x86/iommu: PVH Dom0 workarounds for missing RMRR entries
v5 cover x86/iommu: PVH Dom0 workarounds for missing RMRR entries
Hello,
The following series implement a workaround for missing RMRR entries for a PVH Dom0. It's based on the iommu_inclusive_mapping VTd option.
The PVH workaround identity maps all regions marked as reserved in the memory map.
Note that this workaround is enabled by default on Intel hardware. It's also available to AMD hardware, although it's disabled by default in that case.
The series can be found at:
git://xenbits.xen.org/people/royger/xen.git iommu_inclusive_v5
Thanks.
Roger Pau Monne (8):
iommu: rename iommu_dom0_strict and iommu_passthrough
iommu: introduce dom0-iommu option
iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
mm: introduce a helper to get the memory type of a page x86/iommu: switch the hwdom mapping function to use page_get_type dom0/pvh: change the order of the MMCFG initialization
vpci: introduce a helper to check MMCFG ranges
x86/iommu: add map-reserved dom0-iommu option to map reserved memory
ranges
docs/misc/xen-command-line.markdown | 39 ++++++++ xen/arch/arm/mm.c | 6 ++ xen/arch/x86/hvm/dom0_build.c | 9 +- xen/arch/x86/hvm/io.c | 5 + xen/arch/x86/mm.c | 34 +++++++ xen/arch/x86/x86_64/mm.c | 3 +- xen/drivers/passthrough/amd/iommu_init.c | 2 +- xen/drivers/passthrough/amd/pci_amd_iommu.c | 11 ++- xen/drivers/passthrough/arm/iommu.c | 4 + xen/drivers/passthrough/arm/smmu.c | 3 + xen/drivers/passthrough/iommu.c | 68 ++++++++++--- xen/drivers/passthrough/vtd/extern.h | 2 - xen/drivers/passthrough/vtd/iommu.c | 25 ++--- xen/drivers/passthrough/vtd/x86/vtd.c | 58 +---------- xen/drivers/passthrough/x86/iommu.c | 103 ++++++++++++++++++++ xen/include/asm-x86/hvm/io.h | 3 + xen/include/xen/iommu.h | 8 +- xen/include/xen/mm.h | 2 + 18 files changed, 293 insertions(+), 92 deletions(-)
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 1/8 iommu: rename iommu_dom0_strict and iommu_passthrough
To iommu_hwdom_strict and iommu_hwdom_passthrough which is more descriptive of their usage. Also change their type from bool_t to bool.
No functional change.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Brian Woods <[email protected]>
Cc: Kevin Tian <[email protected]>
---
Changes since v4:
- New in this version.
--- xen/arch/x86/x86_64/mm.c | 3 ++- xen/drivers/passthrough/amd/iommu_init.c | 2 +- xen/drivers/passthrough/amd/pci_amd_iommu.c | 4 +-- xen/drivers/passthrough/iommu.c | 27 +++++++++++---------- xen/drivers/passthrough/vtd/iommu.c | 16 ++++++------ xen/drivers/passthrough/vtd/x86/vtd.c | 2 +- xen/include/xen/iommu.h | 6 +++-- 7 files changed, 32 insertions(+), 28 deletions(-)
xen/arch/x86/x86_64/mm.c +2 -1
@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)if ( ret ) goto destroy_m2p; -if ( iommu_enabled && !iommu_passthrough && !need_iommu (hardware_domain) ) +if ( iommu_enabled && !iommu_hwdom_passthrough && + !need_iommu (hardware_domain) ) {for ( i = spfn; i < epfn; i++ )if (iommu_map_page (hardware_domain, i, i, IOMMUF_readable|IOMMUF_writable) )
xen/drivers/passthrough/amd/iommu_init.c +1 -1
@@ -1062,7 +1062,7 @@ static void __init amd_iommu_init_cleanup(void)radix_tree_destroy (&ivrs_maps, xfree); iommu_enabled =0 ; - iommu_passthrough =0 ; + iommu_hwdom_passthrough = false; iommu_intremap =0 ; iommuv2_enabled =0 ; }
xen/drivers/passthrough/amd/pci_amd_iommu.c +2 -2
@@ -121,7 +121,7 @@ static void amd_iommu_setup_domain_device(BUG_ON ( !hd->arch.root_table || !hd->arch.paging_mode || !iommu->dev_table.buffer ); -if ( iommu_passthrough &&is_hardware_domain (domain) ) +if ( iommu_hwdom_passthrough &&is_hardware_domain (domain) ) valid =0 ;if ( ats_enabled )
@@ -256,7 +256,7 @@ static void __hwdom_init amd_iommu_hwdom_init(struct domain *d)if (allocate_domain_resources (dom_iommu (d)) )BUG (); -if ( !iommu_passthrough && !need_iommu (d) ) +if ( !iommu_hwdom_passthrough && !need_iommu (d) ) {int rc =0 ;
xen/drivers/passthrough/iommu.c +14 -13
@@ -52,15 +52,16 @@ custom_param("iommu", parse_iommu_param);bool_t __initdata iommu_enable =1 ;bool_t __read_mostly iommu_enabled;bool_t __read_mostly force_iommu; -bool_t __hwdom_initdata iommu_dom0_strict;bool_t __read_mostly iommu_verbose;bool_t __read_mostly iommu_workaround_bios_bug;bool_t __read_mostly iommu_igfx =1 ; -bool_t __read_mostly iommu_passthrough;bool_t __read_mostly iommu_snoop =1 ;bool_t __read_mostly iommu_qinval =1 ;bool_t __read_mostly iommu_intremap =1 ; +bool __hwdom_initdata iommu_hwdom_strict; +bool __read_mostly iommu_hwdom_passthrough; +/* * In the current implementation of VT-d posted interrupts, in some extreme * cases, the per cpu list which saves the blocked vCPU will be very long,
@@ -121,9 +122,9 @@ static int __init parse_iommu_param(const char *s)else if ( !strncmp (s,"amd-iommu-perdev-intremap" , ss - s) ) amd_iommu_perdev_intremap = val;else if ( !strncmp (s,"dom0-passthrough" , ss - s) ) - iommu_passthrough = val; + iommu_hwdom_passthrough = val;else if ( !strncmp (s,"dom0-strict" , ss - s) ) - iommu_dom0_strict = val; + iommu_hwdom_strict = val;else if ( !strncmp (s,"sharept" , ss - s) ) iommu_hap_pt_share = val;else
@@ -158,11 +159,11 @@ static void __hwdom_init check_hwdom_reqs(struct domain *d)arch_iommu_check_autotranslated_hwdom (d); -if ( iommu_passthrough ) +if ( iommu_hwdom_passthrough )panic ("Dom0 uses paging translated mode, dom0-passthrough must not be " "enabled\n" ); - iommu_dom0_strict =1 ; + iommu_hwdom_strict = true; } void __hwdom_init iommu_hwdom_init(struct domain *d)
@@ -175,7 +176,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)return ;register_keyhandler ('o' , &iommu_dump_p2m_table,"dump iommu p2m table" ,0 ); - d->need_iommu = !!iommu_dom0_strict; + d->need_iommu = !!iommu_hwdom_strict;if (need_iommu (d) && !iommu_use_hap_pt (d) ) {struct page_info *page;
