clear-pkgs-linux-iot-lts2018/0491-sos-clean-up-ptdev-msi...

161 lines
4.6 KiB
Diff

From a97e7a5c4426befa89412c5adc43e69cbcf3dda1 Mon Sep 17 00:00:00 2001
From: Shiqing Gao <shiqing.gao@intel.com>
Date: Fri, 31 Aug 2018 10:58:56 +0800
Subject: [PATCH 491/743] sos: clean up ptdev msi-x table ioremap operations
- use list rather than array to store the virtual base address
of MSI-X table in memory space after ioremap
- use physical BDF rather than virtual BDF for index purpose
this could avoid the issue when different UOSs might use same
virtual BDF for different pass-through PCI devices.
Change-Id: Ie8330bc5054ba549a11866b84770df5d1a257a6c
Tracked-On:218445
Signed-off-by: Shiqing Gao <shiqing.gao@intel.com>
Reviewed-on:
Reviewed-by: Chi, Mingqiang <mingqiang.chi@intel.com>
Reviewed-by: Dong, Eddie <eddie.dong@intel.com>
Tested-by: Dong, Eddie <eddie.dong@intel.com>
---
drivers/char/vhm/vhm_dev.c | 70 +++++++++++++++++++++-----------------
1 file changed, 38 insertions(+), 32 deletions(-)
diff --git a/drivers/char/vhm/vhm_dev.c b/drivers/char/vhm/vhm_dev.c
index d7ad18f6a55e..5130fb508bc9 100644
--- a/drivers/char/vhm/vhm_dev.c
+++ b/drivers/char/vhm/vhm_dev.c
@@ -75,6 +75,7 @@
#include <linux/uaccess.h>
#include <linux/io.h>
#include <linux/pci.h>
+#include <linux/list.h>
#include <linux/vhm/acrn_hv_defs.h>
#include <linux/vhm/vhm_ioctl_defs.h>
@@ -92,15 +93,16 @@ static struct device *vhm_device;
static struct tasklet_struct vhm_io_req_tasklet;
static atomic_t ioreq_retry = ATOMIC_INIT(0);
-/* max num of pass-through devices using msix */
-#define MAX_ENTRY 3
-
struct table_iomems {
- /* device's virtual BDF */
- unsigned short virt_bdf;
+ /* list node for this table_iomems */
+ struct list_head list;
+ /* device's physical BDF */
+ unsigned short phys_bdf;
/* virtual base address of MSI-X table in memory space after ioremap */
unsigned long mmap_addr;
-} tables[MAX_ENTRY];
+};
+static LIST_HEAD(table_iomems_list);
+static DEFINE_MUTEX(table_iomems_lock);
static int vhm_dev_open(struct inode *inodep, struct file *filep)
{
@@ -392,7 +394,7 @@ static long vhm_dev_ioctl(struct file *filep,
case IC_SET_PTDEV_INTR_INFO: {
struct acrn_ptdev_irq pt_irq;
- int i;
+ struct table_iomems *new;
if (copy_from_user(&pt_irq,
(void *)ioctl_param, sizeof(pt_irq)))
@@ -406,23 +408,24 @@ static long vhm_dev_ioctl(struct file *filep,
}
if (pt_irq.msix.table_paddr) {
- for (i = 0; i < MAX_ENTRY; i++) {
- if (tables[i].virt_bdf)
- continue;
-
- tables[i].virt_bdf = pt_irq.virt_bdf;
- tables[i].mmap_addr =
- ioremap_nocache(pt_irq.msix.table_paddr,
+ new = kmalloc(sizeof(struct table_iomems), GFP_KERNEL);
+ if (new == NULL)
+ return -EFAULT;
+ new->phys_bdf = pt_irq.phys_bdf;
+ new->mmap_addr = (unsigned long)
+ ioremap_nocache(pt_irq.msix.table_paddr,
pt_irq.msix.table_size);
- break;
- }
+
+ mutex_lock(&table_iomems_lock);
+ list_add(&new->list, &table_iomems_list);
+ mutex_unlock(&table_iomems_lock);
}
break;
}
case IC_RESET_PTDEV_INTR_INFO: {
struct acrn_ptdev_irq pt_irq;
- int i;
+ struct table_iomems *new;
if (copy_from_user(&pt_irq,
(void *)ioctl_param, sizeof(pt_irq)))
@@ -436,16 +439,17 @@ static long vhm_dev_ioctl(struct file *filep,
}
if (pt_irq.msix.table_paddr) {
- for (i = 0; i < MAX_ENTRY; i++) {
- if (tables[i].virt_bdf)
- continue;
-
- tables[i].virt_bdf = pt_irq.virt_bdf;
- tables[i].mmap_addr =
- ioremap_nocache(pt_irq.msix.table_paddr,
+ new = kmalloc(sizeof(struct table_iomems), GFP_KERNEL);
+ if (new == NULL)
+ return -EFAULT;
+ new->phys_bdf = pt_irq.phys_bdf;
+ new->mmap_addr = (unsigned long)
+ ioremap_nocache(pt_irq.msix.table_paddr,
pt_irq.msix.table_size);
- break;
- }
+
+ mutex_lock(&table_iomems_lock);
+ list_add(&new->list, &table_iomems_list);
+ mutex_unlock(&table_iomems_lock);
}
break;
@@ -466,19 +470,21 @@ static long vhm_dev_ioctl(struct file *filep,
if (msix_remap.msix) {
void __iomem *msix_entry;
- int i;
+ struct table_iomems *ptr;
- for (i = 0; i < MAX_ENTRY; i++) {
- if (tables[i].virt_bdf == msix_remap.virt_bdf)
+ mutex_lock(&table_iomems_lock);
+ list_for_each_entry(ptr, &table_iomems_list, list) {
+ if (ptr->phys_bdf == msix_remap.phys_bdf)
break;
}
+ mutex_unlock(&table_iomems_lock);
- if (!tables[i].mmap_addr)
+ if (!ptr->mmap_addr)
return -EFAULT;
- msix_entry = tables[i].mmap_addr +
+ msix_entry = (void __iomem *) (ptr->mmap_addr +
msix_remap.msix_entry_index *
- PCI_MSIX_ENTRY_SIZE;
+ PCI_MSIX_ENTRY_SIZE);
/* mask the entry when setup */
writel(PCI_MSIX_ENTRY_CTRL_MASKBIT,
--
2.19.2