
 	int offset = doe_mb->cap_offset;
-	size_t length;
+	size_t length, remainder;
 	u32 val;
 	int i;
 
@@ -171,7 +164,7 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 		return -EIO;
 
 	/* Length is 2 DW of header + length of payload in DW */
-	length = 2 + task->request_pl_sz / sizeof(u32);
+	length = 2 + DIV_ROUND_UP(task->request_pl_sz, sizeof(u32));
 	if (length > PCI_DOE_MAX_LENGTH)
 		return -EIO;
 	if (length == PCI_DOE_MAX_LENGTH)
@@ -184,10 +177,21 @@ static int pci_doe_send_req(struct pci_doe_mb *doe_mb,
 	pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
 			       FIELD_PREP(PCI_DOE_DATA_OBJECT_HEADER_2_LENGTH,
 					  length));
+
+	/* Write payload */
 	for (i = 0; i < task->request_pl_sz / sizeof(u32); i++)
 		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE,
 				       le32_to_cpu(task->request_pl[i]));
 
+	/* Write last payload dword */
+	remainder = task->request_pl_sz % sizeof(u32);
+	if (remainder) {
+		val = 0;
+		memcpy(&val, &task->request_pl[i], remainder);
+		le32_to_cpus(&val);
+		pci_write_config_dword(pdev, offset + PCI_DOE_WRITE, val);
+	}
+
 	pci_doe_write_ctrl(doe_mb, PCI_DOE_CTRL_GO);
 
 	return 0;
@@ -207,11 +211,11 @@ static bool pci_doe_data_obj_ready(struct pci_doe_mb *doe_mb)
 
 static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *task)
 {
+	size_t length, payload_length, remainder, received;
 	struct pci_dev *pdev = doe_mb->pdev;
 	int offset = doe_mb->cap_offset;
-	size_t length, payload_length;
+	int i = 0;
 	u32 val;
-	int i;
 
 	/* Read the first dword to get the protocol */
 	pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
@@ -238,15 +242,38 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 
 	/* First 2 dwords have already been read */
 	length -= 2;
-	payload_length = min(length, task->response_pl_sz / sizeof(u32));
-	/* Read the rest of the response payload */
-	for (i = 0; i < payload_length; i++) {
+	received = task->response_pl_sz;
+	payload_length = DIV_ROUND_UP(task->response_pl_sz, sizeof(u32));
+	remainder = task->response_pl_sz % sizeof(u32);
+
+	/* remainder signifies number of data bytes in last payload dword */
+	if (!remainder)
+		remainder = sizeof(u32);
+
+	if (length < payload_length) {
+		received = length * sizeof(u32);
+		payload_length = length;
+		remainder = sizeof(u32);
+	}
+
+	if (payload_length) {
+		/* Read all payload dwords except the last */
+		for (; i < payload_length - 1; i++) {
+			pci_read_config_dword(pdev, offset + PCI_DOE_READ,
+					      &val);
+			task->response_pl[i] = cpu_to_le32(val);
+			pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		}
+
+		/* Read last payload dword */
 		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
-		task->response_pl[i] = cpu_to_le32(val);
+		cpu_to_le32s(&val);
+		memcpy(&task->response_pl[i], &val, remainder);
 		/* Prior to the last ack, ensure Data Object Ready */
-		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
+		if (!pci_doe_data_obj_ready(doe_mb))
 			return -EIO;
 		pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		i++;
 	}
 
 	/* Flush excess length */
@@ -260,7 +287,7 @@ static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas
 	if (FIELD_GET(PCI_DOE_STATUS_ERROR, val))
 		return -EIO;
 
-	return min(length, task->response_pl_sz / sizeof(u32)) * sizeof(u32);
+	return received;
 }
 
 static void signal_task_complete(struct pci_doe_task *task, int rv)
@@ -561,14 +588,6 @@ static int pci_doe_submit_task(struct pci_doe_mb *doe_mb,
 	if (!pci_doe_supports_prot(doe_mb, task->prot.vid, task->prot.type))
 		return -EINVAL;
 
-	/*
-	 * DOE requests must be a whole number of DW and the response needs to
-	 * be big enough for at least 1 DW
-	 */
-	if (task->request_pl_sz % sizeof(u32) ||
-	    task->response_pl_sz < sizeof(u32))
-		return -EINVAL;
-
 	if (test_bit(PCI_DOE_FLAG_DEAD, &doe_mb->flags))
 		return -EIO;
 
@@ -596,6 +615,11 @@ static int pci_doe_submit_task(struct pci_doe_mb *doe_mb,
  * without byte-swapping.  If payloads contain little-endian register values,
  * the caller is responsible for conversion with cpu_to_le32() / le32_to_cpu().
  *
+ * For convenience, arbitrary payload sizes are allowed even though PCIe r6.0
+ * sec 6.30.1 specifies the Data Object Header 2 "Length" in dwords.  The last
+ * (partial) dword is copied with byte granularity and padded with zeroes if
+ * necessary.  Callers are thus relieved from using dword-sized bounce buffers.
+ *
  * RETURNS: Length of received response or negative errno.
  * Received data in excess of @response_sz is discarded.
  * The length may be smaller than @response_sz and the caller
Signature · 1 line
2.39.1
Reviewed-by on v3 15/16
#54 lore raw

Re: [PATCH v3 15/16] PCI/DOE: Relax restrictions on request and response size

Quoted 31 lines · Lukas Wunner

An upcoming user of DOE is CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31).

It builds on SPDM (Security Protocol and Data Model):
https://www.dmtf.org/dsp/DSP0274

SPDM message sizes are not always a multiple of dwords. To transport them over DOE without using bounce buffers, allow sending requests and receiving responses whose final dword is only partially populated.

To be clear, PCIe r6.0 sec 6.30.1 specifies the Data Object Header 2 "Length" in dwords and pci_doe_send_req() and pci_doe_recv_resp() read/write dwords. So from a spec point of view, DOE is still specified in dwords and allowing non-dword request/response buffers is merely for the convenience of callers.

---
Changes v2 -> v3:
* Fix byte order of last dword on big endian arches (Ira)
* Explain in commit message and kerneldoc that arbitrary-sized payloads are not stipulated by the spec, but merely for convenience of the caller (Bjorn, Jonathan)
* Add code comment that "remainder" in pci_doe_recv_resp() signifies the number of data bytes in the last payload dword (Ira)

 drivers/pci/doe.c | 74 +++++++++++++++++++++++++++++++----------------
 1 file changed, 49 insertions(+), 25 deletions(-)

......

Quoted 5 lines
drivers/pci/doe.c +5 -0
+
+	if (payload_length) {
+		/* Read all payload dwords except the last */
+		for (; i < payload_length - 1; i++) {
+			pci_read_config_dword(pdev, offset + PCI_DOE_READ,

drivers/pci/doe.c static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas · line 271

Quoted 10 lines
drivers/pci/doe.c +8 -1
+					      &val);
+			task->response_pl[i] = cpu_to_le32(val);
+			pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		}
+
+		/* Read last payload dword */
 		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
-		task->response_pl[i] = cpu_to_le32(val);
+		cpu_to_le32s(&val);
+		memcpy(&task->response_pl[i], &val, remainder);

This "Read last payload dword" seems like making sense only when remainder != sizeof(u32). If remainder == sizeof(u32), it should be read in above reading loops.
But this implementation also looks good to me.

Quoted 9 lines
drivers/pci/doe.c +2 -1
 		/* Prior to the last ack, ensure Data Object Ready */
-		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
+		if (!pci_doe_data_obj_ready(doe_mb))
 			return -EIO;
 		pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		i++;
 	}
 
 	/* Flush excess length */
#55 lore raw

Re: [PATCH v3 15/16] PCI/DOE: Relax restrictions on request and response size

Quoted 10 lines · Ming
Quoted 9 lines · Lukas Wunner

An upcoming user of DOE is CMA (Component Measurement and Authentication, PCIe r6.0 sec 6.31).

It builds on SPDM (Security Protocol and Data Model):
https://www.dmtf.org/dsp/DSP0274

SPDM message sizes are not always a multiple of dwords. To transport them over DOE without using bounce buffers, allow sending requests and receiving responses whose final dword is only partially populated.

[...]

Quoted 10 lines
Quoted 10 lines
drivers/pci/doe.c +10 -0
+	if (payload_length) {
+		/* Read all payload dwords except the last */
+		for (; i < payload_length - 1; i++) {
+			pci_read_config_dword(pdev, offset + PCI_DOE_READ,
+					      &val);
+			task->response_pl[i] = cpu_to_le32(val);
+			pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		}
+
+		/* Read last payload dword */
Quoted 8 lines

drivers/pci/doe.c static int pci_doe_recv_resp(struct pci_doe_mb *doe_mb, struct pci_doe_task *tas · line 271

Quoted 4 lines
drivers/pci/doe.c +2 -1
 		pci_read_config_dword(pdev, offset + PCI_DOE_READ, &val);
-		task->response_pl[i] = cpu_to_le32(val);
+		cpu_to_le32s(&val);
+		memcpy(&task->response_pl[i], &val, remainder);

This "Read last payload dword" seems like making sense only when remainder != sizeof(u32). If remainder == sizeof(u32), it should be read in above reading loops.
But this implementation also looks good to me.

The last payload dword requires special handling anyway because of the Data Object Ready check (see quoted remainder of the function below).

Up until now that special handling is done with an if-clause that checks if the last loop iteration has been reached.

The support for non-dword responses adds more special handling of the last payload dword, which is why I decided to go the full distance and move handling of the last dword out of the loop.

Thanks,

Lukas

Quoted 9 lines
Quoted 9 lines
drivers/pci/doe.c +2 -1
 		/* Prior to the last ack, ensure Data Object Ready */
-		if (i == (payload_length - 1) && !pci_doe_data_obj_ready(doe_mb))
+		if (!pci_doe_data_obj_ready(doe_mb))
 			return -EIO;
 		pci_write_config_dword(pdev, offset + PCI_DOE_READ, 0);
+		i++;
 	}
 
 	/* Flush excess length */
#56 lore raw

PATCH v3 16/16 cxl/pci: Rightsize CDAT response allocation

In reply to Lukas Wunner

Jonathan notes that cxl_cdat_get_length() and cxl_cdat_read_table() allocate 32 dwords for the DOE response even though it may be smaller.

In the case of cxl_cdat_get_length(), only the second dword of the response is of interest (it contains the length). So reduce the allocation to 2 dwords and let DOE discard the remainder.

In the case of cxl_cdat_read_table(), a correctly sized allocation for the full CDAT already exists. Let DOE write each table entry directly into that allocation. There's a snag in that the table entry is preceded by a Table Access Response Header (1 dword). Save the last dword of the previous table entry, let DOE overwrite it with the header of the next entry and restore it afterwards.

The resulting CDAT is preceded by 4 unavoidable useless bytes. Increase the allocation size accordingly and skip these bytes when exposing CDAT in sysfs.

The buffer overflow check in cxl_cdat_read_table() becomes unnecessary because the remaining bytes in the allocation are tracked in "length", which is passed to DOE and limits how many bytes it writes to the allocation. Additionally, cxl_cdat_read_table() bails out if the DOE response is truncated due to insufficient space.

---
Changes v2 -> v3:
* Newly added patch in v3 on popular request (Jonathan)

 drivers/cxl/core/pci.c | 34 ++++++++++++++++++----------------
 drivers/cxl/cxl.h      |  3 ++-
 drivers/cxl/port.c     |  2 +-
 3 files changed, 21 insertions(+), 18 deletions(-)
drivers/cxl/core/pci.c +18 -16
index 1b954783b516..70097cc75302 100644
@@ -471,7 +471,7 @@ static int cxl_cdat_get_length(struct device *dev,
 			       size_t *length)
 {
 	__le32 request = CDAT_DOE_REQ(0);
-	__le32 response[32];
+	__le32 response[2];
 	int rc;
 
 	rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
@@ -495,28 +495,28 @@ static int cxl_cdat_read_table(struct device *dev,
 			       struct pci_doe_mb *cdat_doe,
 			       struct cxl_cdat *cdat)
 {
-	size_t length = cdat->length;
-	u32 *data = cdat->table;
+	size_t length = cdat->length + sizeof(u32);
+	__le32 *data = cdat->table;
 	int entry_handle = 0;
+	__le32 saved_dw = 0;
 
 	do {
 		__le32 request = CDAT_DOE_REQ(entry_handle);
 		struct cdat_entry_header *entry;
-		__le32 response[32];
 		size_t entry_dw;
 		int rc;
 
 		rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
 			     CXL_DOE_PROTOCOL_TABLE_ACCESS,
 			     &request, sizeof(request),
-			     &response, sizeof(response));
+			     data, length);
 		if (rc < 0) {
 			dev_err(dev, "DOE failed: %d", rc);
 			return rc;
 		}
 
 		/* 1 DW Table Access Response Header + CDAT entry */
-		entry = (struct cdat_entry_header *)(response + 1);
+		entry = (struct cdat_entry_header *)(data + 1);
 		if ((entry_handle == 0 &&
 		     rc != sizeof(u32) + sizeof(struct cdat_header)) ||
 		    (entry_handle > 0 &&
@@ -526,21 +526,22 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 le32_to_cpu(response[0]));
+					 le32_to_cpu(data[0]));
 		entry_dw = rc / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
-		entry_dw = min(length / sizeof(u32), entry_dw);
-		/* Prevent length < 1 DW from causing a buffer overflow */
-		if (entry_dw) {
-			memcpy(data, entry, entry_dw * sizeof(u32));
-			length -= entry_dw * sizeof(u32);
-			data += entry_dw;
-		}
+		/*
+		 * Table Access Response Header overwrote the last DW of
+		 * previous entry, so restore that DW
+		 */
+		*data = saved_dw;
+		length -= entry_dw * sizeof(u32);
+		data += entry_dw;
+		saved_dw = *data;
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
 	/* Length in CDAT header may exceed concatenation of CDAT entries */
-	cdat->length -= length;
+	cdat->length -= length - sizeof(u32);
 
 	return 0;
 }
@@ -576,7 +577,8 @@ void read_cdat_data(struct cxl_port *port)
 		return;
 	}
 
-	port->cdat.table = devm_kzalloc(dev, cdat_length, GFP_KERNEL);
+	port->cdat.table = devm_kzalloc(dev, cdat_length + sizeof(u32),
+					GFP_KERNEL);
 	if (!port->cdat.table)
 		return;
 
drivers/cxl/cxl.h +2 -1
index 1b1cf459ac77..78f5cae5134c 100644
@@ -494,7 +494,8 @@ struct cxl_pmem_region {
  * @component_reg_phys: component register capability base address (optional)
  * @dead: last ep has been removed, force port re-creation
  * @depth: How deep this port is relative to the root. depth 0 is the root.
- * @cdat: Cached CDAT data
+ * @cdat: Cached CDAT data (@table is preceded by 4 null bytes, these are not
+ *	  included in @length)
  * @cdat_available: Should a CDAT attribute be available in sysfs
  */
 struct cxl_port {
drivers/cxl/port.c +1 -1
index 5453771bf330..0705343ac5ca 100644
@@ -95,7 +95,7 @@ static ssize_t CDAT_read(struct file *filp, struct kobject *kobj,
 		return 0;
 
 	return memory_read_from_buffer(buf, count, &offset,
-				       port->cdat.table,
+				       port->cdat.table + sizeof(u32),
 				       port->cdat.length);
 }
 
Signature · 1 line
2.39.1
Reviewed-by on v3 16/16
#57 lore raw

Re: [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation

Quoted 7 lines · On Fri, 10 Feb 2023 21:25:16 +0100 Lukas Wunner <[email protected]> wrote:

Jonathan notes that cxl_cdat_get_length() and cxl_cdat_read_table() allocate 32 dwords for the DOE response even though it may be smaller.

In the case of cxl_cdat_get_length(), only the second dword of the response is of interest (it contains the length). So reduce the allocation to 2 dwords and let DOE discard the remainder.

Quoted 6 lines

In the case of cxl_cdat_read_table(), a correctly sized allocation for the full CDAT already exists. Let DOE write each table entry directly into that allocation. There's a snag in that the table entry is preceded by a Table Access Response Header (1 dword). Save the last dword of the previous table entry, let DOE overwrite it with the header of the next entry and restore it afterwards.

Marginally nasty, but looks like it works to me and avoid excessive allocations.

Quoted 11 lines

The resulting CDAT is preceded by 4 unavoidable useless bytes. Increase the allocation size accordingly and skip these bytes when exposing CDAT in sysfs.

The buffer overflow check in cxl_cdat_read_table() becomes unnecessary because the remaining bytes in the allocation are tracked in "length", which is passed to DOE and limits how many bytes it writes to the allocation. Additionally, cxl_cdat_read_table() bails out if the DOE response is truncated due to insufficient space.

Quoted 3 lines

Follow up comment on earlier one inline.

Quoted 23 lines

---
Changes v2 -> v3:
* Newly added patch in v3 on popular request (Jonathan)

 drivers/cxl/core/pci.c | 34 ++++++++++++++++++----------------
 drivers/cxl/cxl.h      |  3 ++-
 drivers/cxl/port.c     |  2 +-
 3 files changed, 21 insertions(+), 18 deletions(-)
drivers/cxl/core/pci.c +1 -1
index 1b954783b516..70097cc75302 100644
@@ -471,7 +471,7 @@ static int cxl_cdat_get_length(struct device *dev,
 			       size_t *length)
 {
 	__le32 request = CDAT_DOE_REQ(0);
-	__le32 response[32];
+	__le32 response[2];
 	int rc;
 
 	rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,

drivers/cxl/core/pci.c static int cxl_cdat_read_table(struct device *dev, · line 499

Quoted 7 lines
drivers/cxl/core/pci.c +2 -2
@@ -495,28 +495,28 @@ static int cxl_cdat_read_table(struct device *dev,
 			       struct pci_doe_mb *cdat_doe,
 			       struct cxl_cdat *cdat)
 {
-	size_t length = cdat->length;
-	u32 *data = cdat->table;
+	size_t length = cdat->length + sizeof(u32);
+	__le32 *data = cdat->table;

Ah. Makes my earlier comment on this type irrelevant.

Quoted 95 lines
drivers/cxl/core/pci.c +15 -13
 	int entry_handle = 0;
+	__le32 saved_dw = 0;
 
 	do {
 		__le32 request = CDAT_DOE_REQ(entry_handle);
 		struct cdat_entry_header *entry;
-		__le32 response[32];
 		size_t entry_dw;
 		int rc;
 
 		rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
 			     CXL_DOE_PROTOCOL_TABLE_ACCESS,
 			     &request, sizeof(request),
-			     &response, sizeof(response));
+			     data, length);
 		if (rc < 0) {
 			dev_err(dev, "DOE failed: %d", rc);
 			return rc;
 		}
 
 		/* 1 DW Table Access Response Header + CDAT entry */
-		entry = (struct cdat_entry_header *)(response + 1);
+		entry = (struct cdat_entry_header *)(data + 1);
 		if ((entry_handle == 0 &&
 		     rc != sizeof(u32) + sizeof(struct cdat_header)) ||
 		    (entry_handle > 0 &&
@@ -526,21 +526,22 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 le32_to_cpu(response[0]));
+					 le32_to_cpu(data[0]));
 		entry_dw = rc / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
-		entry_dw = min(length / sizeof(u32), entry_dw);
-		/* Prevent length < 1 DW from causing a buffer overflow */
-		if (entry_dw) {
-			memcpy(data, entry, entry_dw * sizeof(u32));
-			length -= entry_dw * sizeof(u32);
-			data += entry_dw;
-		}
+		/*
+		 * Table Access Response Header overwrote the last DW of
+		 * previous entry, so restore that DW
+		 */
+		*data = saved_dw;
+		length -= entry_dw * sizeof(u32);
+		data += entry_dw;
+		saved_dw = *data;
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
 	/* Length in CDAT header may exceed concatenation of CDAT entries */
-	cdat->length -= length;
+	cdat->length -= length - sizeof(u32);
 
 	return 0;
 }
@@ -576,7 +577,8 @@ void read_cdat_data(struct cxl_port *port)
 		return;
 	}
 
-	port->cdat.table = devm_kzalloc(dev, cdat_length, GFP_KERNEL);
+	port->cdat.table = devm_kzalloc(dev, cdat_length + sizeof(u32),
+					GFP_KERNEL);
 	if (!port->cdat.table)
 		return;
 
drivers/cxl/cxl.h +2 -1
index 1b1cf459ac77..78f5cae5134c 100644
@@ -494,7 +494,8 @@ struct cxl_pmem_region {
  * @component_reg_phys: component register capability base address (optional)
  * @dead: last ep has been removed, force port re-creation
  * @depth: How deep this port is relative to the root. depth 0 is the root.
- * @cdat: Cached CDAT data
+ * @cdat: Cached CDAT data (@table is preceded by 4 null bytes, these are not
+ *	  included in @length)
  * @cdat_available: Should a CDAT attribute be available in sysfs
  */
 struct cxl_port {
drivers/cxl/port.c +1 -1
index 5453771bf330..0705343ac5ca 100644
@@ -95,7 +95,7 @@ static ssize_t CDAT_read(struct file *filp, struct kobject *kobj,
 		return 0;
 
 	return memory_read_from_buffer(buf, count, &offset,
-				       port->cdat.table,
+				       port->cdat.table + sizeof(u32),
 				       port->cdat.length);
 }
 
#58 lore raw

Re: [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation

In reply to Lukas Wunner
Quoted 11 lines · Lukas Wunner

Jonathan notes that cxl_cdat_get_length() and cxl_cdat_read_table() allocate 32 dwords for the DOE response even though it may be smaller.

In the case of cxl_cdat_get_length(), only the second dword of the response is of interest (it contains the length). So reduce the allocation to 2 dwords and let DOE discard the remainder.

In the case of cxl_cdat_read_table(), a correctly sized allocation for the full CDAT already exists. Let DOE write each table entry directly into that allocation. There's a snag in that the table entry is preceded by a Table Access Response Header (1 dword).

Where is this 'Table Access Response Header' defined?

Ira

Quoted 145 lines

Save the last
dword of the previous table entry, let DOE overwrite it with the header of the next entry and restore it afterwards.

The resulting CDAT is preceded by 4 unavoidable useless bytes. Increase the allocation size accordingly and skip these bytes when exposing CDAT in sysfs.

The buffer overflow check in cxl_cdat_read_table() becomes unnecessary because the remaining bytes in the allocation are tracked in "length", which is passed to DOE and limits how many bytes it writes to the allocation. Additionally, cxl_cdat_read_table() bails out if the DOE response is truncated due to insufficient space.

---
Changes v2 -> v3:
* Newly added patch in v3 on popular request (Jonathan)

 drivers/cxl/core/pci.c | 34 ++++++++++++++++++----------------
 drivers/cxl/cxl.h      |  3 ++-
 drivers/cxl/port.c     |  2 +-
 3 files changed, 21 insertions(+), 18 deletions(-)
drivers/cxl/core/pci.c +18 -16
index 1b954783b516..70097cc75302 100644
@@ -471,7 +471,7 @@ static int cxl_cdat_get_length(struct device *dev,
 			       size_t *length)
 {
 	__le32 request = CDAT_DOE_REQ(0);
-	__le32 response[32];
+	__le32 response[2];
 	int rc;
 
 	rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
@@ -495,28 +495,28 @@ static int cxl_cdat_read_table(struct device *dev,
 			       struct pci_doe_mb *cdat_doe,
 			       struct cxl_cdat *cdat)
 {
-	size_t length = cdat->length;
-	u32 *data = cdat->table;
+	size_t length = cdat->length + sizeof(u32);
+	__le32 *data = cdat->table;
 	int entry_handle = 0;
+	__le32 saved_dw = 0;
 
 	do {
 		__le32 request = CDAT_DOE_REQ(entry_handle);
 		struct cdat_entry_header *entry;
-		__le32 response[32];
 		size_t entry_dw;
 		int rc;
 
 		rc = pci_doe(cdat_doe, PCI_DVSEC_VENDOR_ID_CXL,
 			     CXL_DOE_PROTOCOL_TABLE_ACCESS,
 			     &request, sizeof(request),
-			     &response, sizeof(response));
+			     data, length);
 		if (rc < 0) {
 			dev_err(dev, "DOE failed: %d", rc);
 			return rc;
 		}
 
 		/* 1 DW Table Access Response Header + CDAT entry */
-		entry = (struct cdat_entry_header *)(response + 1);
+		entry = (struct cdat_entry_header *)(data + 1);
 		if ((entry_handle == 0 &&
 		     rc != sizeof(u32) + sizeof(struct cdat_header)) ||
 		    (entry_handle > 0 &&
@@ -526,21 +526,22 @@ static int cxl_cdat_read_table(struct device *dev,
 
 		/* Get the CXL table access header entry handle */
 		entry_handle = FIELD_GET(CXL_DOE_TABLE_ACCESS_ENTRY_HANDLE,
-					 le32_to_cpu(response[0]));
+					 le32_to_cpu(data[0]));
 		entry_dw = rc / sizeof(u32);
 		/* Skip Header */
 		entry_dw -= 1;
-		entry_dw = min(length / sizeof(u32), entry_dw);
-		/* Prevent length < 1 DW from causing a buffer overflow */
-		if (entry_dw) {
-			memcpy(data, entry, entry_dw * sizeof(u32));
-			length -= entry_dw * sizeof(u32);
-			data += entry_dw;
-		}
+		/*
+		 * Table Access Response Header overwrote the last DW of
+		 * previous entry, so restore that DW
+		 */
+		*data = saved_dw;
+		length -= entry_dw * sizeof(u32);
+		data += entry_dw;
+		saved_dw = *data;
 	} while (entry_handle != CXL_DOE_TABLE_ACCESS_LAST_ENTRY);
 
 	/* Length in CDAT header may exceed concatenation of CDAT entries */
-	cdat->length -= length;
+	cdat->length -= length - sizeof(u32);
 
 	return 0;
 }
@@ -576,7 +577,8 @@ void read_cdat_data(struct cxl_port *port)
 		return;
 	}
 
-	port->cdat.table = devm_kzalloc(dev, cdat_length, GFP_KERNEL);
+	port->cdat.table = devm_kzalloc(dev, cdat_length + sizeof(u32),
+					GFP_KERNEL);
 	if (!port->cdat.table)
 		return;
 
drivers/cxl/cxl.h +2 -1
index 1b1cf459ac77..78f5cae5134c 100644
@@ -494,7 +494,8 @@ struct cxl_pmem_region {
  * @component_reg_phys: component register capability base address (optional)
  * @dead: last ep has been removed, force port re-creation
  * @depth: How deep this port is relative to the root. depth 0 is the root.
- * @cdat: Cached CDAT data
+ * @cdat: Cached CDAT data (@table is preceded by 4 null bytes, these are not
+ *	  included in @length)
  * @cdat_available: Should a CDAT attribute be available in sysfs
  */
 struct cxl_port {
drivers/cxl/port.c +1 -1
index 5453771bf330..0705343ac5ca 100644
@@ -95,7 +95,7 @@ static ssize_t CDAT_read(struct file *filp, struct kobject *kobj,
 		return 0;
 
 	return memory_read_from_buffer(buf, count, &offset,
-				       port->cdat.table,
+				       port->cdat.table + sizeof(u32),
 				       port->cdat.length);
 }
 
Signature · 1 line
2.39.1
#59 lore raw

Re: [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation

Quoted 8 lines · Ira Weiny
Quoted 7 lines · Lukas Wunner

Jonathan notes that cxl_cdat_get_length() and cxl_cdat_read_table() allocate 32 dwords for the DOE response even though it may be smaller.

In the case of cxl_cdat_get_length(), only the second dword of the response is of interest (it contains the length). So reduce the allocation to 2 dwords and let DOE discard the remainder.

Quoted 6 lines
Quoted 4 lines

In the case of cxl_cdat_read_table(), a correctly sized allocation for the full CDAT already exists. Let DOE write each table entry directly into that allocation. There's a snag in that the table entry is preceded by a Table Access Response Header (1 dword).

Where is this 'Table Access Response Header' defined?

CXL r3.0 table 8-14 (sec 8.1.11.1, page 399).

I'll amend the commit message with a reference to the spec.

Thanks,

Lukas

#60 lore raw

Re: [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation

In reply to Lukas Wunner
Quoted 24 lines · Lukas Wunner

Jonathan notes that cxl_cdat_get_length() and cxl_cdat_read_table() allocate 32 dwords for the DOE response even though it may be smaller.

In the case of cxl_cdat_get_length(), only the second dword of the response is of interest (it contains the length). So reduce the allocation to 2 dwords and let DOE discard the remainder.

In the case of cxl_cdat_read_table(), a correctly sized allocation for the full CDAT already exists. Let DOE write each table entry directly into that allocation. There's a snag in that the table entry is preceded by a Table Access Response Header (1 dword). Save the last dword of the previous table entry, let DOE overwrite it with the header of the next entry and restore it afterwards.

The resulting CDAT is preceded by 4 unavoidable useless bytes. Increase the allocation size accordingly and skip these bytes when exposing CDAT in sysfs.

The buffer overflow check in cxl_cdat_read_table() becomes unnecessary because the remaining bytes in the allocation are tracked in "length", which is passed to DOE and limits how many bytes it writes to the allocation. Additionally, cxl_cdat_read_table() bails out if the DOE response is truncated due to insufficient space.

Quoted 8 lines

---
Changes v2 -> v3:
* Newly added patch in v3 on popular request (Jonathan)

  drivers/cxl/core/pci.c | 34 ++++++++++++++++++----------------

Almost no change from this patchset applied cleanly to drivers/cxl/core/pci.c, what is the patchset based on? Thanks,

Signature · 1 line
Alexey
#61 lore raw

Re: [PATCH v3 16/16] cxl/pci: Rightsize CDAT response allocation

Quoted 2 lines · Alexey Kardashevskiy

Almost no change from this patchset applied cleanly to drivers/cxl/core/pci.c, what is the patchset based on? Thanks,

This series is based on pci/next, i.e. v6.2-rc1 plus PCI changes for v6.3:

https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git/log/?h=next

It should apply cleanly to cxl/next as well, save for a trivial merge conflict in patch [13/16] due to a context change. Here's a branch containing v3 of this series based on cxl/next:

https://github.com/l1k/linux/commits/doe_rework_v3_cxl_next

That said, you're probably better off just using the development branch for CMA/SPDM, it incorporates all the changes requested during review of v3 and will form the basis for v4 (slated for submission in a few days):

https://github.com/l1k/linux/commits/doe

Thanks,

Lukas