pau: Add support for OpenCAPI Persistent Memory devices.

Lowest Point of Coherency (LPC) memory allows the host to access memory on
an OpenCAPI device.

When the P10 chip accesses memory addresses on the AFU, the Real Address
on the PowerBus must hit a BAR in the PAU such as GPU-Memory BAR. The BAR
defines the range of Real Addresses that represent AFU memory.

The two existing OPAL calls, OPAL_NPU_MEM_ALLOC and OPAL_NPU_MEM_RELEASE
are used to manage the AFU momory.

Signed-off-by: Christophe Lombard <clombard@linux.vnet.ibm.com>
Signed-off-by: Vasant Hegde <hegdevasant@linux.vnet.ibm.com>

7 files changed, 152 insertions, 16 deletions

diff --git a/hw/npu-opal.c b/hw/npu-opal.c
index 2e455dc..1b66857 100644
--- a/hw/npu-opal.c
+++ b/hw/npu-opal.c
@@ -239,3 +239,38 @@ static int64_t opal_npu_tl_set(uint64_t phb_id, uint32_t bdfn,
 	return OPAL_PARAMETER;
 }
 opal_call(OPAL_NPU_TL_SET, opal_npu_tl_set, 5);
+
+static int64_t opal_npu_mem_alloc(uint64_t phb_id, uint32_t bdfn,
+				  uint64_t size, uint64_t *bar)
+{
+	struct phb *phb = pci_get_phb(phb_id);
+
+	if (!phb)
+		return OPAL_PARAMETER;
+
+	if (phb->phb_type == phb_type_npu_v2_opencapi)
+		return npu2_opencapi_mem_alloc(phb, bdfn, size, bar);
+
+	if (phb->phb_type == phb_type_pau_opencapi)
+		return pau_opencapi_mem_alloc(phb, bdfn, size, bar);
+
+	return OPAL_PARAMETER;
+}
+opal_call(OPAL_NPU_MEM_ALLOC, opal_npu_mem_alloc, 4);
+
+static int64_t opal_npu_mem_release(uint64_t phb_id, uint32_t bdfn)
+{
+	struct phb *phb = pci_get_phb(phb_id);;
+
+	if (!phb)
+		return OPAL_PARAMETER;
+
+	if (phb->phb_type == phb_type_npu_v2_opencapi)
+		return npu2_opencapi_mem_release(phb, bdfn);
+
+	if (phb->phb_type == phb_type_pau_opencapi)
+		return pau_opencapi_mem_release(phb, bdfn);
+
+	return OPAL_PARAMETER;
+}
+opal_call(OPAL_NPU_MEM_RELEASE, opal_npu_mem_release, 2);
diff --git a/hw/npu2-opencapi.c b/hw/npu2-opencapi.c
index 272f924..8e7bcca 100644
--- a/hw/npu2-opencapi.c
+++ b/hw/npu2-opencapi.c
@@ -2295,18 +2295,13 @@ out:
 	return rc;
 }
 
-static int64_t opal_npu_mem_alloc(uint64_t phb_id, uint32_t __unused bdfn,
-				  uint64_t size, __be64 *__bar)
+int64_t npu2_opencapi_mem_alloc(struct phb *phb, uint32_t __unused bdfn,
+				uint64_t size, uint64_t *__bar)
 {
-	struct phb *phb = pci_get_phb(phb_id);
 	struct npu2_dev *dev;
 	uint64_t bar;
 	int64_t rc;
 
-
-	if (!phb || phb->phb_type != phb_type_npu_v2_opencapi)
-		return OPAL_PARAMETER;
-
 	dev = phb_to_npu2_dev_ocapi(phb);
 	if (!dev)
 		return OPAL_PARAMETER;
@@ -2320,21 +2315,14 @@ static int64_t opal_npu_mem_alloc(uint64_t phb_id, uint32_t __unused bdfn,
 
 	return rc;
 }
-opal_call(OPAL_NPU_MEM_ALLOC, opal_npu_mem_alloc, 4);
 
-static int64_t opal_npu_mem_release(uint64_t phb_id, uint32_t __unused bdfn)
+int64_t npu2_opencapi_mem_release(struct phb *phb, uint32_t __unused bdfn)
 {
-	struct phb *phb = pci_get_phb(phb_id);
 	struct npu2_dev *dev;
 
-
-	if (!phb || phb->phb_type != phb_type_npu_v2_opencapi)
-		return OPAL_PARAMETER;
-
 	dev = phb_to_npu2_dev_ocapi(phb);
 	if (!dev)
 		return OPAL_PARAMETER;
 
 	return release_mem_bar(dev);
 }
-opal_call(OPAL_NPU_MEM_RELEASE, opal_npu_mem_release, 2);
diff --git a/hw/pau.c b/hw/pau.c
index 9f6a709..a34d6da 100644
--- a/hw/pau.c
+++ b/hw/pau.c
@@ -442,6 +442,99 @@ int64_t pau_opencapi_tl_set(struct phb *phb, uint32_t __unused bdfn,
 	return OPAL_SUCCESS;
 }
 
