core, configs, tools: Prepare for variable IOMMU register set sizes

Introduce a size field to struct jailhouse_iommu and fill it via the
config generator. The information can be retrieved from the ACPI tables
for AMD. On Intel, we need to study the Linux mappings, thus we need to
demand that DMAR is enabled now while retrieving system information.

Based on patches by Valentine Sinitsyn.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

diff --git a/configs/f2a88xm-hd3.c b/configs/f2a88xm-hd3.c
index 825452aacfa0d4ef9afcec2f9ae5d180d43187da..d11ebd8a3b5ecc7df98abc727d8693c10d44fb21 100644 (file)
--- a/configs/f2a88xm-hd3.c
+++ b/configs/f2a88xm-hd3.c
@@ -47,6 +47,7 @@ struct {
                        .iommu_units = {
                                {
                                        .base = 0xfeb80000,
+                                       .size = 0x80000,
                                },
                        },
                },

diff --git a/configs/h87i.c b/configs/h87i.c
index 2015729c90f5b3ecf5f8cc933baa5a00a7912957..5869fb112820374cce1090d6bf6429f5ba57a40a 100644 (file)
--- a/configs/h87i.c
+++ b/configs/h87i.c
@@ -42,9 +42,11 @@ struct {
                        .iommu_units = {
                                {
                                        .base = 0xfed90000,
+                                       .size = 0x1000,
                                },
                                {
                                        .base = 0xfed91000,
+                                       .size = 0x1000,
                                },
                        },
                },

diff --git a/configs/qemu-vm.c b/configs/qemu-vm.c
index 1d7481e8a35f2bd5c35581a09676f66fab85ea2e..64b9a2568270506f2b0271f7aa2872d890f05337 100644 (file)
--- a/configs/qemu-vm.c
+++ b/configs/qemu-vm.c
@@ -56,6 +56,7 @@ struct {
                        .iommu_units = {
                                {
                                        .base = 0xfed90000,
+                                       .size = 0x1000,
                                },
                        },
                },

diff --git a/hypervisor/include/jailhouse/cell-config.h b/hypervisor/include/jailhouse/cell-config.h
index 84ffa5c28478a083b6959b41b6e5fed75020df27..b21bf60de761c7e4005a0be18f383f79c0ea354f 100644 (file)
--- a/hypervisor/include/jailhouse/cell-config.h
+++ b/hypervisor/include/jailhouse/cell-config.h
@@ -141,6 +141,7 @@ struct jailhouse_pci_capability {
 
 struct jailhouse_iommu {
        __u64 base;
+       __u32 size;
 } __attribute__((packed));
 
 #define JAILHOUSE_SYSTEM_SIGNATURE     "JAILSYST"

diff --git a/tools/jailhouse-config-create b/tools/jailhouse-config-create
index 935a5c14d802f6987a11d67f9ebabc24f33cbbb1..b7207fb4145cb0ee181aaab0c6ebdf150fbdc701 100755 (executable)
--- a/tools/jailhouse-config-create
+++ b/tools/jailhouse-config-create
@@ -2,7 +2,13 @@
 #
 # Jailhouse, a Linux-based partitioning hypervisor
 #
-# Copyright (c) Siemens AG, 2014
+# Copyright (c) Siemens AG, 2014-2016
+# Copyright (c) Valentine Sinitsyn, 2014-2015
+#
+# Authors:
+#  Henning Schild <henning.schild@siemens.com>
+#  Jan Kiszka <jan.kiszka@siemens.com>
+#  Valentine Sinitsyn <valentine.sinitsyn@gmail.com>
 #
 # This work is licensed under the terms of the GNU GPL, version 2.  See
 # the COPYING file in the top-level directory.
@@ -532,8 +538,7 @@ class IOMemRegionTree:
             # blacklisted on all levels
             if (
                 (s.find('PCI MMCONFIG') >= 0) or
-                (s.find('APIC') >= 0) or  # covers both APIC and IOAPIC
-                (s.find('dmar') >= 0)
+                (s.find('APIC') >= 0)  # covers both APIC and IOAPIC
             ):
                 continue
 
@@ -553,6 +558,12 @@ class IOMemRegionTree:
         return regions
 
 
+class IOMMUConfig(object):
+    def __init__(self, props):
+        self.base_addr = props['base_addr']
+        self.mmio_size = props['mmio_size']
+
+
 def parse_iomem(pcidevices):
     regions = IOMemRegionTree.parse_iomem_tree(
         IOMemRegionTree.parse_iomem_file())
@@ -561,6 +572,7 @@ def parse_iomem(pcidevices):
     add_rom_region = False
 
     ret = []
+    dmar_regions = []
     for r in regions:
         append_r = True
         # filter the list for MSI-X pages
@@ -580,6 +592,10 @@ def parse_iomem(pcidevices):
         if (r.start >= rom_region.start and r.stop <= rom_region.stop):
             add_rom_region = True
             append_r = False
+        # filter out and save DMAR regions
+        if r.typestr.find('dmar') >= 0:
+            dmar_regions.append(r)
+            append_r = False
         if append_r:
             ret.append(r)
 
@@ -592,7 +608,7 @@ def parse_iomem(pcidevices):
     if (ret[0].typestr == 'System RAM' and ret[0].start == 0x1000):
         ret[0].start = 0
 
-    return ret
+    return ret, dmar_regions
 
 
 def parse_pcidevices():
@@ -704,7 +720,7 @@ def parse_dmar_devscope(f):
 
 # parsing of DMAR ACPI Table
 # see Intel VT-d Spec chapter 8
-def parse_dmar(pcidevices, ioapics):
+def parse_dmar(pcidevices, ioapics, dmar_regions):
     f = input_open('/sys/firmware/acpi/tables/DMAR', 'rb')
     signature = f.read(4)
     if signature != b'DMAR':
@@ -729,7 +745,20 @@ def parse_dmar(pcidevices, ioapics):
             if len(units) >= 8:
                 raise RuntimeError('Too many DMAR units. '
                                    'Raise JAILHOUSE_MAX_IOMMU_UNITS.')
-            units.append(base)
+            size = 0
+            for r in dmar_regions:
+                if base == r.start:
+                    size = r.size()
+            if size == 0:
+                raise RuntimeError('DMAR region size cannot be identified.\n'
+                                   'Target Linux must run with Intel IOMMU '
+                                   'enabled.')
+            if size > 0x3000:
+                raise RuntimeError('Unexpectedly large DMAR region.')
+            units.append(IOMMUConfig({
+                'base_addr': base,
+                'mmio_size': size
+            }))
             if flags & 1:
                 for d in pcidevices:
                     if d.iommu is None:
@@ -825,19 +854,21 @@ def parse_ivrs(pcidevices, ioapics):
     regions = []
     # BDF of devices that are permitted outside IOMMU: root complex
     iommu_skiplist = set([0x0])
+    ivhd_blocks = 0
     while length > 0:
         (block_type, block_length) = struct.unpack('<BxH', f.read(4))
         if block_type in [0x10, 0x11]:
+            ivhd_blocks += 1
+            if ivhd_blocks > 1:
+                raise RuntimeError('Jailhouse doesn\'t support more than one '
+                                   'AMD IOMMU per PCI function.')
             # IVHD block
-            (iommu_id, base_addr, pci_seg) = \
-                struct.unpack('<HxxQH', f.read(14))
-            length -= block_length
-            block_length -= 18
+            ivhd_fields = struct.unpack('<HHQHxxL', f.read(20))
+            (iommu_bdf, base_cap_ofs,
+             base_addr, pci_seg, iommu_feat) = ivhd_fields
 
-            # IOMMU EFR image and reserved area
-            skip_bytes = 6 if block_type == 0x10 else 22
-            f.seek(skip_bytes, os.SEEK_CUR)
-            block_length -= skip_bytes
+            length -= block_length
+            block_length -= 24
 
             if pci_seg != 0:
                 raise RuntimeError('We do not support multiple PCI segments')
@@ -846,12 +877,22 @@ def parse_ivrs(pcidevices, ioapics):
                 raise RuntimeError('Too many IOMMU units. '
                                    'Raise JAILHOUSE_MAX_IOMMU_UNITS.')
 
-            # We shouldn't map IOMMU to the cells
             for i, d in enumerate(pcidevices):
-                if d.bdf() == iommu_id:
+                if d.bdf() == iommu_bdf:
+                    # We must not map IOMMU to the cells
                     del pcidevices[i]
 
-            units.append(base_addr)
+            if (iommu_feat & (0xF << 13)) and (iommu_feat & (0x3F << 17)):
+                # Performance Counters are supported, allocate 512K
+                mmio_size = 524288
+            else:
+                # Allocate 16K
+                mmio_size = 16384
+
+            units.append(IOMMUConfig({
+                'base_addr': base_addr,
+                'mmio_size': mmio_size,
+            }))
 
             bdf_start_range = None
             while block_length > 0:
@@ -908,9 +949,9 @@ def parse_ivrs(pcidevices, ioapics):
 
         elif type in [0x20, 0x21, 0x22]:
             # IVMD block
+            ivmd_fields = struct.unpack('<BBHHHxxxxxxxxQQ', f.read(32))
             (block_type, block_flags, block_length,
-             device_id, aux_data, mem_addr, mem_len) = struct.unpack(
-                 '<BBHHHxxxxxxxxQQ')
+             device_id, aux_data, mem_addr, mem_len) = ivmd_fields
             length -= block_length
 
             if int(block_flags):
@@ -1042,7 +1083,7 @@ product = [input_readline('/sys/class/dmi/id/sys_vendor',
 inmatemem = kmg_multiply_str(options.mem_inmates)
 hvmem = [0, kmg_multiply_str(options.mem_hv)]
 
-regions = parse_iomem(pcidevices)
+(regions, dmar_regions) = parse_iomem(pcidevices)
 ourmem = parse_kernel_cmdline()
 total = hvmem[1] + inmatemem
 
@@ -1052,7 +1093,8 @@ ioapics = parse_madt()
 
 vendor = get_cpu_vendor()
 if vendor == 'GenuineIntel':
-    (iommu_units, extra_memregs) = parse_dmar(pcidevices, ioapics)
+    (iommu_units, extra_memregs) = parse_dmar(pcidevices, ioapics,
+                                              dmar_regions)
 else:
     (iommu_units, extra_memregs) = parse_ivrs(pcidevices, ioapics)
 regions += extra_memregs

diff --git a/tools/root-cell-config.c.tmpl b/tools/root-cell-config.c.tmpl
index 631fcd9d014e557071f7d047939d8045c6bced36..1e5f9f1079b8e5169582f72754873405b8ac5b7c 100644 (file)
--- a/tools/root-cell-config.c.tmpl
+++ b/tools/root-cell-config.c.tmpl
@@ -68,9 +68,10 @@ struct {
                        .pm_timer_address = ${hex(pm_timer_base)},
                        % if iommu_units:
                        .iommu_units = {
-                               % for d in iommu_units:
+                               % for unit in iommu_units:
                                {
-                                       .base = ${hex(d)},
+                                       .base = ${hex(unit.base_addr)},
+                                       .size = ${hex(unit.mmio_size)},
                                },
                                % endfor
                        },