soc: qualcomm: Add Shared Memory Manager driver
authorRamon Fried <ramon.fried@gmail.com>
Sun, 1 Jul 2018 23:57:56 +0000 (02:57 +0300)
committerTom Rini <trini@konsulko.com>
Thu, 19 Jul 2018 20:31:37 +0000 (16:31 -0400)
The Shared Memory Manager driver implements an interface for allocating
and accessing items in the memory area shared among all of the
processors in a Qualcomm platform.

Adapted from the Linux driver (4.17)

Changes from the original Linux driver:
* Removed HW spinlock mechanism, which is irrelevant
in U-boot particualar use case, which is just reading from the smem.
* Adapted from Linux driver model to U-Boot's.

Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Ramon Fried <ramon.fried@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
MAINTAINERS
arch/arm/Kconfig
drivers/Kconfig
drivers/smem/Kconfig
drivers/smem/Makefile
drivers/smem/msm_smem.c [new file with mode: 0644]

index 91e4ad7c75a1b909282f1032c84c3a538421f77c..a2293b7c88d4513f9ddccee4f21f70f3473437bd 100644 (file)
@@ -206,6 +206,7 @@ ARM SNAPDRAGON
 M:     Ramon Fried <ramon.fried@gmail.com>
 S:     Maintained
 F:     arch/arm/mach-snapdragon/
+F:     drivers/smem/msm_smem.c
 
 ARM STI
 M:     Patrice Chotard <patrice.chotard@st.com>
index 64d58a624182049d96dd76f6b55f874316ff2c4d..376851ef7aa94e42eb2b5c7fcc25d1275f1ce7a4 100644 (file)
@@ -754,6 +754,8 @@ config ARCH_SNAPDRAGON
        select SPMI
        select OF_CONTROL
        select OF_SEPARATE
+       select SMEM
+       select MSM_SMEM
 
 config ARCH_SOCFPGA
        bool "Altera SOCFPGA family"
index 9e21b28750c3dbbcd8d948a9aa30e9de125040c7..c72abf893297094cc7a143a613ad914d535c99af 100644 (file)
@@ -84,6 +84,8 @@ source "drivers/scsi/Kconfig"
 
 source "drivers/serial/Kconfig"
 
+source "drivers/smem/Kconfig"
+
 source "drivers/sound/Kconfig"
 
 source "drivers/spi/Kconfig"
index 64337a8b8ea3e50f37d88b87f6ee12fad7898cf7..6cf5a4e61e345309e69c9d5cfad56e5fdaed9876 100644 (file)
@@ -1,2 +1,15 @@
 menuconfig SMEM
        bool  "SMEM (Shared Memory mamanger) support"
+
+if SMEM
+
+config MSM_SMEM
+    bool "Qualcomm Shared Memory Manager (SMEM)"
+    depends on DM
+    depends on ARCH_SNAPDRAGON
+    help
+      Enable support for the Qualcomm Shared Memory Manager.
+      The driver provides an interface to items in a heap shared among all
+      processors in a Qualcomm platform.
+
+endif # menu "SMEM Support"
index ca55c4512de7072db9a3837345bbb52b25743727..605b8fc290a3bb81a933204dfe7f27526bf9084e 100644 (file)
@@ -3,3 +3,4 @@
 # Makefile for the U-Boot SMEM interface drivers
 
 obj-$(CONFIG_SMEM) += smem-uclass.o
+obj-$(CONFIG_MSM_SMEM) += msm_smem.o
diff --git a/drivers/smem/msm_smem.c b/drivers/smem/msm_smem.c
new file mode 100644 (file)
index 0000000..9fa653a
--- /dev/null
@@ -0,0 +1,932 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (c) 2015, Sony Mobile Communications AB.
+ * Copyright (c) 2012-2013, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2018, Ramon Fried <ramon.fried@gmail.com>
+ */
+
+#include <common.h>
+#include <errno.h>
+#include <dm.h>
+#include <dm/of_access.h>
+#include <dm/of_addr.h>
+#include <asm/io.h>
+#include <linux/ioport.h>
+#include <linux/io.h>
+#include <smem.h>
+
+DECLARE_GLOBAL_DATA_PTR;
+
+/*
+ * The Qualcomm shared memory system is an allocate-only heap structure that
+ * consists of one of more memory areas that can be accessed by the processors
+ * in the SoC.
+ *
+ * All systems contains a global heap, accessible by all processors in the SoC,
+ * with a table of contents data structure (@smem_header) at the beginning of
+ * the main shared memory block.
+ *
+ * The global header contains meta data for allocations as well as a fixed list
+ * of 512 entries (@smem_global_entry) that can be initialized to reference
+ * parts of the shared memory space.
+ *
+ *
+ * In addition to this global heap, a set of "private" heaps can be set up at
+ * boot time with access restrictions so that only certain processor pairs can
+ * access the data.
+ *
+ * These partitions are referenced from an optional partition table
+ * (@smem_ptable), that is found 4kB from the end of the main smem region. The
+ * partition table entries (@smem_ptable_entry) lists the involved processors
+ * (or hosts) and their location in the main shared memory region.
+ *
+ * Each partition starts with a header (@smem_partition_header) that identifies
+ * the partition and holds properties for the two internal memory regions. The
+ * two regions are cached and non-cached memory respectively. Each region
+ * contain a link list of allocation headers (@smem_private_entry) followed by
+ * their data.
+ *
+ * Items in the non-cached region are allocated from the start of the partition
+ * while items in the cached region are allocated from the end. The free area
+ * is hence the region between the cached and non-cached offsets. The header of
+ * cached items comes after the data.
+ *
+ * Version 12 (SMEM_GLOBAL_PART_VERSION) changes the item alloc/get procedure
+ * for the global heap. A new global partition is created from the global heap
+ * region with partition type (SMEM_GLOBAL_HOST) and the max smem item count is
+ * set by the bootloader.
+ *
+ */
+
+/*
+ * The version member of the smem header contains an array of versions for the
+ * various software components in the SoC. We verify that the boot loader
+ * version is a valid version as a sanity check.
+ */
+#define SMEM_MASTER_SBL_VERSION_INDEX  7
+#define SMEM_GLOBAL_HEAP_VERSION       11
+#define SMEM_GLOBAL_PART_VERSION       12
+
+/*
+ * The first 8 items are only to be allocated by the boot loader while
+ * initializing the heap.
+ */
+#define SMEM_ITEM_LAST_FIXED   8
+
+/* Highest accepted item number, for both global and private heaps */
+#define SMEM_ITEM_COUNT                512
+
+/* Processor/host identifier for the application processor */
+#define SMEM_HOST_APPS         0
+
+/* Processor/host identifier for the global partition */
+#define SMEM_GLOBAL_HOST       0xfffe
+
+/* Max number of processors/hosts in a system */
+#define SMEM_HOST_COUNT                10
+
+/**
+ * struct smem_proc_comm - proc_comm communication struct (legacy)
+ * @command:   current command to be executed
+ * @status:    status of the currently requested command
+ * @params:    parameters to the command
+ */
+struct smem_proc_comm {
+       __le32 command;
+       __le32 status;
+       __le32 params[2];
+};
+
+/**
+ * struct smem_global_entry - entry to reference smem items on the heap
+ * @allocated: boolean to indicate if this entry is used
+ * @offset:    offset to the allocated space
+ * @size:      size of the allocated space, 8 byte aligned
+ * @aux_base:  base address for the memory region used by this unit, or 0 for
+ *             the default region. bits 0,1 are reserved
+ */
+struct smem_global_entry {
+       __le32 allocated;
+       __le32 offset;
+       __le32 size;
+       __le32 aux_base; /* bits 1:0 reserved */
+};
+#define AUX_BASE_MASK          0xfffffffc
+
+/**
+ * struct smem_header - header found in beginning of primary smem region
+ * @proc_comm:         proc_comm communication interface (legacy)
+ * @version:           array of versions for the various subsystems
+ * @initialized:       boolean to indicate that smem is initialized
+ * @free_offset:       index of the first unallocated byte in smem
+ * @available:         number of bytes available for allocation
+ * @reserved:          reserved field, must be 0
+ * toc:                        array of references to items
+ */
+struct smem_header {
+       struct smem_proc_comm proc_comm[4];
+       __le32 version[32];
+       __le32 initialized;
+       __le32 free_offset;
+       __le32 available;
+       __le32 reserved;
+       struct smem_global_entry toc[SMEM_ITEM_COUNT];
+};
+
+/**
+ * struct smem_ptable_entry - one entry in the @smem_ptable list
+ * @offset:    offset, within the main shared memory region, of the partition
+ * @size:      size of the partition
+ * @flags:     flags for the partition (currently unused)
+ * @host0:     first processor/host with access to this partition
+ * @host1:     second processor/host with access to this partition
+ * @cacheline: alignment for "cached" entries
+ * @reserved:  reserved entries for later use
+ */
+struct smem_ptable_entry {
+       __le32 offset;
+       __le32 size;
+       __le32 flags;
+       __le16 host0;
+       __le16 host1;
+       __le32 cacheline;
+       __le32 reserved[7];
+};
+
+/**
+ * struct smem_ptable - partition table for the private partitions
+ * @magic:     magic number, must be SMEM_PTABLE_MAGIC
+ * @version:   version of the partition table
+ * @num_entries: number of partitions in the table
+ * @reserved:  for now reserved entries
+ * @entry:     list of @smem_ptable_entry for the @num_entries partitions
+ */
+struct smem_ptable {
+       u8 magic[4];
+       __le32 version;
+       __le32 num_entries;
+       __le32 reserved[5];
+       struct smem_ptable_entry entry[];
+};
+
+static const u8 SMEM_PTABLE_MAGIC[] = { 0x24, 0x54, 0x4f, 0x43 }; /* "$TOC" */
+
+/**
+ * struct smem_partition_header - header of the partitions
+ * @magic:     magic number, must be SMEM_PART_MAGIC
+ * @host0:     first processor/host with access to this partition
+ * @host1:     second processor/host with access to this partition
+ * @size:      size of the partition
+ * @offset_free_uncached: offset to the first free byte of uncached memory in
+ *             this partition
+ * @offset_free_cached: offset to the first free byte of cached memory in this
+ *             partition
+ * @reserved:  for now reserved entries
+ */
+struct smem_partition_header {
+       u8 magic[4];
+       __le16 host0;
+       __le16 host1;
+       __le32 size;
+       __le32 offset_free_uncached;
+       __le32 offset_free_cached;
+       __le32 reserved[3];
+};
+
+static const u8 SMEM_PART_MAGIC[] = { 0x24, 0x50, 0x52, 0x54 };
+
+/**
+ * struct smem_private_entry - header of each item in the private partition
+ * @canary:    magic number, must be SMEM_PRIVATE_CANARY
+ * @item:      identifying number of the smem item
+ * @size:      size of the data, including padding bytes
+ * @padding_data: number of bytes of padding of data
+ * @padding_hdr: number of bytes of padding between the header and the data
+ * @reserved:  for now reserved entry
+ */
+struct smem_private_entry {
+       u16 canary; /* bytes are the same so no swapping needed */
+       __le16 item;
+       __le32 size; /* includes padding bytes */
+       __le16 padding_data;
+       __le16 padding_hdr;
+       __le32 reserved;
+};
+#define SMEM_PRIVATE_CANARY    0xa5a5
+
+/**
+ * struct smem_info - smem region info located after the table of contents
+ * @magic:     magic number, must be SMEM_INFO_MAGIC
+ * @size:      size of the smem region
+ * @base_addr: base address of the smem region
+ * @reserved:  for now reserved entry
+ * @num_items: highest accepted item number
+ */
+struct smem_info {
+       u8 magic[4];
+       __le32 size;
+       __le32 base_addr;
+       __le32 reserved;
+       __le16 num_items;
+};
+
+static const u8 SMEM_INFO_MAGIC[] = { 0x53, 0x49, 0x49, 0x49 }; /* SIII */
+
+/**
+ * struct smem_region - representation of a chunk of memory used for smem
+ * @aux_base:  identifier of aux_mem base
+ * @virt_base: virtual base address of memory with this aux_mem identifier
+ * @size:      size of the memory region
+ */
+struct smem_region {
+       u32 aux_base;
+       void __iomem *virt_base;
+       size_t size;
+};
+
+/**
+ * struct qcom_smem - device data for the smem device
+ * @dev:       device pointer
+ * @global_partition:  pointer to global partition when in use
+ * @global_cacheline:  cacheline size for global partition
+ * @partitions:        list of pointers to partitions affecting the current
+ *             processor/host
+ * @cacheline: list of cacheline sizes for each host
+ * @item_count: max accepted item number
+ * @num_regions: number of @regions
+ * @regions:   list of the memory regions defining the shared memory
+ */
+struct qcom_smem {
+       struct udevice *dev;
+
+       struct smem_partition_header *global_partition;
+       size_t global_cacheline;
+       struct smem_partition_header *partitions[SMEM_HOST_COUNT];
+       size_t cacheline[SMEM_HOST_COUNT];
+       u32 item_count;
+
+       unsigned int num_regions;
+       struct smem_region regions[0];
+};
+
+static struct smem_private_entry *
+phdr_to_last_uncached_entry(struct smem_partition_header *phdr)
+{
+       void *p = phdr;
+
+       return p + le32_to_cpu(phdr->offset_free_uncached);
+}
+
+static void *phdr_to_first_cached_entry(struct smem_partition_header *phdr,
+                                       size_t cacheline)
+{
+       void *p = phdr;
+
+       return p + le32_to_cpu(phdr->size) - ALIGN(sizeof(*phdr), cacheline);
+}
+
+static void *phdr_to_last_cached_entry(struct smem_partition_header *phdr)
+{
+       void *p = phdr;
+
+       return p + le32_to_cpu(phdr->offset_free_cached);
+}
+
+static struct smem_private_entry *
+phdr_to_first_uncached_entry(struct smem_partition_header *phdr)
+{
+       void *p = phdr;
+
+       return p + sizeof(*phdr);
+}
+
+static struct smem_private_entry *
+uncached_entry_next(struct smem_private_entry *e)
+{
+       void *p = e;
+
+       return p + sizeof(*e) + le16_to_cpu(e->padding_hdr) +
+              le32_to_cpu(e->size);
+}
+
+static struct smem_private_entry *
+cached_entry_next(struct smem_private_entry *e, size_t cacheline)
+{
+       void *p = e;
+
+       return p - le32_to_cpu(e->size) - ALIGN(sizeof(*e), cacheline);
+}
+
+static void *uncached_entry_to_item(struct smem_private_entry *e)
+{
+       void *p = e;
+
+       return p + sizeof(*e) + le16_to_cpu(e->padding_hdr);
+}
+
+static void *cached_entry_to_item(struct smem_private_entry *e)
+{
+       void *p = e;
+
+       return p - le32_to_cpu(e->size);
+}
+
+/* Pointer to the one and only smem handle */
+static struct qcom_smem *__smem;
+
+static int qcom_smem_alloc_private(struct qcom_smem *smem,
+                                  struct smem_partition_header *phdr,
+                                  unsigned int item,
+                                  size_t size)
+{
+       struct smem_private_entry *hdr, *end;
+       size_t alloc_size;
+       void *cached;
+
+       hdr = phdr_to_first_uncached_entry(phdr);
+       end = phdr_to_last_uncached_entry(phdr);
+       cached = phdr_to_last_cached_entry(phdr);
+
+       while (hdr < end) {
+               if (hdr->canary != SMEM_PRIVATE_CANARY) {
+                       dev_err(smem->dev,
+                               "Found invalid canary in hosts %d:%d partition\n",
+                               phdr->host0, phdr->host1);
+                       return -EINVAL;
+               }
+
+               if (le16_to_cpu(hdr->item) == item)
+                       return -EEXIST;
+
+               hdr = uncached_entry_next(hdr);
+       }
+
+       /* Check that we don't grow into the cached region */
+       alloc_size = sizeof(*hdr) + ALIGN(size, 8);
+       if ((void *)hdr + alloc_size >= cached) {
+               dev_err(smem->dev, "Out of memory\n");
+               return -ENOSPC;
+       }
+
+       hdr->canary = SMEM_PRIVATE_CANARY;
+       hdr->item = cpu_to_le16(item);
+       hdr->size = cpu_to_le32(ALIGN(size, 8));
+       hdr->padding_data = cpu_to_le16(le32_to_cpu(hdr->size) - size);
+       hdr->padding_hdr = 0;
+
+       /*
+        * Ensure the header is written before we advance the free offset, so
+        * that remote processors that does not take the remote spinlock still
+        * gets a consistent view of the linked list.
+        */
+       dmb();
+       le32_add_cpu(&phdr->offset_free_uncached, alloc_size);
+
+       return 0;
+}
+
+static int qcom_smem_alloc_global(struct qcom_smem *smem,
+                                 unsigned int item,
+                                 size_t size)
+{
+       struct smem_global_entry *entry;
+       struct smem_header *header;
+
+       header = smem->regions[0].virt_base;
+       entry = &header->toc[item];
+       if (entry->allocated)
+               return -EEXIST;
+
+       size = ALIGN(size, 8);
+       if (WARN_ON(size > le32_to_cpu(header->available)))
+               return -ENOMEM;
+
+       entry->offset = header->free_offset;
+       entry->size = cpu_to_le32(size);
+
+       /*
+        * Ensure the header is consistent before we mark the item allocated,
+        * so that remote processors will get a consistent view of the item
+        * even though they do not take the spinlock on read.
+        */
+       dmb();
+       entry->allocated = cpu_to_le32(1);
+
+       le32_add_cpu(&header->free_offset, size);
+       le32_add_cpu(&header->available, -size);
+
+       return 0;
+}
+
+/**
+ * qcom_smem_alloc() - allocate space for a smem item
+ * @host:      remote processor id, or -1
+ * @item:      smem item handle
+ * @size:      number of bytes to be allocated
+ *
+ * Allocate space for a given smem item of size @size, given that the item is
+ * not yet allocated.
+ */
+static int qcom_smem_alloc(unsigned int host, unsigned int item, size_t size)
+{
+       struct smem_partition_header *phdr;
+       int ret;
+
+       if (!__smem)
+               return -EPROBE_DEFER;
+
+       if (item < SMEM_ITEM_LAST_FIXED) {
+               dev_err(__smem->dev,
+                       "Rejecting allocation of static entry %d\n", item);
+               return -EINVAL;
+       }
+
+       if (WARN_ON(item >= __smem->item_count))
+               return -EINVAL;
+
+       if (host < SMEM_HOST_COUNT && __smem->partitions[host]) {
+               phdr = __smem->partitions[host];
+               ret = qcom_smem_alloc_private(__smem, phdr, item, size);
+       } else if (__smem->global_partition) {
+               phdr = __smem->global_partition;
+               ret = qcom_smem_alloc_private(__smem, phdr, item, size);
+       } else {
+               ret = qcom_smem_alloc_global(__smem, item, size);
+       }
+
+       return ret;
+}
+
+static void *qcom_smem_get_global(struct qcom_smem *smem,
+                                 unsigned int item,
+                                 size_t *size)
+{
+       struct smem_header *header;
+       struct smem_region *area;
+       struct smem_global_entry *entry;
+       u32 aux_base;
+       unsigned int i;
+
+       header = smem->regions[0].virt_base;
+       entry = &header->toc[item];
+       if (!entry->allocated)
+               return ERR_PTR(-ENXIO);
+
+       aux_base = le32_to_cpu(entry->aux_base) & AUX_BASE_MASK;
+
+       for (i = 0; i < smem->num_regions; i++) {
+               area = &smem->regions[i];
+
+               if (area->aux_base == aux_base || !aux_base) {
+                       if (size != NULL)
+                               *size = le32_to_cpu(entry->size);
+                       return area->virt_base + le32_to_cpu(entry->offset);
+               }
+       }
+
+       return ERR_PTR(-ENOENT);
+}
+
+static void *qcom_smem_get_private(struct qcom_smem *smem,
+                                  struct smem_partition_header *phdr,
+                                  size_t cacheline,
+                                  unsigned int item,
+                                  size_t *size)
+{
+       struct smem_private_entry *e, *end;
+
+       e = phdr_to_first_uncached_entry(phdr);
+       end = phdr_to_last_uncached_entry(phdr);
+
+       while (e < end) {
+               if (e->canary != SMEM_PRIVATE_CANARY)
+                       goto invalid_canary;
+
+               if (le16_to_cpu(e->item) == item) {
+                       if (size != NULL)
+                               *size = le32_to_cpu(e->size) -
+                                       le16_to_cpu(e->padding_data);
+
+                       return uncached_entry_to_item(e);
+               }
+
+               e = uncached_entry_next(e);
+       }
+
+       /* Item was not found in the uncached list, search the cached list */
+
+       e = phdr_to_first_cached_entry(phdr, cacheline);
+       end = phdr_to_last_cached_entry(phdr);
+
+       while (e > end) {
+               if (e->canary != SMEM_PRIVATE_CANARY)
+                       goto invalid_canary;
+
+               if (le16_to_cpu(e->item) == item) {
+                       if (size != NULL)
+                               *size = le32_to_cpu(e->size) -
+                                       le16_to_cpu(e->padding_data);
+
+                       return cached_entry_to_item(e);
+               }
+
+               e = cached_entry_next(e, cacheline);
+       }
+
+       return ERR_PTR(-ENOENT);
+
+invalid_canary:
+       dev_err(smem->dev, "Found invalid canary in hosts %d:%d partition\n",
+                       phdr->host0, phdr->host1);
+
+       return ERR_PTR(-EINVAL);
+}
+
+/**
+ * qcom_smem_get() - resolve ptr of size of a smem item
+ * @host:      the remote processor, or -1
+ * @item:      smem item handle
+ * @size:      pointer to be filled out with size of the item
+ *
+ * Looks up smem item and returns pointer to it. Size of smem
+ * item is returned in @size.
+ */
+static void *qcom_smem_get(unsigned int host, unsigned int item, size_t *size)
+{
+       struct smem_partition_header *phdr;
+       size_t cacheln;
+       void *ptr = ERR_PTR(-EPROBE_DEFER);
+
+       if (!__smem)
+               return ptr;
+
+       if (WARN_ON(item >= __smem->item_count))
+               return ERR_PTR(-EINVAL);
+
+       if (host < SMEM_HOST_COUNT && __smem->partitions[host]) {
+               phdr = __smem->partitions[host];
+               cacheln = __smem->cacheline[host];
+               ptr = qcom_smem_get_private(__smem, phdr, cacheln, item, size);
+       } else if (__smem->global_partition) {
+               phdr = __smem->global_partition;
+               cacheln = __smem->global_cacheline;
+               ptr = qcom_smem_get_private(__smem, phdr, cacheln, item, size);
+       } else {
+               ptr = qcom_smem_get_global(__smem, item, size);
+       }
+
+       return ptr;
+
+}
+
+/**
+ * qcom_smem_get_free_space() - retrieve amount of free space in a partition
+ * @host:      the remote processor identifying a partition, or -1
+ *
+ * To be used by smem clients as a quick way to determine if any new
+ * allocations has been made.
+ */
+static int qcom_smem_get_free_space(unsigned int host)
+{
+       struct smem_partition_header *phdr;
+       struct smem_header *header;
+       unsigned int ret;
+
+       if (!__smem)
+               return -EPROBE_DEFER;
+
+       if (host < SMEM_HOST_COUNT && __smem->partitions[host]) {
+               phdr = __smem->partitions[host];
+               ret = le32_to_cpu(phdr->offset_free_cached) -
+                     le32_to_cpu(phdr->offset_free_uncached);
+       } else if (__smem->global_partition) {
+               phdr = __smem->global_partition;
+               ret = le32_to_cpu(phdr->offset_free_cached) -
+                     le32_to_cpu(phdr->offset_free_uncached);
+       } else {
+               header = __smem->regions[0].virt_base;
+               ret = le32_to_cpu(header->available);
+       }
+
+       return ret;
+}
+
+static int qcom_smem_get_sbl_version(struct qcom_smem *smem)
+{
+       struct smem_header *header;
+       __le32 *versions;
+
+       header = smem->regions[0].virt_base;
+       versions = header->version;
+
+       return le32_to_cpu(versions[SMEM_MASTER_SBL_VERSION_INDEX]);
+}
+
+static struct smem_ptable *qcom_smem_get_ptable(struct qcom_smem *smem)
+{
+       struct smem_ptable *ptable;
+       u32 version;
+
+       ptable = smem->regions[0].virt_base + smem->regions[0].size - SZ_4K;
+       if (memcmp(ptable->magic, SMEM_PTABLE_MAGIC, sizeof(ptable->magic)))
+               return ERR_PTR(-ENOENT);
+
+       version = le32_to_cpu(ptable->version);
+       if (version != 1) {
+               dev_err(smem->dev,
+                       "Unsupported partition header version %d\n", version);
+               return ERR_PTR(-EINVAL);
+       }
+       return ptable;
+}
+
+static u32 qcom_smem_get_item_count(struct qcom_smem *smem)
+{
+       struct smem_ptable *ptable;
+       struct smem_info *info;
+
+       ptable = qcom_smem_get_ptable(smem);
+       if (IS_ERR_OR_NULL(ptable))
+               return SMEM_ITEM_COUNT;
+
+       info = (struct smem_info *)&ptable->entry[ptable->num_entries];
+       if (memcmp(info->magic, SMEM_INFO_MAGIC, sizeof(info->magic)))
+               return SMEM_ITEM_COUNT;
+
+       return le16_to_cpu(info->num_items);
+}
+
+static int qcom_smem_set_global_partition(struct qcom_smem *smem)
+{
+       struct smem_partition_header *header;
+       struct smem_ptable_entry *entry = NULL;
+       struct smem_ptable *ptable;
+       u32 host0, host1, size;
+       int i;
+
+       ptable = qcom_smem_get_ptable(smem);
+       if (IS_ERR(ptable))
+               return PTR_ERR(ptable);
+
+       for (i = 0; i < le32_to_cpu(ptable->num_entries); i++) {
+               entry = &ptable->entry[i];
+               host0 = le16_to_cpu(entry->host0);
+               host1 = le16_to_cpu(entry->host1);
+
+               if (host0 == SMEM_GLOBAL_HOST && host0 == host1)
+                       break;
+       }
+
+       if (!entry) {
+               dev_err(smem->dev, "Missing entry for global partition\n");
+               return -EINVAL;
+       }
+
+       if (!le32_to_cpu(entry->offset) || !le32_to_cpu(entry->size)) {
+               dev_err(smem->dev, "Invalid entry for global partition\n");
+               return -EINVAL;
+       }
+
+       if (smem->global_partition) {
+               dev_err(smem->dev, "Already found the global partition\n");
+               return -EINVAL;
+       }
+
+       header = smem->regions[0].virt_base + le32_to_cpu(entry->offset);
+       host0 = le16_to_cpu(header->host0);
+       host1 = le16_to_cpu(header->host1);
+
+       if (memcmp(header->magic, SMEM_PART_MAGIC, sizeof(header->magic))) {
+               dev_err(smem->dev, "Global partition has invalid magic\n");
+               return -EINVAL;
+       }
+
+       if (host0 != SMEM_GLOBAL_HOST && host1 != SMEM_GLOBAL_HOST) {
+               dev_err(smem->dev, "Global partition hosts are invalid\n");
+               return -EINVAL;
+       }
+
+       if (le32_to_cpu(header->size) != le32_to_cpu(entry->size)) {
+               dev_err(smem->dev, "Global partition has invalid size\n");
+               return -EINVAL;
+       }
+
+       size = le32_to_cpu(header->offset_free_uncached);
+       if (size > le32_to_cpu(header->size)) {
+               dev_err(smem->dev,
+                       "Global partition has invalid free pointer\n");
+               return -EINVAL;
+       }
+
+       smem->global_partition = header;
+       smem->global_cacheline = le32_to_cpu(entry->cacheline);
+
+       return 0;
+}
+
+static int qcom_smem_enumerate_partitions(struct qcom_smem *smem,
+                                         unsigned int local_host)
+{
+       struct smem_partition_header *header;
+       struct smem_ptable_entry *entry;
+       struct smem_ptable *ptable;
+       unsigned int remote_host;
+       u32 host0, host1;
+       int i;
+
+       ptable = qcom_smem_get_ptable(smem);
+       if (IS_ERR(ptable))
+               return PTR_ERR(ptable);
+
+       for (i = 0; i < le32_to_cpu(ptable->num_entries); i++) {
+               entry = &ptable->entry[i];
+               host0 = le16_to_cpu(entry->host0);
+               host1 = le16_to_cpu(entry->host1);
+
+               if (host0 != local_host && host1 != local_host)
+                       continue;
+
+               if (!le32_to_cpu(entry->offset))
+                       continue;
+
+               if (!le32_to_cpu(entry->size))
+                       continue;
+
+               if (host0 == local_host)
+                       remote_host = host1;
+               else
+                       remote_host = host0;
+
+               if (remote_host >= SMEM_HOST_COUNT) {
+                       dev_err(smem->dev,
+                               "Invalid remote host %d\n",
+                               remote_host);
+                       return -EINVAL;
+               }
+
+               if (smem->partitions[remote_host]) {
+                       dev_err(smem->dev,
+                               "Already found a partition for host %d\n",
+                               remote_host);
+                       return -EINVAL;
+               }
+
+               header = smem->regions[0].virt_base + le32_to_cpu(entry->offset);
+               host0 = le16_to_cpu(header->host0);
+               host1 = le16_to_cpu(header->host1);
+
+               if (memcmp(header->magic, SMEM_PART_MAGIC,
+                           sizeof(header->magic))) {
+                       dev_err(smem->dev,
+                               "Partition %d has invalid magic\n", i);
+                       return -EINVAL;
+               }
+
+               if (host0 != local_host && host1 != local_host) {
+                       dev_err(smem->dev,
+                               "Partition %d hosts are invalid\n", i);
+                       return -EINVAL;
+               }
+
+               if (host0 != remote_host && host1 != remote_host) {
+                       dev_err(smem->dev,
+                               "Partition %d hosts are invalid\n", i);
+                       return -EINVAL;
+               }
+
+               if (le32_to_cpu(header->size) != le32_to_cpu(entry->size)) {
+                       dev_err(smem->dev,
+                               "Partition %d has invalid size\n", i);
+                       return -EINVAL;
+               }
+
+               if (le32_to_cpu(header->offset_free_uncached) > le32_to_cpu(header->size)) {
+                       dev_err(smem->dev,
+                               "Partition %d has invalid free pointer\n", i);
+                       return -EINVAL;
+               }
+
+               smem->partitions[remote_host] = header;
+               smem->cacheline[remote_host] = le32_to_cpu(entry->cacheline);
+       }
+
+       return 0;
+}
+
+static int qcom_smem_map_memory(struct qcom_smem *smem, struct udevice *dev,
+                               const char *name, int i)
+{
+       struct fdt_resource r;
+       int ret;
+       int node = dev_of_offset(dev);
+
+       ret = fdtdec_lookup_phandle(gd->fdt_blob, node, name);
+       if (ret < 0) {
+               dev_err(dev, "No %s specified\n", name);
+               return -EINVAL;
+       }
+
+       ret = fdt_get_resource(gd->fdt_blob, ret, "reg", 0, &r);
+       if (ret)
+               return ret;
+
+       smem->regions[i].aux_base = (u32)r.start;
+       smem->regions[i].size = fdt_resource_size(&r);
+       smem->regions[i].virt_base = devm_ioremap(dev, r.start, fdt_resource_size(&r));
+       if (!smem->regions[i].virt_base)
+               return -ENOMEM;
+
+       return 0;
+}
+
+static int qcom_smem_probe(struct udevice *dev)
+{
+       struct smem_header *header;
+       struct qcom_smem *smem;
+       size_t array_size;
+       int num_regions;
+       u32 version;
+       int ret;
+       int node = dev_of_offset(dev);
+
+       num_regions = 1;
+       if (fdtdec_lookup_phandle(gd->fdt_blob, node, "qcomrpm-msg-ram") >= 0)
+               num_regions++;
+
+       array_size = num_regions * sizeof(struct smem_region);
+       smem = devm_kzalloc(dev, sizeof(*smem) + array_size, GFP_KERNEL);
+       if (!smem)
+               return -ENOMEM;
+
+       smem->dev = dev;
+       smem->num_regions = num_regions;
+
+       ret = qcom_smem_map_memory(smem, dev, "memory-region", 0);
+       if (ret)
+               return ret;
+
+       if (num_regions > 1) {
+               ret = qcom_smem_map_memory(smem, dev,
+                                       "qcom,rpm-msg-ram", 1);
+               if (ret)
+                       return ret;
+       }
+
+       header = smem->regions[0].virt_base;
+       if (le32_to_cpu(header->initialized) != 1 ||
+           le32_to_cpu(header->reserved)) {
+               dev_err(&pdev->dev, "SMEM is not initialized by SBL\n");
+               return -EINVAL;
+       }
+
+       version = qcom_smem_get_sbl_version(smem);
+       switch (version >> 16) {
+       case SMEM_GLOBAL_PART_VERSION:
+               ret = qcom_smem_set_global_partition(smem);
+               if (ret < 0)
+                       return ret;
+               smem->item_count = qcom_smem_get_item_count(smem);
+               break;
+       case SMEM_GLOBAL_HEAP_VERSION:
+               smem->item_count = SMEM_ITEM_COUNT;
+               break;
+       default:
+               dev_err(dev, "Unsupported SMEM version 0x%x\n", version);
+               return -EINVAL;
+       }
+
+       ret = qcom_smem_enumerate_partitions(smem, SMEM_HOST_APPS);
+       if (ret < 0 && ret != -ENOENT)
+               return ret;
+
+       __smem = smem;
+
+       return 0;
+}
+
+static int qcom_smem_remove(struct udevice *dev)
+{
+       __smem = NULL;
+
+       return 0;
+}
+
+const struct udevice_id qcom_smem_of_match[] = {
+       { .compatible = "qcom,smem" },
+       { }
+};
+
+static const struct smem_ops msm_smem_ops = {
+       .alloc = qcom_smem_alloc,
+       .get = qcom_smem_get,
+       .get_free_space = qcom_smem_get_free_space,
+};
+
+U_BOOT_DRIVER(qcom_smem) = {
+       .name   = "qcom_smem",
+       .id     = UCLASS_SMEM,
+       .of_match = qcom_smem_of_match,
+       .ops = &msm_smem_ops,
+       .probe = qcom_smem_probe,
+       .remove = qcom_smem_remove,
+};