+static int64_t pau_opencapi_afu_memory_bars(struct pau_dev *dev,
+					    uint64_t size,
+					    uint64_t *bar)
+{
+	struct pau *pau = dev->pau;
+	uint64_t addr, psize;
+	uint64_t reg, val;
+
+	PAUDEVDBG(dev, "Setup AFU Memory BARs\n");
+
+	if (dev->memory_bar.enable) {
+		PAUDEVERR(dev, "AFU memory allocation failed - BAR already in use\n");
+		return OPAL_RESOURCE;
+	}
+
+	phys_map_get(pau->chip_id, OCAPI_MEM,
+		     dev->index,
+		     &addr, &psize);
+
+	if (size > psize) {
+		PAUDEVERR(dev, "Invalid AFU memory BAR allocation size "
+			       "requested: 0x%llx bytes (limit 0x%llx)\n",
+			  size, psize);
+		return OPAL_PARAMETER;
+	}
+
+	if (size < (1 << 30))
+		size = 1 << 30;
+
+	dev->memory_bar.enable = true;
+	dev->memory_bar.addr = addr;
+	dev->memory_bar.size = size;
+
+	reg = PAU_GPU_MEM_BAR(dev->index);
+	val = PAU_GPU_MEM_BAR_ENABLE |
+	      PAU_GPU_MEM_BAR_POISON;
+	val = SETFIELD(PAU_GPU_MEM_BAR_ADDR, val, addr >> 30);
+	if (!is_pow2(size))
+		size = 1ull << (ilog2(size) + 1);
+
+	size = (size >> 30) - 1;
+	val = SETFIELD(PAU_GPU_MEM_BAR_SIZE, val, size);
+	pau_write(pau, reg, val);
+
+	reg = PAU_CTL_MISC_GPU_MEM_BAR(dev->index);
+	pau_write(pau, reg, val);
+
+	reg = PAU_XSL_GPU_MEM_BAR(dev->index);
+	pau_write(pau, reg, val);
+
+	*bar = addr;
+	return OPAL_SUCCESS;
+}
+
+int64_t pau_opencapi_mem_alloc(struct phb *phb, uint32_t __unused bdfn,
+			       uint64_t size, uint64_t *bar)
+{
+	struct pau_dev *dev = pau_phb_to_opencapi_dev(phb);
+	int64_t rc;
+
+	if (!dev)
+		return OPAL_PARAMETER;
+
+	if (!opal_addr_valid(bar))
+		return OPAL_PARAMETER;
+
+	lock(&dev->pau->lock);
+	rc = pau_opencapi_afu_memory_bars(dev, size, bar);
+
+	unlock(&dev->pau->lock);
+	return rc;
+}
+
+int64_t pau_opencapi_mem_release(struct phb *phb, uint32_t __unused bdfn)
+{
+	struct pau_dev *dev = pau_phb_to_opencapi_dev(phb);
+
+	if (!dev)
+		return OPAL_PARAMETER;
+
+	lock(&dev->pau->lock);
+	pau_write(dev->pau, PAU_GPU_MEM_BAR(dev->index), 0ull);
+	pau_write(dev->pau, PAU_CTL_MISC_GPU_MEM_BAR(dev->index), 0ull);
+	pau_write(dev->pau, PAU_XSL_GPU_MEM_BAR(dev->index), 0ull);
+
+	dev->memory_bar.enable = false;
+	dev->memory_bar.addr = 0ull;
+	dev->memory_bar.size = 0ull;
+	unlock(&dev->pau->lock);
+
+	return OPAL_SUCCESS;
+}
+
 #define CQ_CTL_STATUS_TIMEOUT  10 /* milliseconds */
 
 static int pau_opencapi_set_fence_control(struct pau_dev *dev,
diff --git a/hw/phys-map.c b/hw/phys-map.c
index 711d626..cfce4ab 100644
--- a/hw/phys-map.c
+++ b/hw/phys-map.c
@@ -30,7 +30,12 @@ static const struct phys_map_entry phys_map_table_p10[] = {
 	/* System memory upto 4TB minus GPU memory */
 	{ SYSTEM_MEM,      0, 0x0000000000000000ull, 0x0000034000000000ull },
 
-	/* TODO: Figure out GPU memory */
+	/* Configure OpenCapi memory as before with the old chips.
+	 * Keep in mind that we would need to rework this by allocating
+	 * a new topology id.
+	 */
+	{ OCAPI_MEM,	   0, 0x0002000000000000ull, 0x0000040000000000ull },
+	{ OCAPI_MEM,	   1, 0x0002040000000000ull, 0x0000040000000000ull },
 
 	/* 0 TB offset @ MMIO 0x0006000000000000ull */
 	{ PHB5_64BIT_MMIO, 0, 0x0006000000000000ull, 0x0000004000000000ull },
diff --git a/include/npu2.h b/include/npu2.h
index a12bf98..b302108 100644
--- a/include/npu2.h
+++ b/include/npu2.h
@@ -277,5 +277,8 @@ int64_t npu2_opencapi_spa_clear_cache(struct phb *phb, uint32_t __unused bdfn,
 				      uint64_t PE_handle);
 int64_t npu2_opencapi_tl_set(struct phb *phb, uint32_t __unused bdfn,
 		    long capabilities, char *rate);
+int64_t npu2_opencapi_mem_alloc(struct phb *phb, uint32_t __unused bdfn,
+				uint64_t size, uint64_t *bar);
+int64_t npu2_opencapi_mem_release(struct phb *phb, uint32_t __unused bdfn);
 
 #endif /* __NPU2_H */
diff --git a/include/pau-regs.h b/include/pau-regs.h
index da83ad4..45c3603 100644
--- a/include/pau-regs.h
+++ b/include/pau-regs.h
@@ -64,6 +64,12 @@
 #define   PAU_SNP_MISC_CFG0_ENABLE_PBUS		PPC_BIT(2)
 #define   PAU_SNP_MISC_CFG0_OCAPI_MODE		PPC_BITMASK(32, 36)
 #define   PAU_SNP_MISC_CFG0_OCAPI_C2		PPC_BITMASK(45, 49)
+#define PAU_GPU_MEM_BAR(brk)			(PAU_BLOCK_CQ_SM(0) + 0x190 + (brk) * 8)
+#define   PAU_GPU_MEM_BAR_ENABLE		PPC_BIT(0)
+#define   PAU_GPU_MEM_BAR_ADDR_MASK		PPC_BITMASK(1, 35)
+#define     PAU_GPU_MEM_BAR_ADDR		PPC_BITMASK(1, 21)
+#define     PAU_GPU_MEM_BAR_SIZE		PPC_BITMASK(22, 35)
+#define   PAU_GPU_MEM_BAR_POISON		PPC_BIT(45)
 #define PAU_NTL_BAR(brk)			(PAU_BLOCK_CQ_SM(0) + 0x1b8 + (brk) * 8)
 #define   PAU_NTL_BAR_ENABLE			PPC_BIT(0)
 #define   PAU_NTL_BAR_ADDR			PPC_BITMASK(3, 35)
@@ -88,6 +94,7 @@
 #define   PAU_CTL_MISC_CFG2_OCAPI_MEM_OS_BIT	PPC_BITMASK(25, 29)
 #define PAU_CTL_MISC_STATUS(brk)		(PAU_BLOCK_CQ_CTL + 0x060 + (brk) * 8)
 #define   PAU_CTL_MISC_STATUS_AM_FENCED(brk)	(PPC_BITMASK(41, 42) << ((brk)*32))
+#define PAU_CTL_MISC_GPU_MEM_BAR(brk)		(PAU_BLOCK_CQ_CTL + 0x070 + (brk) * 8)
 #define PAU_CTL_MISC_MMIOPA_CONFIG(brk)		(PAU_BLOCK_CQ_CTL + 0x098 + (brk) * 8)
 #define   PAU_CTL_MISC_MMIOPA_CONFIG_BAR_ADDR	PPC_BITMASK(1, 35)
 #define   PAU_CTL_MISC_MMIOPA_CONFIG_BAR_SIZE	PPC_BITMASK(39, 43)
@@ -159,6 +166,7 @@
 /* XSL block registers */
 #define PAU_XSL_OSL_SPAP_AN(brk)		(PAU_BLOCK_XSL + 0x000 + (brk) * 8)
 #define   PAU_XSL_OSL_SPAP_AN_EN		PPC_BIT(63)
+#define PAU_XSL_GPU_MEM_BAR(brk)		(PAU_BLOCK_XSL + 0x0D0 + (brk) * 8)
 #define PAU_XSL_WRAP_CFG			(PAU_BLOCK_XSL + 0x100)
 #define   PAU_XSL_WRAP_CFG_CLOCK_ENABLE		PPC_BIT(0)
 #define PAU_XSL_OSL_XLATE_CFG(brk)		(PAU_BLOCK_XSL + 0x040 + (brk) * 8)
diff --git a/include/pau.h b/include/pau.h
index a70058f..4a6087c 100644
--- a/include/pau.h
+++ b/include/pau.h
@@ -45,6 +45,7 @@ struct pau_dev {
 
 	struct pau_bar		ntl_bar;
 	struct pau_bar		genid_bar;
+	struct pau_bar		memory_bar;
 
 	/* Associated I2C information */
 	uint8_t			i2c_bus_id;
@@ -209,6 +210,9 @@ int64_t pau_opencapi_spa_clear_cache(struct phb *phb,
 				     uint64_t PE_handle);
 int64_t pau_opencapi_tl_set(struct phb *phb, uint32_t __unused bdfn,
 			    long capabilities, char *rate_buf);
+int64_t pau_opencapi_mem_alloc(struct phb *phb, uint32_t __unused bdfn,
+			       uint64_t size, uint64_t *bar);
+int64_t pau_opencapi_mem_release(struct phb *phb, uint32_t __unused bdfn);
 
 /* PHY */
 int pau_dev_phy_reset(struct pau_dev *dev);