[MERGE] Adding possibility to use shared atomic counter for IOp ack & to access register through PCIe BAR0 (#10)

* [FEAT] adds ability to use shared atomic through mmap to send ack
* [FEAT] update AMI driver to allow mmap of PF0 BAR0 from user application
* [CHORE] ami driver is compatible with backend not using BAR0 registers or shared_atomic_cnt so moving to its version 3.1.1

Author: pgardratzama

sw/AMI/driver/amc_proxy.c

@@ -19,7 +19,6 @@
 #include "ami_program.h"  /* Need some #defines from here */
 #include "ami_iop_ack_proc.h"
 
-extern wait_queue_head_t wait_iop_q;
 /*****************************************************************************/
 /* Defines                                                                   */
 /*****************************************************************************/
@@ -854,14 +853,12 @@ static int complete_response_thread(void *data)
                             PR_DBG("iopAck[%d] = 0x%x", i, iop_ack_table[i]);
                         }
                         atomic_add(iop_ack_table_size, iop_ack_cnt_atomic);
-                        PR_DBG("Add %d to iop_ack_cnt_atomic to %d", iop_ack_table_size, atomic_read(iop_ack_cnt_atomic));
 
                         // Acknowledge queue consumption
                         ackq_tail += ackq_used_words;
                         iowrite32(ackq_tail, amc_proxy_inst->amc_ctrl_ctxt->gcq_payload_base_virt_addr + AMC_IOPACK_ADDR_TAIL);
-                        // signaling ami_top that waiting reader can be woken
-                        wake_up(&wait_iop_q);
                         kfree(iopAck);
+                        PR_INFO("added %d to iop_ack_cnt_atomic of device %d", iop_ack_table_size, cdev_num);
                 }
 
                 /* Check if a stop has been requested */

sw/AMI/driver/ami.h

@@ -11,7 +11,7 @@
 #include <linux/printk.h>
 
 /* Meta Information */
-#define MDL_VERSION     "3.1.0-zama"
+#define MDL_VERSION     "3.1.1-zama"
 #define MDL_DESCRIPTION "AVED Management Interface (AMI) is used to manage AVED-based devices through PCIe"
 #define MDL_AUTHOR      "AMD, Inc."
 #define MDL_RELDATE     "2023"

sw/AMI/driver/ami_cdev.c

@@ -86,6 +86,45 @@ int dev_close(struct inode *inode, struct file *filp)
     return 0;
 }
 
+int dev_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+    struct pf_dev_struct *pf_dev = NULL;
+
+    pf_dev = filp->private_data;
+
+    /* Check device data */
+    if (!pf_dev) {
+        PR_ERR("dev_mmap: unable to find card");
+        return -ENODEV;
+    }
+    // 1. Get the physical address of BAR0 (stored in your pdev struct)
+    unsigned long phys_addr = pci_resource_start(pf_dev->pci, 0);
+    unsigned long size = vma->vm_end - vma->vm_start;
+    unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+
+    PR_INFO("mmap BAR0 phys_addr 0x%lx size 0x%lx on 0x%lx", phys_addr + offset, size, vma->vm_start);
+
+    // 2. Safety Check: Ensure user isn't asking for more than the BAR size
+    unsigned long bar_len = pci_resource_len(pf_dev->pci, 0);
+    if (offset + size > bar_len) {
+        PR_ERR("dev_mmap failed because offset 0x%lx + size 0x%lx goes beyond bar size 0x%lx", offset, size, bar_len);
+        return -EINVAL;
+    }
+
+    // 3. Mark the memory as "uncached" (Crucial for hardware registers!)
+    vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+
+    if (remap_pfn_range(vma,
+                        vma->vm_start,           // User virtual address start
+                        (phys_addr + offset) >> PAGE_SHIFT, // Physical PFN (addr / 4096)
+                        size,                    // Size
+                        vma->vm_page_prot)) {    // Protection flags
+        return -EAGAIN;
+    }
+
+    return 0;
+}
+
 /*
  * This function will be called when we use IOCTL with command on the Device file
  */

sw/AMI/driver/ami_cdev.h

@@ -259,6 +259,7 @@ struct drv_cdev_struct {
 /* Standard Linux callbacks */
 int dev_open(struct inode *inode, struct file *filp);
 int dev_close(struct inode *inode, struct file *filp);
+int dev_mmap(struct file *filp, struct vm_area_struct *vma);
 long dev_unlocked_ioctl(struct file *filp, unsigned int cmd, unsigned long arg);
 
 /**

sw/AMI/driver/ami_driver_version.h

@@ -5,10 +5,10 @@
 #ifndef _AMI_DRIVER_VERSION_H_
 #define _AMI_DRIVER_VERSION_H_
 
-#define GIT_TAG                   "3.0.0"
+#define GIT_TAG                   "3.1.1"
 #define GIT_TAG_VER_MAJOR         (3)
-#define GIT_TAG_VER_MINOR         (0)
-#define GIT_TAG_VER_PATCH         (0)
+#define GIT_TAG_VER_MINOR         (1)
+#define GIT_TAG_VER_PATCH         (1)
 #define GIT_TAG_VER_DEV_COMMITS   (0)
 
 #define GIT_HASH                  "0bab29e568f64a25f17425c0ffd1c0e89609b6d1"

sw/AMI/driver/ami_iop_ack_proc.c

@@ -10,8 +10,6 @@
 #define PROC_FILENAME_MAXLENGTH 20
 
 static ack_proc_file *ack_proc_file_list = NULL;
-/* Queue of processes who want data */
-DECLARE_WAIT_QUEUE_HEAD(wait_iop_q);
 /* Queue of processes who want our file */
 static DECLARE_WAIT_QUEUE_HEAD(waitq);
 
@@ -154,9 +152,40 @@ static int ami_close(struct inode *inode, struct file *file)
     return 0; /* success */
 }
 
+static int ami_mmap(struct file *file, struct vm_area_struct *vma) {
+    unsigned long pfn;
+    ack_proc_file *read_apf=NULL;
+
+    if (file == NULL) {
+        PR_ERR("ami_mmap: file* %p is NULL", file);
+        return -ENODEV;
+    }
+
+    read_apf = find_ack_proc_file_by_file(file);
+    if (read_apf == NULL) {
+        PR_ERR("ami_mmap: Ack file corresponding to file* %p not found", file);
+        return -EINVAL;
+    }
+    // Get Physical Frame Number (PFN) of our allocated page
+    pfn = virt_to_phys(read_apf) >> PAGE_SHIFT;
+
+    if ((vma->vm_end - vma->vm_start) > PAGE_SIZE) {
+        PR_ERR("ami_mmap: mmap request bigger than a page rejected");
+        return -EINVAL;
+    }
+    // Map it to user space
+    if (remap_pfn_range(vma, vma->vm_start, pfn,
+                        PAGE_SIZE, vma->vm_page_prot)) {
+        return -EAGAIN;
+    }
+
+    return 0;
+}
+
 static const struct proc_ops file_ops_4_ami_proc_file = {
     .proc_read = ami_output, /* "read" from the file */
     .proc_write = NULL, /* "write" to the file */
+    .proc_mmap = ami_mmap,
     .proc_open = ami_open, /* called when the /proc file is opened */
     .proc_release = ami_close, /* called when it's closed */
     .proc_lseek = noop_llseek, /* return file->f_pos */
@@ -178,7 +207,8 @@ int create_proc_file(unsigned dev_index)
     proc_set_size(ami_proc_file, 80);
     proc_set_user(ami_proc_file, GLOBAL_ROOT_UID, GLOBAL_ROOT_GID);
 
-    new_ack_proc_file = kzalloc(sizeof(struct ack_proc_file), GFP_KERNEL);
+    unsigned long page_addr = get_zeroed_page(GFP_KERNEL);
+    new_ack_proc_file = (ack_proc_file*) page_addr;
     new_ack_proc_file->minor_cdev_number = dev_index;
     new_ack_proc_file->ami_proc_file = ami_proc_file;
     atomic_set(&new_ack_proc_file->iop_ack_cnt_atomic, 0);
@@ -216,8 +246,12 @@ unsigned remove_ack_proc_file_by_cdevn(unsigned target_minor_cdev_number) {
                 prev_apf->next = current_apf->next;
             }
 
+            if (current_apf->ami_proc_file) {
+                proc_remove(current_apf->ami_proc_file);
+                current_apf->ami_proc_file = NULL;
+            }
             PR_INFO("Removed /proc node with minor_cdev_number: %u", current_apf->minor_cdev_number);
-            kfree(current_apf);
+            free_page((unsigned long)current_apf);
             return 0;
         }
 
@@ -279,17 +313,10 @@ ack_proc_file* find_ack_proc_file_by_file(struct file *file) {
 
 int delete_proc_file(unsigned dev_index)
 {
-    char proc_filename[PROC_FILENAME_MAXLENGTH];
-
     if (remove_ack_proc_file_by_cdevn(dev_index)) {
         // the proc file to delete has not been found
         return 1;
     }
 
-    snprintf(proc_filename,PROC_FILENAME_MAXLENGTH,"%s_%d",PROC_ENTRY_FILENAME, dev_index);
-
-    remove_proc_entry(proc_filename, NULL);
-    PR_INFO("/proc/%s removed\n", proc_filename);
-
     return 0;
 }

sw/AMI/driver/ami_iop_ack_proc.h

@@ -20,9 +20,9 @@
 #define MESSAGE_LENGTH 80
 
 typedef struct ack_proc_file {
+    atomic_t               iop_ack_cnt_atomic;
     unsigned int           minor_cdev_number;
     struct proc_dir_entry *ami_proc_file;
-    atomic_t               iop_ack_cnt_atomic;
     atomic_t               in_use;
     struct ack_proc_file  *next;
 } ack_proc_file;

sw/AMI/driver/ami_top.c

@@ -71,10 +71,11 @@ void pcie_device_remove(struct pci_dev *dev);
 int register_driver_pcie(void);
 
 static struct file_operations dev_fops = {
-    .owner        = THIS_MODULE,
-    .open        = dev_open,
-    .release    = dev_close,
+    .owner          = THIS_MODULE,
+    .open           = dev_open,
+    .release        = dev_close,
     .unlocked_ioctl = dev_unlocked_ioctl,
+    .mmap           = dev_mmap,
 };
 
 int register_driver_kernel(void)