Jan Beulich
2011-Jul-12 13:26 UTC
[Xen-devel] [PATCH, RFC] properly determine BAR values for SR-IOV virtual functions
As was discussed a couple of time on this list, SR-IOV virtual
functions have their BARs read as zero - the physical function''s SR-IOV
capability structure must be consulted instead.
The patch is only a first draft, and may not apply as-is to current
-unstable. On some machines it in particular depends on some additional
changes to the MMCONFIG initialization, which I am yet to post (still
missing some parts of it), as accessing the SR-IOV capability structure
requires the extended config space to be readable by Xen.
Issues:
Can we expect doing the BAR sizing from pci_add_device() to be
compatible with all Dom0 OSes? In particular, is the assumption that
virtual functions aren''t enabled at this point reasonable?
Is using dev->devfn for obtaining the VF index really correct (this is
how I see Linux handling it)? How would VFs beyond 255 be expressed?
This is only compile tested, as I have no SR-IOV device available for
testing.
--- 2011-06-19.orig/xen/arch/x86/msi.c 2011-06-22 16:23:01.000000000 +0200
+++ 2011-06-19/xen/arch/x86/msi.c 2011-07-12 15:02:42.000000000 +0200
@@ -522,12 +522,31 @@ static int msi_capability_init(struct pc
return 0;
}
-static u64 read_pci_mem_bar(u8 bus, u8 slot, u8 func, u8 bir)
+static u64 read_pci_mem_bar(u8 bus, u8 slot, u8 func, u8 bir, int vf)
{
u8 limit;
- u32 addr;
+ u32 addr, base = PCI_BASE_ADDRESS_0, disp = 0;
- switch ( pci_conf_read8(bus, slot, func, PCI_HEADER_TYPE) & 0x7f )
+ if ( vf >= 0 )
+ {
+ struct pci_dev *pdev = pci_get_pdev(bus, PCI_DEVFN(slot, func));
+ unsigned int pos = pci_find_ext_capability(0, bus,
+ PCI_DEVFN(slot, func),
+ PCI_EXT_CAP_ID_SRIOV);
+ u16 ctrl = pci_conf_read16(bus, slot, func, pos + PCI_SRIOV_CTRL);
+
+ if ( !pdev || !pos ||
+ !(ctrl & PCI_SRIOV_CTRL_VFE) ||
+ !(ctrl & PCI_SRIOV_CTRL_MSE) ||
+ bir >= PCI_SRIOV_NUM_BARS ||
+ !pdev->vf_rlen[bir] )
+ return 0;
+ base = pos + PCI_SRIOV_BAR;
+ BUILD_BUG_ON(ARRAY_SIZE(pdev->vf_rlen) != PCI_SRIOV_NUM_BARS);
+ disp = vf * pdev->vf_rlen[bir];
+ limit = PCI_SRIOV_NUM_BARS;
+ }
+ else switch ( pci_conf_read8(bus, slot, func, PCI_HEADER_TYPE) & 0x7f )
{
case PCI_HEADER_TYPE_NORMAL:
limit = 6;
@@ -544,7 +563,7 @@ static u64 read_pci_mem_bar(u8 bus, u8 s
if ( bir >= limit )
return 0;
- addr = pci_conf_read32(bus, slot, func, PCI_BASE_ADDRESS_0 + bir * 4);
+ addr = pci_conf_read32(bus, slot, func, base + bir * 4);
if ( (addr & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO )
return 0;
if ( (addr & PCI_BASE_ADDRESS_MEM_TYPE_MASK) ==
PCI_BASE_ADDRESS_MEM_TYPE_64 )
@@ -553,10 +572,9 @@ static u64 read_pci_mem_bar(u8 bus, u8 s
if ( ++bir >= limit )
return 0;
return addr |
- ((u64)pci_conf_read32(bus, slot, func,
- PCI_BASE_ADDRESS_0 + bir * 4) <<
32);
+ ((u64)pci_conf_read32(bus, slot, func, base + bir * 4) <<
32);
}
- return addr & PCI_BASE_ADDRESS_MEM_MASK;
+ return (addr & PCI_BASE_ADDRESS_MEM_MASK) + disp;
}
/**
@@ -629,11 +647,29 @@ static int msix_capability_init(struct p
if ( !dev->msix_nr_entries )
{
+ u8 pbus, pslot, pfunc;
+ int vf;
u64 pba_paddr;
u32 pba_offset;
+ if ( !dev->info.is_virtfn )
+ {
+ pbus = bus;
+ pslot = slot;
+ pfunc = func;
+ vf = -1;
+ }
+ else
+ {
+ pbus = dev->info.physfn.bus;
+ pslot = PCI_SLOT(dev->info.physfn.devfn);
+ pfunc = PCI_FUNC(dev->info.physfn.devfn);
+ vf = dev->devfn;
+ }
+
ASSERT(!dev->msix_used_entries);
- WARN_ON(msi->table_base != read_pci_mem_bar(bus, slot, func, bir));
+ WARN_ON(msi->table_base !+ read_pci_mem_bar(pbus,
pslot, pfunc, bir, vf));
dev->msix_nr_entries = nr_entries;
dev->msix_table.first = PFN_DOWN(table_paddr);
@@ -645,7 +681,7 @@ static int msix_capability_init(struct p
pba_offset = pci_conf_read32(bus, slot, func,
msix_pba_offset_reg(pos));
bir = (u8)(pba_offset & PCI_MSIX_BIRMASK);
- pba_paddr = read_pci_mem_bar(bus, slot, func, bir);
+ pba_paddr = read_pci_mem_bar(pbus, pslot, pfunc, bir, vf);
WARN_ON(!pba_paddr);
pba_paddr += pba_offset & ~PCI_MSIX_BIRMASK;
--- 2011-06-19.orig/xen/drivers/passthrough/pci.c 2011-07-12 14:10:12.000000000
+0200
+++ 2011-06-19/xen/drivers/passthrough/pci.c 2011-07-12 14:59:37.000000000 +0200
@@ -145,6 +145,7 @@ void pci_enable_acs(struct pci_dev *pdev
int pci_add_device(u8 bus, u8 devfn, const struct pci_dev_info *info)
{
struct pci_dev *pdev;
+ unsigned int slot = PCI_SLOT(devfn), func = PCI_FUNC(devfn);
const char *pdev_type;
int ret = -ENOMEM;
@@ -153,7 +154,14 @@ int pci_add_device(u8 bus, u8 devfn, con
else if (info->is_extfn)
pdev_type = "extended function";
else if (info->is_virtfn)
+ {
+ spin_lock(&pcidevs_lock);
+ pdev = pci_get_pdev(info->physfn.bus, info->physfn.devfn)
+ spin_unlock(&pcidevs_lock);
+ if ( !pdev )
+ pci_add_device(info->physfn.bus, info->physfn.devfn, NULL);
pdev_type = "virtual function";
+ }
else
return -EINVAL;
@@ -164,6 +172,67 @@ int pci_add_device(u8 bus, u8 devfn, con
if ( info )
pdev->info = *info;
+ else if ( !pdev->vf_rlen[0] )
+ {
+ unsigned int pos = pci_find_ext_capability(0, bus, devfn,
+ PCI_EXT_CAP_ID_SRIOV);
+ u16 ctrl = pci_conf_read16(bus, slot, func, pos + PCI_SRIOV_CTRL);
+
+ if ( !pos )
+ /* Nothing */;
+ else if ( !(ctrl & (PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE)) )
+ {
+ unsigned int i;
+
+ for ( i = 0; i < PCI_SRIOV_NUM_BARS, ++i )
+ {
+ unsigned int idx = pos + PCI_SRIOV_BAR + i * 4;
+ u32 bar = pci_conf_read32(bus, slot, func, idx);
+ u32 hi = 0;
+
+ if ( (bar & PCI_BASE_ADDRESS_SPACE) =+
PCI_BASE_ADDRESS_SPACE_IO )
+ {
+ printk(XENLOG_WARNING "SR-IOV device %02x:%02x.%x with
vf"
+ " BAR%u in IO space\n",
+ bus, slot, func, i);
+ continue;
+ }
+ pci_conf_write32(bus, slot, func, idx, ~0);
+ if ( (bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK) =+
PCI_BASE_ADDRESS_MEM_TYPE_64 )
+ {
+ if ( i >= PCI_SRIOV_NUM_BARS )
+ {
+ printk(XENLOG_WARNING "SR-IOV device %02x:%02x.%x
with"
+ " 64-bit vf BAR%u in last
slot\n",
+ bus, slot, func);
+ break;
+ }
+ hi = pci_conf_read32(bus, slot, func, idx + 4);
+ pci_conf_write32(bus, slot, func, idx + 4, ~0);
+ }
+ pdev->vf_rlen[i] = pci_conf_read32(bus, slot, func, idx)
&
+ PCI_BASE_ADDRESS_MEM_MASK;
+ if ( (bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK) =+
PCI_BASE_ADDRESS_MEM_TYPE_64 )
+ {
+ pdev->vf_rlen[i] |= ((u64)pci_conf_read32(bus, slot,
func,
+ idx + 4))
<< 32;
+ pci_conf_write32(bus, slot, func, idx + 4, hi);
+ ++i;
+ }
+ else
+ pdev->vf_rlen[i] |= (u64)~0 << 32;
+ pci_conf_write32(bus, slot, func, idx, bar);
+ pdev->vf_rlen[i] = -pdev->vf_rlen[i];
+ }
+ }
+ else
+ printk(XENLOG_WARNING "SR-IOV device %02x:%02x.%x has its
virtual"
+ " functions already enabled
(%04x)\n",
+ bus, slot, func, ctrl);
+ }
ret = 0;
if ( !pdev->domain )
@@ -183,7 +252,7 @@ int pci_add_device(u8 bus, u8 devfn, con
out:
spin_unlock(&pcidevs_lock);
printk(XENLOG_DEBUG "PCI add %s %02x:%02x.%x\n", pdev_type,
- bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+ bus, slot, func);
return ret;
}
--- 2011-06-19.orig/xen/include/xen/pci.h 2011-07-12 14:13:28.000000000 +0200
+++ 2011-06-19/xen/include/xen/pci.h 2011-07-12 14:27:33.000000000 +0200
@@ -57,6 +57,7 @@ struct pci_dev {
const u8 bus;
const u8 devfn;
struct pci_dev_info info;
+ u64 vf_rlen[6];
};
#define for_each_pdev(domain, pdev) \
--- 2011-06-19.orig/xen/include/xen/pci_regs.h 2011-06-22 16:23:01.000000000
+0200
+++ 2011-06-19/xen/include/xen/pci_regs.h 2011-07-12 14:15:50.000000000 +0200
@@ -425,7 +425,7 @@
#define PCI_EXT_CAP_ID_ACS 13
#define PCI_EXT_CAP_ID_ARI 14
#define PCI_EXT_CAP_ID_ATS 15
-#define PCI_EXT_CAP_ID_IOV 16
+#define PCI_EXT_CAP_ID_SRIOV 16
/* Advanced Error Reporting */
#define PCI_ERR_UNCOR_STATUS 4 /* Uncorrectable Error Status */
@@ -545,4 +545,35 @@
#define PCI_ACS_CTRL 0x06 /* ACS Control Register */
#define PCI_ACS_EGRESS_CTL_V 0x08 /* ACS Egress Control Vector */
+/* Single Root I/O Virtualization */
+#define PCI_SRIOV_CAP 0x04 /* SR-IOV Capabilities */
+#define PCI_SRIOV_CAP_VFM 0x01 /* VF Migration Capable */
+#define PCI_SRIOV_CAP_INTR(x) ((x) >> 21) /* Interrupt Message Number */
+#define PCI_SRIOV_CTRL 0x08 /* SR-IOV Control */
+#define PCI_SRIOV_CTRL_VFE 0x01 /* VF Enable */
+#define PCI_SRIOV_CTRL_VFM 0x02 /* VF Migration Enable */
+#define PCI_SRIOV_CTRL_INTR 0x04 /* VF Migration Interrupt Enable */
+#define PCI_SRIOV_CTRL_MSE 0x08 /* VF Memory Space Enable */
+#define PCI_SRIOV_CTRL_ARI 0x10 /* ARI Capable Hierarchy */
+#define PCI_SRIOV_STATUS 0x0a /* SR-IOV Status */
+#define PCI_SRIOV_STATUS_VFM 0x01 /* VF Migration Status */
+#define PCI_SRIOV_INITIAL_VF 0x0c /* Initial VFs */
+#define PCI_SRIOV_TOTAL_VF 0x0e /* Total VFs */
+#define PCI_SRIOV_NUM_VF 0x10 /* Number of VFs */
+#define PCI_SRIOV_FUNC_LINK 0x12 /* Function Dependency Link */
+#define PCI_SRIOV_VF_OFFSET 0x14 /* First VF Offset */
+#define PCI_SRIOV_VF_STRIDE 0x16 /* Following VF Stride */
+#define PCI_SRIOV_VF_DID 0x1a /* VF Device ID */
+#define PCI_SRIOV_SUP_PGSIZE 0x1c /* Supported Page Sizes */
+#define PCI_SRIOV_SYS_PGSIZE 0x20 /* System Page Size */
+#define PCI_SRIOV_BAR 0x24 /* VF BAR0 */
+#define PCI_SRIOV_NUM_BARS 6 /* Number of VF BARs */
+#define PCI_SRIOV_VFM 0x3c /* VF Migration State Array Offset*/
+#define PCI_SRIOV_VFM_BIR(x) ((x) & 7) /* State BIR */
+#define PCI_SRIOV_VFM_OFFSET(x) ((x) & ~7) /* State Offset */
+#define PCI_SRIOV_VFM_UA 0x0 /* Inactive.Unavailable */
+#define PCI_SRIOV_VFM_MI 0x1 /* Dormant.MigrateIn */
+#define PCI_SRIOV_VFM_MO 0x2 /* Active.MigrateOut */
+#define PCI_SRIOV_VFM_AV 0x3 /* Active.Available */
+
#endif /* LINUX_PCI_REGS_H */
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel