libxl: choose an appropriate default for passthrough...
...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
xl: libxl_create.c:589: libxl__domain_make: Assertion `info->passthrough != LIBXL_PASSTHROUGH_ENABLED' failed.
This is because libxl__domain_create_info_setdefault() currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.
---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Juergen Gross <[email protected]>
--- tools/libxl/libxl_create.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
tools/libxl/libxl_create.c +5 -1
@@ -68,7 +68,11 @@ int libxl__domain_create_info_setdefault(libxl__gc *gc, c_info->passthrough = ((c_info->type ==LIBXL_DOMAIN_TYPE_PV ) || !info.cap_iommu_hap_pt_share) ?LIBXL_PASSTHROUGH_SYNC_PT :LIBXL_PASSTHROUGH_SHARE_PT ; - } + }else if (!info.cap_hvm_directio) + c_info->passthrough =LIBXL_PASSTHROUGH_DISABLED ; + +/* An explicit setting should now have been chosen */ +assert (c_info->passthrough !=LIBXL_PASSTHROUGH_ENABLED );return 0 ; }
Signature · 7 lines
2.20.1.2.gb21ebb671 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 6 lines · Paul Durrant
...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
xl: libxl_create.c:589:
libxl__domain_make: Assertion `info->passthrough !=LIBXL_PASSTHROUGH_ENABLED' failed.
Quoted 8 lines
This is because
libxl__domain_create_info_setdefault()currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.Signed-off-by Paul Durrant <[email protected]>
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Paul Durrant writes ("[PATCH-for-4.13] libxl: choose an appropriate default for passthrough..."):
Quoted 4 lines
...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
Quoted 8 lines
xl: libxl_create.c:589:
libxl__domain_make: Assertion `info->passthrough !=LIBXL_PASSTHROUGH_ENABLED' failed.This is because
libxl__domain_create_info_setdefault()currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.
It's weird that after this patch "enabled" can mean DISABLED. Surely if you say `passthrough="enabled"' and the host has no PT support (eg it's disabled in the bios) it should fail ?
Normally libxl config options have an "unknown" or "default" option.
Also it is anomalous that xl is doing the complex work of choosing a default. I think all the complex code
+ switch (c_info->passthrough) {
+ case LIBXL_PASSTHROUGH_ENABLED:
in xl_parse.c should be in libxl. (Some of it is there already.)
I'm sorry that I wasn't didn't review babde47a3fed...
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Quoted 23 lines
-----Original Message-----
From: Ian Jackson <[email protected]>
Sent: 01 October 2019 11:39
To: Paul Durrant <[email protected]>
Cc: [email protected]; Wei Liu <[email protected]>; Anthony Perard <[email protected]>; Juergen Gross <[email protected]>
Subject: Re: [PATCH-for-4.13] libxl: choose an appropriate default for passthrough...Paul Durrant writes ("[PATCH-for-4.13] libxl: choose an appropriate default for passthrough..."):
Quoted 5 lines
...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
xl: libxl_create.c:589:
libxl__domain_make: Assertion `info->passthrough !=
LIBXL_PASSTHROUGH_ENABLED' failed.Quoted 7 lines
This is because
libxl__domain_create_info_setdefault()currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.
Quoted 3 lines
It's weird that after this patch "enabled" can mean DISABLED. Surely if you say `passthrough="enabled"' and the host has no PT support (eg it's disabled in the bios) it should fail ?
Indeed, and xl will do exactly that.
Quoted 9 lines
Normally libxl config options have an "unknown" or "default" option.
Also it is anomalous that xl is doing the complex work of choosing a default. I think all the complex code
(quoted diff) +2
-0+ switch (c_info->passthrough) { + case LIBXL_PASSTHROUGH_ENABLED:
Quoted 4 lines
in
xl_parse.cshould be in libxl. (Some of it is there already.)I'm sorry that I wasn't didn't review
babde47a3fed...
So, would you advocate an 'unknown' value then? That's essentially just a rename operation on 'enabled'.
The code in xl_parse.c is there for a reason though; the appropriate amount of extra memory for the IOMMU page tables has to be determined there because the 'build' parts of libxl seem to be largely firewalled from the 'create' parts and thus the relevant information is not available to decide the appropriate overhead.
Paul
Quoted 1 line
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Ping? Can I get a response on this (w.r.t. 'enabled' vs. 'unknown') before doing a v2? This issue is currently blocking a push, I believe.
Quoted 48 lines · Paul Durrant
Quoted 24 lines
-----Original Message-----
From: Ian Jackson <[email protected]>
Sent: 01 October 2019 11:39
To: Paul Durrant <[email protected]>
Cc: [email protected]; Wei Liu <[email protected]>; Anthony Perard <[email protected]>; Juergen Gross <[email protected]>
Subject: Re: [PATCH-for-4.13] libxl: choose an appropriate default for passthrough...Paul Durrant writes ("[PATCH-for-4.13] libxl: choose an appropriate default for passthrough..."):
Quoted 5 lines
...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
xl: libxl_create.c:589:
libxl__domain_make: Assertion `info->passthrough !=
LIBXL_PASSTHROUGH_ENABLED' failed.Quoted 7 lines
This is because
libxl__domain_create_info_setdefault()currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.Quoted 3 lines
It's weird that after this patch "enabled" can mean DISABLED. Surely if you say `passthrough="enabled"' and the host has no PT support (eg it's disabled in the bios) it should fail ?
Indeed, and xl will do exactly that.
Quoted 9 lines
Normally libxl config options have an "unknown" or "default" option.
Also it is anomalous that xl is doing the complex work of choosing a default. I think all the complex code
(quoted diff) +2
-0+ switch (c_info->passthrough) { + case LIBXL_PASSTHROUGH_ENABLED:Quoted 4 lines
in
xl_parse.cshould be in libxl. (Some of it is there already.)I'm sorry that I wasn't didn't review
babde47a3fed...So, would you advocate an 'unknown' value then? That's essentially just a rename operation on 'enabled'.
The code in
xl_parse.cis there for a reason though; the appropriate amount of extra memory for the IOMMU page tables has to be determined there because the 'build' parts of libxl seem to be largely firewalled from the 'create' parts and thus the relevant information is not available to decide the appropriate overhead.
Quoted 8 lines
Paul
Quoted 1 line
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Paul Durrant writes ("Re: [Xen-devel] [PATCH-for-4.13] libxl: choose an appropriate default for passthrough..."):
Quoted 2 lines
Ping? Can I get a response on this (w.r.t. 'enabled' vs. 'unknown') before doing a v2? This issue is currently blocking a push, I believe.
I definitely think we should introduce "unknown".
Quoted 1 line
Quoted 1 line
So, would you advocate an 'unknown' value then? That's essentially just a rename operation on 'enabled'.
I think we probably want "enabled" as well but that can wait. If for now you rename "unknown" that will do.
Quoted 1 line
Quoted 1 line
The code in
xl_parse.cis there for a reason though; the appropriate amount of extra memory for the IOMMU page tables has to be determined there because the 'build' parts of libxl seem to be largely firewalled from the 'create' parts and thus the relevant information is not available to decide the appropriate overhead.
I want to look into this more deeply.
Thanks,
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
A message that is not in the archive (sent off-list or not yet received): CACCGGhAQURtCRt5Zn9a6uDvUDKdtAyV_pqv_HQ5CpZ20pBNf1A@mail.gmail.com
Re: [Xen-devel] [PATCH-for-4.13] libxl: choose an appropriate default for passthrough... [and 1 more messages]
Ian Jackson writes ("Re: [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 3 lines
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
Quoted 7 lines · Paul Durrant
Ok. Specifically
libxl_domain_need_memory()only has access to the build info, butlibxl__domain_build_info_setdefault()does not have access to the create info, so it can't choose a sensible default. To avoid re-writing lots of code I went with having xl calulate a default. (Prior to my patch there was no calculated overhead anyway so a host without shared EPT and autoballooning on was already playing russian roulette).
So some restructuring may be needed. I'll see how invasive it looks.
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
v2 cover libxl: fix assertion failure
This was originally a single patch, which is now patch #2 of this series.
Paul Durrant (2):
libxl: replace 'enabled' with 'unknown' in libxl_passthrough
enumeration
libxl: choose an appropriate default for passthrough...
tools/libxl/libxl_create.c | 10 +++++++--- tools/libxl/libxl_types.idl | 2 +- tools/xl/xl_parse.c | 26 +++++++++++++++----------- 3 files changed, 23 insertions(+), 15 deletions(-)
---
Cc: Anthony PERARD <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Juergen Gross <[email protected]>
Cc: Wei Liu <[email protected]>
Signature · 7 lines
2.20.1.2.gb21ebb671 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v2 1/2 libxl: replace 'enabled' with 'unknown' in libxl_passthrough enumeration
This is mostly a cosmetic patch to avoid the default enumeration value being 'enabled'. The only non-cosmetic parts are in xl_parse.c where it now becomes necessary to explicitly parse the 'enabled' value for xl.cfg 'passthrough' option, and error on the value 'unknown', because there is no longer a direct mapping between valid xl.cfg values and the enumeration.
---
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Juergen Gross <[email protected]>
v2:
- new in v2
--- tools/libxl/libxl_create.c | 4 ++-- tools/libxl/libxl_types.idl | 2 +- tools/xl/xl_parse.c | 26 +++++++++++++++----------- 3 files changed, 18 insertions(+), 14 deletions(-)
tools/libxl/libxl_create.c +2 -2
@@ -64,7 +64,7 @@ int libxl__domain_create_info_setdefault(libxl__gc *gc, c_info->ssidref =SECINITSID_DOMU ;if (info.cap_hvm_directio && - (c_info->passthrough ==LIBXL_PASSTHROUGH_ENABLED )) { + (c_info->passthrough ==LIBXL_PASSTHROUGH_UNKNOWN )) { c_info->passthrough = ((c_info->type ==LIBXL_DOMAIN_TYPE_PV ) || !info.cap_iommu_hap_pt_share) ?LIBXL_PASSTHROUGH_SYNC_PT :LIBXL_PASSTHROUGH_SHARE_PT ;
@@ -586,7 +586,7 @@ int libxl__domain_make(libxl__gc *gc, libxl_domain_config *d_config,libxl_defbool_val (info->oos) ?0 : XEN_DOMCTL_CDF_oos_off; } -assert (info->passthrough !=LIBXL_PASSTHROUGH_ENABLED ); +assert (info->passthrough !=LIBXL_PASSTHROUGH_UNKNOWN );LOG (DETAIL ,"passthrough: %s" ,libxl_passthrough_to_string (info->passthrough));
tools/libxl/libxl_types.idl +1 -1
@@ -264,7 +264,7 @@ libxl_vkb_backend = Enumeration("vkb_backend", [ ]) libxl_passthrough = Enumeration("passthrough", [ - (0, "enabled"), + (0, "unknown"), (1, "disabled"), (2, "sync_pt"), (3, "share_pt"),
tools/xl/xl_parse.c +15 -11
@@ -1512,23 +1512,27 @@ void parse_config_data(const char *config_source,if (xlu_cfg_get_string (config,"passthrough" , &buf,0 )) { c_info->passthrough = (d_config->num_pcidevs || d_config->num_dtdevs) - ?LIBXL_PASSTHROUGH_ENABLED :LIBXL_PASSTHROUGH_DISABLED ; + ?LIBXL_PASSTHROUGH_UNKNOWN :LIBXL_PASSTHROUGH_DISABLED ; }else { -libxl_passthrough o; +if (!strcasecmp ("enabled" , buf)) + c_info->passthrough =LIBXL_PASSTHROUGH_UNKNOWN ; +else { +libxl_passthrough o; - e =libxl_passthrough_from_string (buf, &o); -if (e) { -fprintf (stderr, -"ERROR: unknown passthrough option '%s'\n" , - buf); -exit (-ERROR_FAIL ); - } + e =libxl_passthrough_from_string (buf, &o); +if (e || !strcasecmp ("unknown" , buf)) { +fprintf (stderr, +"ERROR: unknown passthrough option '%s'\n" , + buf); +exit (-ERROR_FAIL ); + } - c_info->passthrough = o; + c_info->passthrough = o; + } }switch (c_info->passthrough) { -case LIBXL_PASSTHROUGH_ENABLED : +case LIBXL_PASSTHROUGH_UNKNOWN :/* * Choose a suitable default. libxl would also do this but * choosing here allows the code calculating 'iommu_memkb'
Signature · 7 lines
2.20.1.2.gb21ebb671 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v2 2/2 libxl: choose an appropriate default for passthrough...
In reply to Paul Durrant...if there is no IOMMU or it is globally disabled.
Without this patch, the following assertion may be hit:
xl: libxl_create.c:589: libxl__domain_make: Assertion `info->passthrough != LIBXL_PASSTHROUGH_UNKNOWN' failed.
This is because libxl__domain_create_info_setdefault() currently only sets an appropriate value for 'passthrough' in the case that 'cap_hvm_directio' is true, which is not the case unless an IOMMU is present and enabled in the system. This is normally masked by xl choosing a default value, but that will not happen if xl is not used (e.g. when using libvirt) or when a stub domain is being created.
---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Juergen Gross <[email protected]>
v2:
- re-base
--- tools/libxl/libxl_create.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
tools/libxl/libxl_create.c +5 -1
@@ -68,7 +68,11 @@ int libxl__domain_create_info_setdefault(libxl__gc *gc, c_info->passthrough = ((c_info->type ==LIBXL_DOMAIN_TYPE_PV ) || !info.cap_iommu_hap_pt_share) ?LIBXL_PASSTHROUGH_SYNC_PT :LIBXL_PASSTHROUGH_SHARE_PT ; - } + }else if (!info.cap_hvm_directio) + c_info->passthrough =LIBXL_PASSTHROUGH_DISABLED ; + +/* An explicit setting should now have been chosen */ +assert (c_info->passthrough !=LIBXL_PASSTHROUGH_UNKNOWN );return 0 ; }
Signature · 7 lines
2.20.1.2.gb21ebb671 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
In reply to Paul DurrantQuoted 11 lines · Paul Durrant
This was originally a single patch, which is now patch #2 of this series.
Paul Durrant (2):
libxl: replace 'enabled' with 'unknown' inlibxl_passthrough
enumeration
libxl: choose an appropriate default for passthrough...tools/libxl/libxl_create.c | 10 +++++++--- tools/libxl/libxl_types.idl | 2 +- tools/xl/xl_parse.c | 26 +++++++++++++++----------- 3 files changed, 23 insertions(+), 15 deletions(-)
For the series:
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
In reply to Paul DurrantPaul Durrant writes ("[PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines
This was originally a single patch, which is now patch #2 of this series.
Paul Durrant (2):
libxl: replace 'enabled' with 'unknown' inlibxl_passthrough
enumeration
libxl: choose an appropriate default for passthrough...
Thanks. I have applied these, and also a style fixup (below).
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
Ian.
From b01b1dc046da70a2621a4d1f032ddb22b0cdde6b Mon Sep 17 00:00:00 2001
From: Ian Jackson <[email protected]>
Date: Wed, 2 Oct 2019 16:55:47 +0100
Subject: [PATCH] libxl: create: style: Add a pair of missing { ]
From CODING_STYLE:
Every indented statement is braced, but blocks that contain just one statement may have the braces omitted. To avoid confusion, either all the blocks in an if...else chain have braces, or none of them do.
--- tools/libxl/libxl_create.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
tools/libxl/libxl_create.c +2 -1
@@ -68,8 +68,9 @@ int libxl__domain_create_info_setdefault(libxl__gc *gc, c_info->passthrough = ((c_info->type ==LIBXL_DOMAIN_TYPE_PV ) || !info.cap_iommu_hap_pt_share) ?LIBXL_PASSTHROUGH_SYNC_PT :LIBXL_PASSTHROUGH_SHARE_PT ; - }else if (!info.cap_hvm_directio) + }else if (!info.cap_hvm_directio) { c_info->passthrough =LIBXL_PASSTHROUGH_DISABLED ; + }/* An explicit setting should now have been chosen */ assert (c_info->passthrough !=LIBXL_PASSTHROUGH_UNKNOWN );
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
Quoted 11 lines · Ian Jackson
Paul Durrant writes ("[PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines
This was originally a single patch, which is now patch #2 of this series.
Paul Durrant (2):
libxl: replace 'enabled' with 'unknown' inlibxl_passthrough
enumeration
libxl: choose an appropriate default for passthrough...Thanks. I have applied these, and also a style fixup (below).
Cool.
Quoted 4 lines
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
That would indeed be beneficial for the likes of libvirt. Perhaps it would be reasonable to unify the create and build info at a libxl level (even though they may feed into distinct domctls underneath for the moment)?
Cheers,
Paul
Quoted 38 lines
Ian.
From
b01b1dc046da70a2621a4d1f032ddb22b0cdde6bMon Sep 17 00:00:00 2001
From: Ian Jackson <[email protected]>
Date: Wed, 2 Oct 2019 16:55:47 +0100
Subject: [PATCH] libxl: create: style: Add a pair of missing { ]From
CODING_STYLE:Every indented statement is braced, but blocks that contain just one statement may have the braces omitted. To avoid confusion, either all the blocks in an if...else chain have braces, or none of them do.
CC Paul Durrant <[email protected]>Signed-off-by Ian Jackson <[email protected]>--- tools/libxl/libxl_create.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)tools/libxl/libxl_create.c +2
-2@@ -68,8 +68,9 @@ int libxl__domain_create_info_setdefault(libxl__gc *gc, c_info->passthrough = ((c_info->type ==LIBXL_DOMAIN_TYPE_PV ) || !info.cap_iommu_hap_pt_share) ?LIBXL_PASSTHROUGH_SYNC_PT :LIBXL_PASSTHROUGH_SHARE_PT ; - }else if (!info.cap_hvm_directio) + }else if (!info.cap_hvm_directio) { c_info->passthrough =LIBXL_PASSTHROUGH_DISABLED ; + }/* An explicit setting should now have been chosen */ assert (c_info->passthrough !=LIBXL_PASSTHROUGH_UNKNOWN ); --2.11.0
Quoted 4 lines
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
Paul Durrant writes ("Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines · Ian Jackson
Quoted 3 lines
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
That would indeed be beneficial for the likes of libvirt.
I propose the following plan for 4.13:
* Move the default calculations of b_info->shadow_memkb and b_info->iommu_memkb from xl_vmcontrol.c into libxl, in a new function libxl__need_memory_setdefault, called from initiate_domain_create. That has access to the whole of c_info and b_info.
* Change the API/ABI for libxl_domain_need_memory to take a libxl_domain_config. Internally, this will call an implementation function libxl__domain_need_memory which takes the b_info and c_info separately, and which calls libxl__need_memory_setdefault. (This is the only other call site for libxl__domain_build_info_setdefault.)
* There will be the usual backward compatible arrangement: here, a function libxl_domain_need_memory_0x040c00, which will pass NULL for c_info. The code in libxl__need_memory_setdefault will use 0 for the two additional memory amounts when c_info is NULL.
* The overall effect is that old callers will get the old behaviour. New callers get the new right behaviour. This is the same as the present libxl 4.13 code. Note that libxl_domain_need_memory already has an API stability caveat.
* Consequently, the need for libxl_get_required_shadow_memory and libxl_get_required_iommu_memory goes away. Delete them (they have not been in any release so we can just do this).
* Invent a new value for c_info->passthrough "enabled". Defaulting will be 1. turn "unknown" into "disabled" or "enabled" according to the current logic based on pcidevs/dtdefs; 2. turn "enabled" into something specific according to the current logic based on type, hap_pt_share, etc. Make sure this is all correct inside libxl.
* Delete the defaulting code in xl. xl can just leave settings not specified by the user as blank, and libxl will DTRT with them.
What do people think ? I really want to fix this for 4.13 because the current 4.13 API is not one I want to support.
Quoted 3 lines
Perhaps it would be reasonable to unify the create and build info at a libxl level (even though they may feed into distinct domctls underneath for the moment)?
Yes, but we are probably too late for such an API change at this point for 4.13.
Thanks,
Ian.
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
Quoted 28 lines · Ian Jackson
Paul Durrant writes ("Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines · Ian Jackson
Quoted 3 lines
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
That would indeed be beneficial for the likes of libvirt.
I propose the following plan for 4.13:
* Move the default calculations of
b_info->shadow_memkbandb_info->iommu_memkbfromxl_vmcontrol.cinto libxl, in a new functionlibxl__need_memory_setdefault, called frominitiate_domain_create. That has access to the whole ofc_infoandb_info.* Change the API/ABI for
libxl_domain_need_memoryto take alibxl_domain_config. Internally, this will call an implementation functionlibxl__domain_need_memorywhich takes theb_infoandc_infoseparately, and which callslibxl__need_memory_setdefault. (This is the only other call site forlibxl__domain_build_info_setdefault.)* There will be the usual backward compatible arrangement: here, a function
libxl_domain_need_memory_0x040c00, which will pass NULL forc_info. The code inlibxl__need_memory_setdefaultwill use 0 for the two additional memory amounts whenc_infois NULL.
Quoted 8 lines
* The overall effect is that old callers will get the old behaviour. New callers get the new right behaviour. This is the same as the present libxl 4.13 code. Note that
libxl_domain_need_memoryalready has an API stability caveat.* Consequently, the need for
libxl_get_required_shadow_memoryandlibxl_get_required_iommu_memorygoes away. Delete them (they have not been in any release so we can just do this).
libxl_get_required_shadow_memory is old, and libvirt is using. Only libxl_get_required_iommu_memory is new.
Quoted 11 lines
* Invent a new value for
c_info->passthrough"enabled". Defaulting will be 1. turn "unknown" into "disabled" or "enabled" according to the current logic based on pcidevs/dtdefs; 2. turn "enabled" into something specific according to the current logic based on type,hap_pt_share, etc. Make sure this is all correct inside libxl.* Delete the defaulting code in xl. xl can just leave settings not specified by the user as blank, and libxl will DTRT with them.
What do people think ? I really want to fix this for 4.13 because the current 4.13 API is not one I want to support.
That plan sound fine to me.
Signature · 6 lines
Anthony PERARD _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
In reply to Ian JacksonQuoted 9 lines · Ian Jackson
Paul Durrant writes ("Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines · Ian Jackson
Quoted 3 lines
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
That would indeed be beneficial for the likes of libvirt.
I propose the following plan for 4.13:
Fine with me.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH-for-4.13 v2 0/2] libxl: fix assertion failure
In reply to Ian JacksonQuoted 13 lines · Ian Jackson
Paul Durrant writes ("[PATCH-for-4.13 v2 0/2] libxl: fix assertion failure"):
Quoted 6 lines
This was originally a single patch, which is now patch #2 of this series.
Paul Durrant (2):
libxl: replace 'enabled' with 'unknown' inlibxl_passthrough
enumeration
libxl: choose an appropriate default for passthrough...Thanks. I have applied these, and also a style fixup (below).
I am continuing to look at the defaulting and config management here with a view to getting rid of some of the duplicated code and moving it all into libxl.
Quoted 6 lines
Ian.
From
b01b1dc046da70a2621a4d1f032ddb22b0cdde6bMon Sep 17 00:00:00 2001
From: Ian Jackson <[email protected]>
Date: Wed, 2 Oct 2019 16:55:47 +0100
Subject: [PATCH] libxl: create: style: Add a pair of missing { ]
s/]/}/
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel