Skip to contentSkip to conversation

enlarge intr remapping table

PATCH In review Zhai, Edwin 2 participants 5 messages Sep 25, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 enlarge intr remapping table

This patch enlarges interrupt remapping table to fix the out-of range table access when using many multiple-function PCI devices.

Signature · 2 lines
best rgds,
edwin
xen/drivers/passthrough/vtd/intremap.c +3 -3
Index: hv/xen/drivers/passthrough/vtd/intremap.c
===================================================================
@@ -208,7 +208,7 @@ static int ioapic_rte_to_remap_entry(str
     {
         dprintk(XENLOG_ERR VTDPREFIX,
                 "%s: intremap index (%d) is larger than"
-                " the maximum index (%ld)!\n",
+                " the maximum index (%d)!\n",
                 __func__, index, IREMAP_ENTRY_NR - 1);
         spin_unlock_irqrestore(&ir_ctrl->iremap_lock, flags);
         return -EFAULT;
@@ -502,7 +502,7 @@ static int msi_msg_to_remap_entry(
     {
         dprintk(XENLOG_ERR VTDPREFIX,
                 "%s: intremap index (%d) is larger than"
-                " the maximum index (%ld)!\n",
+                " the maximum index (%d)!\n",
                 __func__, index, IREMAP_ENTRY_NR - 1);
         msi_desc->remap_index = -1;
         spin_unlock_irqrestore(&ir_ctrl->iremap_lock, flags);
@@ -617,7 +617,7 @@ int enable_intremap(struct iommu *iommu)
     if ( ir_ctrl->iremap_maddr == 0 )
     {
         drhd = iommu_to_drhd(iommu);
-        ir_ctrl->iremap_maddr = alloc_pgtable_maddr(drhd, 1);
+        ir_ctrl->iremap_maddr = alloc_pgtable_maddr(drhd, IREMAP_PAGE_NR );
         if ( ir_ctrl->iremap_maddr == 0 )
         {
             dprintk(XENLOG_WARNING VTDPREFIX,
xen/drivers/passthrough/vtd/iommu.h +11 -4
Index: hv/xen/drivers/passthrough/vtd/iommu.h
===================================================================
@@ -302,7 +302,14 @@ struct iremap_entry {
     }hi;
   };
 };
-#define IREMAP_ENTRY_NR (PAGE_SIZE_4K/sizeof(struct iremap_entry))
+
+/* Max intr remapping table page order is 8, as max number of IRTEs is 64K */
+#define IREMAP_PAGE_ORDER  8
+#define IREMAP_PAGE_NR     ( 1 << IREMAP_PAGE_ORDER )
+
+/* Each entry is 16 byte */
+#define IREMAP_ENTRY_NR    ( 1 << ( IREMAP_PAGE_ORDER + 8 ) )
+
 #define iremap_present(v) ((v).lo & 1)
 #define iremap_fault_disable(v) (((v).lo >> 1) & 1)
 
@@ -429,9 +436,9 @@ struct qinval_entry {
 #define IEC_GLOBAL_INVL         0
 #define IEC_INDEX_INVL          1
 #define IRTA_REG_EIME_SHIFT     11
-#define IRTA_REG_TABLE_SIZE     7    // 4k page = 256 * 16 byte entries
-                                     // 2^^(IRTA_REG_TABLE_SIZE + 1) = 256
-                                     // IRTA_REG_TABLE_SIZE = 7
+
+/* 2^(IRTA_REG_TABLE_SIZE + 1) = IREMAP_ENTRY_NR */
+#define IRTA_REG_TABLE_SIZE     ( IREMAP_PAGE_ORDER + 7 )
 
 #define VTD_PAGE_TABLE_LEVEL_3  3
 #define VTD_PAGE_TABLE_LEVEL_4  4

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#1 lore raw
Quoted 5 lines
Quoted 1 line
Quoted 1 line

"Zhai, Edwin" <[email protected]> 25.09.09 09:11 >>>

This patch enlarges interrupt remapping table to fix the out-of range table access when using many multiple-function PCI devices.

This can't be working: Wherever iremap_maddr gets mapped, only a single page is being mapped.

Jan

#2 lore raw
Quoted 11 lines · Jan Beulich
Quoted 7 lines
Quoted 2 lines
Quoted 2 lines

"Zhai, Edwin" <[email protected]> 25.09.09 09:11 >>>

This patch enlarges interrupt remapping table to fix the out-of range table access when using many multiple-function PCI devices.

This can't be working: Wherever iremap_maddr gets mapped, only a single page is being mapped.

Is there any API can be used to map a series of continuous pages?

Quoted 3 lines

Jan

Signature · 2 lines
best rgds,
edwin
#3 lore raw
Quoted 2 lines
Quoted 1 line
Quoted 1 line

"Zhai, Edwin" <[email protected]> 25.09.09 10:36 >>>

Is there any API can be used to map a series of continuous pages?

No, you'll have to figure which one you really need. And even if there was, you surely wouldn't want to map all 256 at one time.

Jan

#4 lore raw

Thanks,
I'll modify my patch.

Quoted 11 lines · Jan Beulich
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines

"Zhai, Edwin" <[email protected]> 25.09.09 10:36 >>>

Is there any API can be used to map a series of continuous pages?

No, you'll have to figure which one you really need. And even if there was, you surely wouldn't want to map all 256 at one time.

Jan

Signature · 2 lines
best rgds,
edwin