@@ -371,8 +372,8 @@ int __init iommu_setup(void)int rc = -ENODEV ;bool_t force_intremap = force_iommu && iommu_intremap; -if ( iommu_dom0_strict ) - iommu_passthrough =0 ; +if ( iommu_hwdom_strict ) + iommu_hwdom_passthrough = false;if ( iommu_enable ) {
@@ -393,15 +394,15 @@ int __init iommu_setup(void)if ( !iommu_enabled ) { iommu_snoop =0 ; - iommu_passthrough =0 ; - iommu_dom0_strict =0 ; + iommu_hwdom_passthrough = false; + iommu_hwdom_strict = false; }printk ("I/O virtualisation %sabled\n" , iommu_enabled ?"en" :"dis" );if ( iommu_enabled ) {printk (" - Dom0 mode: %s\n" , - iommu_passthrough ?"Passthrough" : - iommu_dom0_strict ?"Strict" :"Relaxed" ); + iommu_hwdom_passthrough ?"Passthrough" : + iommu_hwdom_strict ?"Strict" :"Relaxed" );printk ("Interrupt remapping %sabled\n" , iommu_intremap ?"en" :"dis" );tasklet_init (&iommu_pt_cleanup_tasklet, iommu_free_pagetables,0 ); }
xen/drivers/passthrough/vtd/iommu.c +8 -8
@@ -1304,7 +1304,7 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) {struct acpi_drhd_unit *drhd; -if ( !iommu_passthrough &&is_pv_domain (d) ) +if ( !iommu_hwdom_passthrough &&is_pv_domain (d) ) {/* Set up 1:1 page table for hardware domain. */ vtd_set_hwdom_mapping (d);
@@ -1391,7 +1391,7 @@ int domain_context_mapping_one(return res; } -if ( iommu_passthrough &&is_hardware_domain (domain) ) +if ( iommu_hwdom_passthrough &&is_hardware_domain (domain) ) {context_set_translation_type (*context,CONTEXT_TT_PASS_THRU ); agaw =level_to_agaw (iommu->nr_pt_levels);
@@ -1781,7 +1781,7 @@ static int __must_check intel_iommu_map_page(struct domain *d,return 0 ;/* Do nothing if hardware domain and iommu supports pass thru. */ -if ( iommu_passthrough &&is_hardware_domain (d) ) +if ( iommu_hwdom_passthrough &&is_hardware_domain (d) )return 0 ;spin_lock (&hd->arch.mapping_lock);
@@ -1826,7 +1826,7 @@ static int __must_check intel_iommu_unmap_page(struct domain *d, unsigned long gfn) {/* Do nothing if hardware domain and iommu supports pass thru. */ -if ( iommu_passthrough &&is_hardware_domain (d) ) +if ( iommu_hwdom_passthrough &&is_hardware_domain (d) )return 0 ;return dma_pte_clear_one (d, (paddr_t )gfn <<PAGE_SHIFT_4K );
@@ -2269,8 +2269,8 @@ int __init intel_vtd_setup(void)if ( iommu_snoop && !ecap_snp_ctl (iommu->ecap) ) iommu_snoop =0 ; -if ( iommu_passthrough && !ecap_pass_thru (iommu->ecap) ) - iommu_passthrough =0 ; +if ( iommu_hwdom_passthrough && !ecap_pass_thru (iommu->ecap) ) + iommu_hwdom_passthrough = false;if ( iommu_qinval && !ecap_queued_inval (iommu->ecap) ) iommu_qinval =0 ;
@@ -2308,7 +2308,7 @@ int __init intel_vtd_setup(void)#define P (p,s) printk("Intel VT-d %s %senabled.\n", s, (p)? "" : "not ")P (iommu_snoop,"Snoop Control" ); -P (iommu_passthrough,"Dom0 DMA Passthrough" ); +P (iommu_hwdom_passthrough,"Dom0 DMA Passthrough" );P (iommu_qinval,"Queued Invalidation" );P (iommu_intremap,"Interrupt Remapping" );P (iommu_intpost,"Posted Interrupt" );
@@ -2330,7 +2330,7 @@ int __init intel_vtd_setup(void) error: iommu_enabled =0 ; iommu_snoop =0 ; - iommu_passthrough =0 ; + iommu_hwdom_passthrough = false; iommu_qinval =0 ; iommu_intremap =0 ; iommu_intpost =0 ;
xen/drivers/passthrough/vtd/x86/vtd.c +1 -1
@@ -148,7 +148,7 @@ void __hwdom_init vtd_set_hwdom_mapping(struct domain *d)* If dom0-strict mode is enabled then exclude conventional RAM * and let the common code map dom0's pages. */ -if ( iommu_dom0_strict && +if ( iommu_hwdom_strict &&page_is_ram_type (pfn,RAM_TYPE_CONVENTIONAL ) )continue ;
xen/include/xen/iommu.h +4 -2
@@ -29,13 +29,15 @@#include <asm/iommu.h> extern bool_t iommu_enable, iommu_enabled; -extern bool_t force_iommu, iommu_dom0_strict, iommu_verbose; -extern bool_t iommu_workaround_bios_bug, iommu_igfx, iommu_passthrough; +extern bool_t force_iommu, iommu_verbose; +extern bool_t iommu_workaround_bios_bug, iommu_igfx;extern bool_t iommu_snoop, iommu_qinval, iommu_intremap, iommu_intpost;extern bool_t iommu_hap_pt_share;extern bool_t iommu_debug;extern bool_t amd_iommu_perdev_intremap; +extern bool iommu_hwdom_strict, iommu_hwdom_passthrough; +extern unsigned int iommu_dev_iotlb_timeout;int iommu_setup (void );
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 1/8] iommu: rename iommu_dom0_strict and iommu_passthrough
Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
To
iommu_hwdom_strictandiommu_hwdom_passthroughwhich is more descriptive of their usage. Also change their type frombool_tto bool.No functional change.
Signed-off-by Roger Pau Monné <[email protected]>
with one remark:
xen/drivers/passthrough/iommu.c void __hwdom_init iommu_hwdom_init(struct domain *d) · line 179
Quoted 6 lines
xen/drivers/passthrough/iommu.c +1
-1@@ -175,7 +176,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)return ;register_keyhandler ('o' , &iommu_dump_p2m_table,"dump iommu p2m table" ,0 ); - d->need_iommu = !!iommu_dom0_strict; + d->need_iommu = !!iommu_hwdom_strict;
No need for the !! anymore. Easy enough to adjust while committing.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 1/8] iommu: rename iommu_dom0_strict and iommu_passthrough
In reply to Roger Pau MonneQuoted 10 lines
From: Roger Pau Monne [mailto:[email protected]]
Sent: Tuesday, August 14, 2018 9:43 PMTo
iommu_hwdom_strictandiommu_hwdom_passthroughwhich is more descriptive of their usage. Also change their type frombool_tto bool.No functional change.
Signed-off-by Roger Pau Monné <[email protected]>
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 2/8 iommu: introduce dom0-iommu option
In reply to Roger Pau MonneTo select the iommu configuration used by Dom0. This option supersedes iommu=dom0-strict|dom0-passthrough.
---
Changes since v4:
- Move the iommu_dom0_* variable renaming to a previous patch.
- Use dom0-iommu=passthrough|strict booleans, like the iommu option used.
Changes since v2:
- Change the style and text used in the xen command line document.
- Don't allow none, strict or relaxed to use the no- prefix.
Changes since v1:
- New in this version.
---
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
--- docs/misc/xen-command-line.markdown | 19 +++++++++++++++++++ xen/drivers/passthrough/iommu.c | 26 ++++++++++++++++++++++++++ 2 files changed, 45 insertions(+)
docs/misc/xen-command-line.markdown +19 -0
@@ -681,6 +681,19 @@ Flag that makes a dom0 boot in PVHv2 mode. Flag that makes a dom0 use shadow paging. Only works when "pvh" is enabled. +### dom0-iommu +> `= List of [ passthrough | strict ]` + +This list of booleans control the iommu usage by Dom0: + +* `passthrough`: disables DMA remapping for Dom0. Default is `false`. + +* `strict`: sets up DMA remapping only for the RAM Dom0 actually got assigned. + Default is `false` which means Dom0 will get mappings for all the host + RAM except regions in use by Xen. Note that this option is hard coded to + `true` for a PVH Dom0 and any attempt to overwrite it from the command line + is ignored. + ### dom0\_ioports\_disable (x86) > `= List of <hex>-<hex>`
@@ -1150,12 +1163,18 @@ detection of systems known to misbehave upon accesses to that port. > `dom0-passthrough` +> **WARNING: This command line option is deprecated, and superseded by +> _dom0-iommu=passthrough_ - using both options in combination is undefined.** + > Default: `false` >> Control whether to disable DMA remapping for Dom0. > `dom0-strict` +> **WARNING: This command line option is deprecated, and superseded by +> _dom0-iommu=strict_ - using both options in combination is undefined.** + > Default: `false` >> Control whether to set up DMA remapping only for the memory Dom0 actually
xen/drivers/passthrough/iommu.c +26 -0
@@ -136,6 +136,32 @@ static int __init parse_iommu_param(const char *s)return rc; } +static int __init parse_dom0_iommu_param (const char *s) +{ +const char *ss; +int rc =0 ; + +do { +int val; + + ss =strchr (s,',' ); +if ( !ss ) + ss =strchr (s,'\0' ); + +if ( (val =parse_boolean ("passthrough" , s, ss)) >=0 ) + iommu_hwdom_passthrough = val; +else if ( (val =parse_boolean ("strict" , s, ss)) >=0 ) + iommu_hwdom_strict = val; +else + rc = -EINVAL ; + + s = ss +1 ; + }while ( *ss ); + +return rc; +} +custom_param ("dom0-iommu" , parse_dom0_iommu_param); +int iommu_domain_init (struct domain *d) {struct domain_iommu *hd =dom_iommu (d);
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 2/8] iommu: introduce dom0-iommu option
docs/misc/xen-command-line.markdown Flag that makes a dom0 boot in PVHv2 mode. · line 687
Quoted 11 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
docs/misc/xen-command-line.markdown +4
-0@@ -681,6 +681,19 @@ Flag that makes a dom0 boot in PVHv2 mode. Flag that makes a dom0 use shadow paging. Only works when "pvh" is enabled. +### dom0-iommu +> `= List of [ passthrough | strict ]` + +This list of booleans control the iommu usage by Dom0:
... controls ...
docs/misc/xen-command-line.markdown Flag that makes a dom0 boot in PVHv2 mode. · line 689
Quoted 1 line
docs/misc/xen-command-line.markdown +1
-0+* `passthrough`: disables DMA remapping for Dom0. Default is `false`.
Mind also stating that this must not be enabled for PVH Dom0? Or perhaps ...
docs/misc/xen-command-line.markdown Flag that makes a dom0 boot in PVHv2 mode. · line 695
Quoted 5 lines
docs/misc/xen-command-line.markdown +5
-0+* `strict`: sets up DMA remapping only for the RAM Dom0 actually got assigned. + Default is `false` which means Dom0 will get mappings for all the host + RAM except regions in use by Xen. Note that this option is hard coded to + `true` for a PVH Dom0 and any attempt to overwrite it from the command line + is ignored.
... it could be made act like this, ignoring the request instead of panic()ing?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 3/8 iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
In reply to Roger Pau MonneIntroduce a new dom0-iommu=map-inclusive generic option that supersedes iommu_inclusive_mapping. The previous behavior is preserved and the option should only be enabled by default on Intel hardware.
---
Changes since v4:
- Use an if to set the default option value.
- Set the option to false unconditionally on ARM.
Changes since v2:
- Fix typo in commit message.
- Change style and text of the documentation in xen command line.
- Set the defaults in {intel/amd}_iommu_hwdom_init for inclusive.
- Re-add the iommu_dom0_passthrough || !is_pv_domain(d) check.
Changes since v1:
- Use dom0-iommu instead of the iommu option.
- Only enable by default on Intel hardware.
---
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Brian Woods <[email protected]>
Cc: Kevin Tian <[email protected]>
--- docs/misc/xen-command-line.markdown | 13 ++++- xen/drivers/passthrough/amd/pci_amd_iommu.c | 4 ++ xen/drivers/passthrough/arm/iommu.c | 4 ++ xen/drivers/passthrough/arm/smmu.c | 2 + xen/drivers/passthrough/iommu.c | 12 +++++ xen/drivers/passthrough/vtd/extern.h | 2 - xen/drivers/passthrough/vtd/iommu.c | 8 ++- xen/drivers/passthrough/vtd/x86/vtd.c | 58 +------------------- xen/drivers/passthrough/x86/iommu.c | 59 +++++++++++++++++++++ xen/include/xen/iommu.h | 2 + 10 files changed, 99 insertions(+), 65 deletions(-)
docs/misc/xen-command-line.markdown +12 -1
@@ -682,7 +682,7 @@ Flag that makes a dom0 use shadow paging. Only works when "pvh" is enabled. ### dom0-iommu -> `= List of [ passthrough | strict ]` +> `= List of [ passthrough | strict | map-inclusive ]` This list of booleans control the iommu usage by Dom0:
@@ -694,6 +694,14 @@ This list of booleans control the iommu usage by Dom0: `true` for a PVH Dom0 and any attempt to overwrite it from the command line is ignored. +* `map-inclusive`: sets up DMA remapping for all the non-RAM regions below 4GB + except for unusable ranges. Use this to work around firmware issues providing + incorrect RMRR/IVMD entries. Rather than only mapping RAM pages for IOMMU + accesses for Dom0, with this option all pages up to 4GB, not marked as + unusable in the E820 table, will get a mapping established. Note that this + option is only applicable to a PV Dom0 and is enabled by default on Intel + hardware. + ### dom0\_ioports\_disable (x86) > `= List of <hex>-<hex>`
@@ -1229,6 +1237,9 @@ wait descriptor timed out', try increasing this value. ### iommu\_inclusive\_mapping (VT-d) > `= <boolean>` +**WARNING: This command line option is deprecated, and superseded by +_dom0-iommu=map-inclusive_ - using both options in combination is undefined.** + > Default: `true` Use this to work around firmware issues providing incorrect RMRR entries.
xen/drivers/passthrough/amd/pci_amd_iommu.c +4 -0
@@ -253,6 +253,10 @@ static void __hwdom_init amd_iommu_hwdom_init(struct domain *d)unsigned long i;const struct amd_iommu *iommu; +/* Inclusive IOMMU mappings are disabled by default on AMD hardware. */ +if ( iommu_hwdom_inclusive ==-1 ) + iommu_hwdom_inclusive = false; +if (allocate_domain_resources (dom_iommu (d)) )BUG ();
xen/drivers/passthrough/arm/iommu.c +4 -0
@@ -73,3 +73,7 @@ int arch_iommu_populate_page_table(struct domain *d)/* The IOMMU shares the p2m with the CPU */ return -ENOSYS ; } + +void __hwdom_init arch_iommu_hwdom_init(struct domain *d) +{ +}
xen/drivers/passthrough/arm/smmu.c +2 -0
@@ -2727,6 +2727,8 @@ static int arm_smmu_iommu_domain_init(struct domain *d) static void __hwdom_init arm_smmu_iommu_hwdom_init(struct domain *d) { +/* Set to false options not supported on ARM. */ + iommu_hwdom_inclusive = false; }static void arm_smmu_iommu_domain_teardown (struct domain *d)
xen/drivers/passthrough/iommu.c +12 -0
@@ -61,6 +61,7 @@ bool_t __read_mostly iommu_intremap = 1; bool __hwdom_initdata iommu_hwdom_strict;bool __read_mostly iommu_hwdom_passthrough; +int8_t __hwdom_initdata iommu_hwdom_inclusive = -1;/* * In the current implementation of VT-d posted interrupts, in some extreme
@@ -152,6 +153,8 @@ static int __init parse_dom0_iommu_param(const char *s) iommu_hwdom_passthrough = val; else if ( (val = parse_boolean("strict", s, ss)) >= 0 ) iommu_hwdom_strict = val; + else if ( (val = parse_boolean("map-inclusive", s, ss)) >= 0 ) + iommu_hwdom_inclusive = val; else rc = -EINVAL ;
@@ -235,6 +238,15 @@ void __hwdom_init iommu_hwdom_init(struct domain *d) } hd->platform_ops->hwdom_init (d); + +if ( iommu_hwdom_inclusive && !is_pv_domain (d) ) + { +printk (XENLOG_WARNING +"IOMMU inclusive mappings are only supported on PV Dom0\n" ); + iommu_hwdom_inclusive = false; + } + +arch_iommu_hwdom_init (d); }void iommu_teardown (struct domain *d)
xen/drivers/passthrough/vtd/extern.h +0 -2
@@ -99,6 +99,4 @@ void pci_vtd_quirk(const struct pci_dev *);bool_t platform_supports_intremap (void );bool_t platform_supports_x2apic (void ); -void vtd_set_hwdom_mapping (struct domain *d); -#endif // _VTD_EXTERN_H_
xen/drivers/passthrough/vtd/iommu.c +3 -5
@@ -1304,11 +1304,9 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) {struct acpi_drhd_unit *drhd; -if ( !iommu_hwdom_passthrough &&is_pv_domain (d) ) - { -/* Set up 1:1 page table for hardware domain. */ -vtd_set_hwdom_mapping (d); - } +/* Inclusive mappings are enabled by default on Intel hardware for PV. */ +if ( iommu_hwdom_inclusive ==-1 ) + iommu_hwdom_inclusive =is_pv_domain (d);setup_hwdom_pci_devices (d, setup_hwdom_device);setup_hwdom_rmrr (d);
xen/drivers/passthrough/vtd/x86/vtd.c +1 -57
73 lines not on this page · raw message
xen/drivers/passthrough/x86/iommu.c +59 -0
71 lines not on this page · raw message
xen/include/xen/iommu.h +2 -0
@@ -37,6 +37,7 @@ extern bool_t iommu_debug;extern bool_t amd_iommu_perdev_intremap;extern bool iommu_hwdom_strict, iommu_hwdom_passthrough; +extern int8_t iommu_hwdom_inclusive;extern unsigned int iommu_dev_iotlb_timeout;
@@ -51,6 +52,7 @@ void arch_iommu_domain_destroy(struct domain *d);int arch_iommu_domain_init (struct domain *d);int arch_iommu_populate_page_table (struct domain *d);void arch_iommu_check_autotranslated_hwdom (struct domain *d); +void arch_iommu_hwdom_init (struct domain *d);int iommu_construct (struct domain *d);
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 3/8] iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/vtd/iommu.c +0
-0@@ -1304,11 +1304,9 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) {
xen/drivers/passthrough/vtd/iommu.c static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) · line 1309
Quoted 10 lines
xen/drivers/passthrough/vtd/iommu.c +3
-5struct acpi_drhd_unit *drhd; -if ( !iommu_hwdom_passthrough &&is_pv_domain (d) ) - { -/* Set up 1:1 page table for hardware domain. */ -vtd_set_hwdom_mapping (d); - } +/* Inclusive mappings are enabled by default on Intel hardware for PV. */ +if ( iommu_hwdom_inclusive ==-1 ) + iommu_hwdom_inclusive =is_pv_domain (d);
Hmm, I didn't notice this before, but there's an issue here for the late-hwdom case: What if Dom0 and the actual hardware domain differ in their PV-ness? I think you need to keep it at -1 here and take the still -1 value into account in arch_iommu_hwdom_init(). iommu_hwdom_init() then also should not override it.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 3/8] iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
Quoted 7 lines · Jan Beulich
Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/vtd/iommu.c +0
-0@@ -1304,11 +1304,9 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) {struct acpi_drhd_unit *drhd;
Quoted 14 lines
xen/drivers/passthrough/vtd/iommu.c static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) · line 1309
Quoted 8 lines
xen/drivers/passthrough/vtd/iommu.c +3
-5-if ( !iommu_hwdom_passthrough &&is_pv_domain (d) ) - { -/* Set up 1:1 page table for hardware domain. */ -vtd_set_hwdom_mapping (d); - } +/* Inclusive mappings are enabled by default on Intel hardware for PV. */ +if ( iommu_hwdom_inclusive ==-1 ) + iommu_hwdom_inclusive =is_pv_domain (d);Hmm, I didn't notice this before, but there's an issue here for the late-hwdom case: What if Dom0 and the actual hardware domain differ in their PV-ness? I think you need to keep it at -1 here and take the still -1 value into account in
arch_iommu_hwdom_init().iommu_hwdom_init()then also should not override it.
But isn't the late-hwdom the one that would be passed when calling intel_iommu_hwdom_init?
Or else how is intel_iommu_hwdom_init going to differentiate between Dom0 and a late-hwdom if the function is called for both?
Thanks, Roger.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Jan Beulich
Reviewed-byRe: [PATCH v5 3/8] iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
Quoted 24 lines
Quoted 1 line
Quoted 1 line · On 17.08.18 at 12:32, <[email protected]> wrote:
Quoted 7 lines · Jan Beulich
Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/vtd/iommu.c +0
-0@@ -1304,11 +1304,9 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) {struct acpi_drhd_unit *drhd;Quoted 14 lines
xen/drivers/passthrough/vtd/iommu.c static void __hwdom_init intel_iommu_hwdom_init(struct domain *d) · line 1309
Quoted 8 lines
xen/drivers/passthrough/vtd/iommu.c +3
-5-if ( !iommu_hwdom_passthrough &&is_pv_domain (d) ) - { -/* Set up 1:1 page table for hardware domain. */ -vtd_set_hwdom_mapping (d); - } +/* Inclusive mappings are enabled by default on Intel hardware for PV. */ +if ( iommu_hwdom_inclusive ==-1 ) + iommu_hwdom_inclusive =is_pv_domain (d);Hmm, I didn't notice this before, but there's an issue here for the late-hwdom case: What if Dom0 and the actual hardware domain differ in their PV-ness? I think you need to keep it at -1 here and take the still -1 value into account in
arch_iommu_hwdom_init().iommu_hwdom_init()then also should not override it.
Quoted 5 lines
But isn't the late-hwdom the one that would be passed when calling
intel_iommu_hwdom_init?Or else how is
intel_iommu_hwdom_initgoing to differentiate between Dom0 and a late-hwdom if the function is called for both?
Oh, you're right - I was under the wrong impression that the function would be called for both Dom0 and hwdom in this case (and would act identically, on the assumption that Dom0 is going to go away soon anyway). As a result
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 3/8] iommu: make iommu_inclusive_mapping a suboption of dom0-iommu
In reply to Roger Pau MonneQuoted 9 lines
From: Roger Pau Monne [mailto:[email protected]]
Sent: Tuesday, August 14, 2018 9:44 PMIntroduce a new dom0-iommu=map-inclusive generic option that supersedes
iommu_inclusive_mapping. The previous behavior is preserved and the option should only be enabled by default on Intel hardware.Signed-off-by Roger Pau Monné <[email protected]>Reviewed-by Paul Durrant <[email protected]>
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 4/8 mm: introduce a helper to get the memory type of a page
In reply to Roger Pau MonneThe type is only returned if the full page has the same type. This function is unimplemented for ARM.
---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
--- xen/arch/arm/mm.c | 6 ++++++ xen/arch/x86/mm.c | 34 ++++++++++++++++++++++++++++++++++ xen/include/xen/mm.h | 2 ++ 3 files changed, 42 insertions(+)
xen/arch/arm/mm.c +6 -0
@@ -1181,6 +1181,12 @@ int page_is_ram_type(unsigned long mfn, unsigned long mem_type)return 0 ; } +int page_get_type (unsigned long mfn) +{ +ASSERT_UNREACHABLE (); +return -1 ; +} +unsigned long domain_get_maximum_gpfn (struct domain *d) {return gfn_x (d->arch.p2m.max_mapped_gfn);
xen/arch/x86/mm.c +34 -0
@@ -430,6 +430,40 @@ int page_is_ram_type(unsigned long mfn, unsigned long mem_type)return 0 ; } +int page_get_type (unsigned long mfn) +{ +uint64_t maddr =pfn_to_paddr (mfn); +unsigned int i; + +for ( i =0 ; i < e820.nr_map; i++ ) + { +/* Test the range. */ +if ( (e820.map[i].addr <= maddr) && + ((e820.map[i].addr + e820.map[i].size) >= (maddr +PAGE_SIZE )) ) +switch ( e820.map[i].type ) + { +case E820_RAM : +return RAM_TYPE_CONVENTIONAL ; + +case E820_RESERVED : +return RAM_TYPE_RESERVED ; + +case E820_UNUSABLE : +return RAM_TYPE_UNUSABLE ; + +case E820_ACPI : +case E820_NVS : +return RAM_TYPE_ACPI ; + +default : +ASSERT_UNREACHABLE (); +return -1 ; + } + } + +return -1 ; +} +unsigned long domain_get_maximum_gpfn (struct domain *d) {if (is_hvm_domain (d) )
xen/include/xen/mm.h +2 -0
@@ -597,6 +597,8 @@ int __must_check donate_page(struct domain *d, struct page_info *page,#define RAM_TYPE_ACPI 0x00000008/* TRUE if the whole page at @mfn is of the requested RAM type(s) above. */ int page_is_ram_type (unsigned long mfn,unsigned long mem_type); +/* Returns the page type if the whole page is of the same type. */ +int page_get_type (unsigned long mfn);/* Prepare/destroy a ring for a dom0 helper. Helper with talk * with Xen on behalf of this domain. */
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 4/8] mm: introduce a helper to get the memory type of a page
xen/arch/arm/mm.c int page_is_ram_type(unsigned long mfn, unsigned long mem_type) · line 1188
Quoted 12 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/arch/arm/mm.c +5
-0@@ -1181,6 +1181,12 @@ int page_is_ram_type(unsigned long mfn, unsigned long mem_type)return 0 ; } +int page_get_type (unsigned long mfn) +{ +ASSERT_UNREACHABLE (); +return -1 ; +}
With the assertion, why is the function needed in the first place?
xen/arch/x86/mm.c int page_is_ram_type(unsigned long mfn, unsigned long mem_type) · line 433
Quoted 7 lines
xen/arch/x86/mm.c +1
-0@@ -430,6 +430,40 @@ int page_is_ram_type(unsigned long mfn, unsigned long mem_type)return 0 ; } +int page_get_type (unsigned long mfn)
Considering we already have get_page_type(), this function name needs to change. At the very least page_get_ram_type(), looking at the set of values it may return.
xen/arch/x86/mm.c int page_is_ram_type(unsigned long mfn, unsigned long mem_type) · line 442
Quoted 9 lines
xen/arch/x86/mm.c +9
-0+{ +uint64_t maddr =pfn_to_paddr (mfn); +unsigned int i; + +for ( i =0 ; i < e820.nr_map; i++ ) + { +/* Test the range. */ +if ( (e820.map[i].addr <= maddr) && + ((e820.map[i].addr + e820.map[i].size) >= (maddr +PAGE_SIZE )) )
Would you mind inverting the condition and using continue, both to reduce indentation of the switch and to make it very clear that no braces are needed / wanted?
Quoted 13 lines
xen/arch/x86/mm.c +13
-0+switch ( e820.map[i].type ) + { +case E820_RAM : +return RAM_TYPE_CONVENTIONAL ; + +case E820_RESERVED : +return RAM_TYPE_RESERVED ; + +case E820_UNUSABLE : +return RAM_TYPE_UNUSABLE ; + +case E820_ACPI : +case E820_NVS :
xen/arch/x86/mm.c int page_is_ram_type(unsigned long mfn, unsigned long mem_type) · line 465
Quoted 10 lines
xen/arch/x86/mm.c +10
-0+return RAM_TYPE_ACPI ; + +default : +ASSERT_UNREACHABLE (); +return -1 ; + } + } + +return -1 ; +}
One more case to consider: What about a page part of which is a given type, and the other part of which is simply missing from the E820 table? I'm uncertain whether in that case it might be a good idea in general to report it as having the given type; for the specific purpose you want the function for, that would imo be quite helpful.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 4/8] mm: introduce a helper to get the memory type of a page
Quoted 24 lines
Quoted 24 lines
Quoted 2 lines · On 17.08.18 at 12:17, <[email protected]> wrote:
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/arch/x86/mm.c +22
-0+switch ( e820.map[i].type ) + { +case E820_RAM : +return RAM_TYPE_CONVENTIONAL ; + +case E820_RESERVED : +return RAM_TYPE_RESERVED ; + +case E820_UNUSABLE : +return RAM_TYPE_UNUSABLE ; + +case E820_ACPI : +case E820_NVS : +return RAM_TYPE_ACPI ; + +default : +ASSERT_UNREACHABLE (); +return -1 ; + } + } + +return -1 ;
Quoted 8 lines
Quoted 1 line
+}
One more case to consider: What about a page part of which is a given type, and the other part of which is simply missing from the E820 table? I'm uncertain whether in that case it might be a good idea in general to report it as having the given type; for the specific purpose you want the function for, that would imo be quite helpful.
Considering RAM_TYPE_* are bit masks - perhaps the function should OR together all types found for the requested page, and let the caller go from there? And to account for my earlier remark, add a separate RAM_TYPE_UNKNOWN (and never have the function return -1 or some such; its return type then would better be unsigned int)?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 4/8] mm: introduce a helper to get the memory type of a page
Quoted 33 lines · Jan Beulich
Quoted 24 lines
Quoted 24 lines
Quoted 2 lines · On 17.08.18 at 12:17, <[email protected]> wrote:
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/arch/x86/mm.c +22
-0+switch ( e820.map[i].type ) + { +case E820_RAM : +return RAM_TYPE_CONVENTIONAL ; + +case E820_RESERVED : +return RAM_TYPE_RESERVED ; + +case E820_UNUSABLE : +return RAM_TYPE_UNUSABLE ; + +case E820_ACPI : +case E820_NVS : +return RAM_TYPE_ACPI ; + +default : +ASSERT_UNREACHABLE (); +return -1 ; + } + } + +return -1 ;Quoted 8 lines
Quoted 1 line
+}
One more case to consider: What about a page part of which is a given type, and the other part of which is simply missing from the E820 table? I'm uncertain whether in that case it might be a good idea in general to report it as having the given type; for the specific purpose you want the function for, that would imo be quite helpful.
Quoted 6 lines
Considering RAM_TYPE_* are bit masks - perhaps the function should OR together all types found for the requested page, and let the caller go from there? And to account for my earlier remark, add a separate
RAM_TYPE_UNKNOWN(and never have the function return -1 or some such; its return type then would better be unsigned int)?
I can do something along this lines, but then the functionality of the existing inclusive option is likely to change on boxes having memory types not aligned to a page boundary.
Also, we should likely discuss what to do with a page that for example has both unusable and reserved regions. I would map it in the inclusive case because it has a reserved region, but would like to hear your opinion.
Thanks, Roger.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 4/8] mm: introduce a helper to get the memory type of a page
Quoted 42 lines
Quoted 1 line
Quoted 1 line · On 17.08.18 at 18:37, <[email protected]> wrote:
Quoted 33 lines · Jan Beulich
Quoted 24 lines
Quoted 24 lines
Quoted 2 lines · On 17.08.18 at 12:17, <[email protected]> wrote:
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/arch/x86/mm.c +22
-0+switch ( e820.map[i].type ) + { +case E820_RAM : +return RAM_TYPE_CONVENTIONAL ; + +case E820_RESERVED : +return RAM_TYPE_RESERVED ; + +case E820_UNUSABLE : +return RAM_TYPE_UNUSABLE ; + +case E820_ACPI : +case E820_NVS : +return RAM_TYPE_ACPI ; + +default : +ASSERT_UNREACHABLE (); +return -1 ; + } + } + +return -1 ;Quoted 8 lines
Quoted 1 line
+}
One more case to consider: What about a page part of which is a given type, and the other part of which is simply missing from the E820 table? I'm uncertain whether in that case it might be a good idea in general to report it as having the given type; for the specific purpose you want the function for, that would imo be quite helpful.
Quoted 6 lines
Considering RAM_TYPE_* are bit masks - perhaps the function should OR together all types found for the requested page, and let the caller go from there? And to account for my earlier remark, add a separate
RAM_TYPE_UNKNOWN(and never have the function return -1 or some such; its return type then would better be unsigned int)?
Quoted 3 lines
I can do something along this lines, but then the functionality of the existing inclusive option is likely to change on boxes having memory types not aligned to a page boundary.
That's an expected consequence, yes, and given this is a workaround option anyway I'm afraid we'll have to live with it being unclear whether this would be a change to the better or to the worse.
Quoted 4 lines
Also, we should likely discuss what to do with a page that for example has both unusable and reserved regions. I would map it in the inclusive case because it has a reserved region, but would like to hear your opinion.
Not page-aligned unusable regions are pretty suspicious imo. Along the lines of the above response, I think I'd be fine going either of the two possible routes - we simply can't tell ahead of time which one might be better.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 5/8 x86/iommu: switch the hwdom mapping function to use page_get_type
In reply to Roger Pau MonneThis avoids repeated calls to page_is_ram_type which improves performance and makes the code easier to read.
No functional change.
---
Changes since v4:
- New in this version.
---
Cc: Jan Beulich <[email protected]>
--- xen/drivers/passthrough/x86/iommu.c | 60 +++++++++++++++-------------- 1 file changed, 32 insertions(+), 28 deletions(-)
xen/drivers/passthrough/x86/iommu.c +32 -28
@@ -134,6 +134,37 @@ void arch_iommu_domain_destroy(struct domain *d) { } +static bool __hwdom_init hwdom_iommu_map(const struct domain *d, +unsigned long pfn, +unsigned long max_pfn) +{ +/* +* Set up 1:1 mapping for dom0. Default to include only conventional RAM +* areas and let RMRRs include needed reserved regions. When set, the +* inclusive mapping additionally maps in every pfn up to 4GB except those +* that fall in unusable ranges. +*/ +if ( (pfn > max_pfn && !mfn_valid (_mfn (pfn))) ||xen_in_range (pfn) ) +return false; + +switch (page_get_type (pfn) ) + { +case RAM_TYPE_UNUSABLE : +return false; + +case RAM_TYPE_CONVENTIONAL : +if ( iommu_hwdom_strict ) +return false; +break ; + +default : +if ( !iommu_hwdom_inclusive || pfn > max_pfn ) +return false; + } + +return true; +} + void __hwdom_init arch_iommu_hwdom_init(struct domain *d) {unsigned long i, top, max_pfn;
@@ -149,36 +180,9 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)for ( i =0 ; i < top; i++ ) {unsigned long pfn =pdx_to_pfn (i); -bool map;int rc; -/* -* Set up 1:1 mapping for dom0. Default to include only -* conventional RAM areas and let RMRRs include needed reserved -* regions. When set, the inclusive mapping additionally maps in -* every pfn up to 4GB except those that fall in unusable ranges. -*/ -if ( pfn > max_pfn && !mfn_valid (_mfn (pfn)) ) -continue ; - -if ( iommu_hwdom_inclusive && pfn <= max_pfn ) - map = !page_is_ram_type (pfn,RAM_TYPE_UNUSABLE ); -else - map =page_is_ram_type (pfn,RAM_TYPE_CONVENTIONAL ); - -if ( !map ) -continue ; - -/* Exclude Xen bits */ -if (xen_in_range (pfn) ) -continue ; - -/* -* If dom0-strict mode is enabled then exclude conventional RAM -* and let the common code map dom0's pages. -*/ -if ( iommu_hwdom_strict && -page_is_ram_type (pfn,RAM_TYPE_CONVENTIONAL ) ) +if ( !hwdom_iommu_map (d, pfn, max_pfn) )continue ; rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable);
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 5/8] x86/iommu: switch the hwdom mapping function to use page_get_type
Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
+static bool
__hwdom_inithwdom_iommu_map(const struct domain *d,
Unused function parameter?
Quoted 11 lines
xen/drivers/passthrough/x86/iommu.c +11
-0+ unsigned long pfn, + unsigned long max_pfn) +{ +/* +* Set up 1:1 mapping for dom0. Default to include only conventional RAM +* areas and let RMRRs include needed reserved regions. When set, the +* inclusive mapping additionally maps in every pfn up to 4GB except those +* that fall in unusable ranges. +*/ +if ( (pfn > max_pfn && !mfn_valid (_mfn (pfn))) ||xen_in_range (pfn) ) +return false;
xen/drivers/passthrough/x86/iommu.c void arch_iommu_domain_destroy(struct domain *d) · line 158
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+ +switch (page_get_type (pfn) ) + { +case RAM_TYPE_UNUSABLE : +return false; + +case RAM_TYPE_CONVENTIONAL : +if ( iommu_hwdom_strict ) +return false; +break ;
Simply "return !iommu_hwdom_strict", to make it yet easier to follow what happens in which case? But depending on your thoughts on my remarks on the previous patch, this may change all over again anyway.
xen/drivers/passthrough/x86/iommu.c void arch_iommu_domain_destroy(struct domain *d) · line 162
Quoted 3 lines
xen/drivers/passthrough/x86/iommu.c +3
-0+default : +if ( !iommu_hwdom_inclusive || pfn > max_pfn ) +return false;
And then a plain return statement here too?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 5/8] x86/iommu: switch the hwdom mapping function to use page_get_type
In reply to Roger Pau MonneQuoted 24 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/x86/iommu.c +17
-0@@ -134,6 +134,37 @@ void arch_iommu_domain_destroy(struct domain *d) { } +static bool __hwdom_init hwdom_iommu_map(const struct domain *d, +unsigned long pfn, +unsigned long max_pfn) +{ +/* +* Set up 1:1 mapping for dom0. Default to include only conventional RAM +* areas and let RMRRs include needed reserved regions. When set, the +* inclusive mapping additionally maps in every pfn up to 4GB except those +* that fall in unusable ranges. +*/ +if ( (pfn > max_pfn && !mfn_valid (_mfn (pfn))) ||xen_in_range (pfn) ) +return false; + +switch (page_get_type (pfn) ) + { +case RAM_TYPE_UNUSABLE : +return false;
xen/drivers/passthrough/x86/iommu.c void arch_iommu_domain_destroy(struct domain *d) · line 163
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+ +case RAM_TYPE_CONVENTIONAL : +if ( iommu_hwdom_strict ) +return false; +break ; + +default : +if ( !iommu_hwdom_inclusive || pfn > max_pfn ) +return false; + }
Ah, in patch 8 it becomes clear why you want execution to fall out of the switch() in the "true" case. Please disregard the respective earlier comments then.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 6/8 dom0/pvh: change the order of the MMCFG initialization
In reply to Roger Pau MonneSo it's done before the iommu is initialized. This is required in order to be able to fetch the MMCFG regions from the domain struct.
No functional change.
---
Changes since v1:
- New in this version.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
--- xen/arch/x86/hvm/dom0_build.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
xen/arch/x86/hvm/dom0_build.c +7 -2
@@ -1100,6 +1100,13 @@ int __init dom0_construct_pvh(struct domain *d, const module_t *image,return rc; } +/* +* NB: MMCFG initialization needs to be performed before iommu +* initialization so the iommu code can fetch the MMCFG regions used by the +* domain. +*/ +pvh_setup_mmcfg (d); +iommu_hwdom_init (d); rc = pvh_load_kernel(d, image, image_headroom, initrd, bootstrap_map(image),
@@ -1124,8 +1131,6 @@ int __init dom0_construct_pvh(struct domain *d, const module_t *image,return rc; } -pvh_setup_mmcfg (d); -printk ("WARNING: PVH is an experimental mode with limited functionality\n" );return 0 ; }
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 6/8] dom0/pvh: change the order of the MMCFG initialization
Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
So it's done before the iommu is initialized. This is required in order to be able to fetch the MMCFG regions from the domain struct.
No functional change.
Signed-off-by Roger Pau Monné <[email protected]>
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 7/8 vpci: introduce a helper to check MMCFG ranges
In reply to Roger Pau MonneThe helpers returns whether a given memory address belongs to a domain MMCFG range.
---
Changes since v4:
- New in this version.
---
Cc: Paul Durrant <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
--- xen/arch/x86/hvm/io.c | 5 +++++ xen/include/asm-x86/hvm/io.h | 3 +++ 2 files changed, 8 insertions(+)
xen/arch/x86/hvm/io.c +5 -0
@@ -404,6 +404,11 @@ static const struct hvm_mmcfg *vpci_mmcfg_find(const struct domain *d,return NULL ; } +bool vpci_is_mmcfg_address (const struct domain *d,paddr_t addr) +{ +return vpci_mmcfg_find (d, addr); +} +static unsigned int vpci_mmcfg_decode_addr (const struct hvm_mmcfg *mmcfg,paddr_t addr,pci_sbdf_t *sbdf) {
xen/include/asm-x86/hvm/io.h +3 -0
@@ -180,6 +180,9 @@ int register_vpci_mmcfg_handler(struct domain *d, paddr_t addr,/* Destroy tracked MMCFG areas. */ void destroy_vpci_mmcfg (struct domain *d); +/* Check if an address is between a MMCFG region for a domain. */ +bool vpci_is_mmcfg_address (const struct domain *d,paddr_t addr); +#endif /* __ASM_X86_HVM_IO_H__ */
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 7/8] vpci: introduce a helper to check MMCFG ranges
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
The helpers returns whether a given memory address belongs to a domain MMCFG range.
Signed-off-by Roger Pau Monné <[email protected]>
I'd prefer if this was part of the patch actually using the new function.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v5 8/8 x86/iommu: add map-reserved dom0-iommu option to map reserved memory ranges
In reply to Roger Pau MonneSeveral people have reported hardware issues (malfunctioning USB controllers) due to iommu page faults on Intel hardware. Those faults are caused by missing RMRR (VTd) entries in the ACPI tables. Those can be worked around on VTd hardware by manually adding RMRR entries on the command line, this is however limited to Intel hardware and quite cumbersome to do.
In order to solve those issues add a new dom0-iommu=map-reserved option that identity maps all regions marked as reserved in the memory map. Note that regions used by devices emulated by Xen (LAPIC, IO-APIC or PCIe MCFG regions) are specifically avoided. Note that this option is available to a PVH Dom0 (as opposed to the inclusive option which only works for PV Dom0).
---
Changes since v4:
- Use pfn_to_paddr.
- Rebase on top of previous changes.
- Change the default option setting to use if instead of a ternary operator.
- Rename to map-reserved.
Changes since v3:
- Add mappings if the iommu page tables are shared.
Changes since v2:
- Fix comment regarding dom0-strict.
- Change documentation style of xen command line.
- Rename iommu_map to hwdom_iommu_map.
- Move all the checks to hwdom_iommu_map.
Changes since v1:
- Introduce a new reserved option instead of abusing the inclusive option.
- Use the same helper function for PV and PVH in order to decide if a page should be added to the domain page tables.
- Use the data inside of the domain struct to detect overlaps with emulated MMIO regions.
---
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
--- docs/misc/xen-command-line.markdown | 9 ++++ xen/drivers/passthrough/amd/pci_amd_iommu.c | 3 ++ xen/drivers/passthrough/arm/smmu.c | 1 + xen/drivers/passthrough/iommu.c | 3 ++ xen/drivers/passthrough/vtd/iommu.c | 3 ++ xen/drivers/passthrough/x86/iommu.c | 50 ++++++++++++++++++--- xen/include/xen/iommu.h | 2 +- 7 files changed, 65 insertions(+), 6 deletions(-)
docs/misc/xen-command-line.markdown +9 -0
@@ -702,6 +702,15 @@ This list of booleans control the iommu usage by Dom0: option is only applicable to a PV Dom0 and is enabled by default on Intel hardware. +* `map-reserved`: sets up DMA remapping for all the reserved regions in the + memory map for Dom0. Use this to work around firmware issues providing + incorrect RMRR/IVMD entries. Rather than only mapping RAM pages for IOMMU + accesses for Dom0, all memory regions marked as reserved in the memory map + that don't overlap with any MMIO region from emulated devices will be + identity mapped. This option maps a subset of the memory that would be + mapped when using the `map-inclusive` option. This option is available to a + PVH Dom0 and is enabled by default on Intel hardware. + ### dom0\_ioports\_disable (x86) > `= List of <hex>-<hex>`
xen/drivers/passthrough/amd/pci_amd_iommu.c +3 -0
@@ -256,6 +256,9 @@ static void __hwdom_init amd_iommu_hwdom_init(struct domain *d)/* Inclusive IOMMU mappings are disabled by default on AMD hardware. */ if ( iommu_hwdom_inclusive ==-1 ) iommu_hwdom_inclusive = false; +/* Reserved IOMMU mappings are disabled by default on AMD hardware. */ +if ( iommu_hwdom_reserved ==-1 ) + iommu_hwdom_reserved = false;if (allocate_domain_resources (dom_iommu (d)) )BUG ();
xen/drivers/passthrough/arm/smmu.c +1 -0
@@ -2729,6 +2729,7 @@ static void __hwdom_init arm_smmu_iommu_hwdom_init(struct domain *d) {/* Set to false options not supported on ARM. */ iommu_hwdom_inclusive = false; + iommu_hwdom_reserved = false; }static void arm_smmu_iommu_domain_teardown (struct domain *d)
xen/drivers/passthrough/iommu.c +3 -0
@@ -62,6 +62,7 @@ bool_t __read_mostly iommu_intremap = 1; bool __hwdom_initdata iommu_hwdom_strict;bool __read_mostly iommu_hwdom_passthrough; int8_t __hwdom_initdata iommu_hwdom_inclusive = -1; +int8_t __hwdom_initdata iommu_hwdom_reserved = -1;/* * In the current implementation of VT-d posted interrupts, in some extreme
@@ -155,6 +156,8 @@ static int __init parse_dom0_iommu_param(const char *s) iommu_hwdom_strict = val; else if ( (val = parse_boolean("map-inclusive", s, ss)) >= 0 ) iommu_hwdom_inclusive = val; + else if ( (val = parse_boolean("map-reserved", s, ss)) >= 0 ) + iommu_hwdom_inclusive = val; else rc = -EINVAL ;
xen/drivers/passthrough/vtd/iommu.c +3 -0
@@ -1307,6 +1307,9 @@ static void __hwdom_init intel_iommu_hwdom_init(struct domain *d)/* Inclusive mappings are enabled by default on Intel hardware for PV. */ if ( iommu_hwdom_inclusive ==-1 ) iommu_hwdom_inclusive =is_pv_domain (d); +/* Reserved IOMMU mappings are enabled by default on Intel hardware. */ +if ( iommu_hwdom_reserved ==-1 ) + iommu_hwdom_reserved = true;setup_hwdom_pci_devices (d, setup_hwdom_device);setup_hwdom_rmrr (d);
xen/drivers/passthrough/x86/iommu.c +45 -5
89 lines not on this page · raw message
xen/include/xen/iommu.h +1 -1
@@ -37,7 +37,7 @@ extern bool_t iommu_debug;extern bool_t amd_iommu_perdev_intremap;extern bool iommu_hwdom_strict, iommu_hwdom_passthrough; -extern int8_t iommu_hwdom_inclusive; +extern int8_t iommu_hwdom_inclusive, iommu_hwdom_reserved;extern unsigned int iommu_dev_iotlb_timeout;
Signature · 7 lines
2.18.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 8/8] x86/iommu: add map-reserved dom0-iommu option to map reserved memory ranges
Quoted 11 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/x86/iommu.c +2
-0@@ -138,13 +139,20 @@ static bool __hwdom_init hwdom_iommu_map(const struct domain *d, unsigned long pfn, unsigned long max_pfn) { +unsigned int i; +/* * Set up 1:1 mapping for dom0. Default to include only conventional RAM * areas and let RMRRs include needed reserved regions. When set, the * inclusive mapping additionally maps in every pfn up to 4GB except those
xen/drivers/passthrough/x86/iommu.c static bool __hwdom_init hwdom_iommu_map(const struct domain *d, · line 155
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +7
-2-* that fall in unusable ranges. +* that fall in unusable ranges for PV Dom0. */ -if ( (pfn > max_pfn && !mfn_valid (_mfn (pfn))) ||xen_in_range (pfn) ) +if ( (pfn > max_pfn && !mfn_valid (_mfn (pfn))) ||xen_in_range (pfn) || +/* +* Ignore any address below 1MB, that's already identity mapped by the +* domain builder for HVM. +*/ + (!d->domain_id &&is_hvm_domain (d) && pfn <PFN_DOWN (MB (1 ))) )
Would you mind clarifying the comment by saying "Dom0 builder"? The respective source files have been renamed to dom0_build.c for the same reason, quite some time ago.
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+/* Check that it doesn't overlap with the LAPIC */ +if (has_vlapic (d) ) + { +const struct vcpu *v; + +for_each_vcpu (d, v) +if ( pfn ==PFN_DOWN (vlapic_base_address (vcpu_vlapic (v))) ) +return false; + } +/* ... or the IO-APIC */
xen/drivers/passthrough/x86/iommu.c static bool __hwdom_init hwdom_iommu_map(const struct domain *d, · line 197
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+for ( i =0 ;has_vioapic (d) && i < d->arch.hvm_domain.nr_vioapics; i++ ) +if ( pfn ==PFN_DOWN (domain_vioapic (d, i)->base_address) ) +return false; +/* +* ... or the PCIe MCFG regions. +* TODO: runtime added MMCFG regions are not checked to make sure they +* don't overlap with already mapped regions, thus preventing trapping. +*/ +if (has_vpci (d) &&vpci_is_mmcfg_address (d,pfn_to_paddr (pfn)) ) +return false;
Didn't we agree to have such a TODO also in the LAPIC loop?
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 228
Quoted 12 lines
xen/drivers/passthrough/x86/iommu.c +7
-1@@ -185,7 +219,13 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)if ( !hwdom_iommu_map (d, pfn, max_pfn) )continue ; - rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable); +if (iommu_use_hap_pt (d) ) + { +ASSERT (is_hvm_domain (d)); + rc =set_identity_p2m_entry (d, pfn, p2m_access_rw,0 ); + } +else + rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable);
Why iommu_use_hap_pt()? Shouldn't HAP with or without shared page tables as well as shadow all get the same in-sync p2m and IOMMU mappings?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 8/8] x86/iommu: add map-reserved dom0-iommu option to map reserved memory ranges
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/x86/iommu.c +0
-0@@ -185,7 +219,13 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)if ( !hwdom_iommu_map (d, pfn, max_pfn) )continue ;
Quoted 12 lines
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 228
Quoted 8 lines
xen/drivers/passthrough/x86/iommu.c +7
-1- rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable); +if (iommu_use_hap_pt (d) ) + { +ASSERT (is_hvm_domain (d)); + rc =set_identity_p2m_entry (d, pfn, p2m_access_rw,0 ); + } +else + rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable);Why
iommu_use_hap_pt()? Shouldn't HAP with or without shared page tables as well as shadow all get the same in-sync p2m and IOMMU mappings?
iommu_map_page is a noop if iommu_use_hap_pt is true (see intel_iommu_map_page for example). Hence in the case the IOMMU page tables are shared with HAP the pages must be added to the p2m.
I could switch this to use set_identity_p2m_entry if the guest is auto-translated and only use iommu_map_page for non-autotranslated guests.
Thanks, Roger.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v5 8/8] x86/iommu: add map-reserved dom0-iommu option to map reserved memory ranges
Quoted 20 lines
Quoted 1 line
Quoted 1 line · On 17.08.18 at 13:17, <[email protected]> wrote:
Quoted 5 lines · Jan Beulich
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 14.08.18 at 15:43, <[email protected]> wrote:
xen/drivers/passthrough/x86/iommu.c +0
-0@@ -185,7 +219,13 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)if ( !hwdom_iommu_map (d, pfn, max_pfn) )continue ;Quoted 12 lines
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 228
Quoted 8 lines
xen/drivers/passthrough/x86/iommu.c +7
-1- rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable); +if (iommu_use_hap_pt (d) ) + { +ASSERT (is_hvm_domain (d)); + rc =set_identity_p2m_entry (d, pfn, p2m_access_rw,0 ); + } +else + rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable);Why
iommu_use_hap_pt()? Shouldn't HAP with or without shared page tables as well as shadow all get the same in-sync p2m and IOMMU mappings?
Quoted 3 lines
iommu_map_pageis a noop ifiommu_use_hap_ptis true (seeintel_iommu_map_pagefor example). Hence in the case the IOMMU page tables are shared with HAP the pages must be added to the p2m.
This wasn't what I'm after, while ...
Quoted 3 lines
I could switch this to use
set_identity_p2m_entryif the guest is auto-translated and only useiommu_map_pagefor non-autotranslated guests.
.... indeed this is what I would prefer (unless there a reasons for not doing so).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel