[PATCH v6 0/4] x86/pvh: Support relocating dom0 kernel

[PATCH v6 0/4] x86/pvh: Support relocating dom0 kernel

[Jason Andryuk]

Xen tries to load a PVH dom0 kernel at the fixed guest physical address
from the elf headers.  For Linux, this defaults to 0x1000000 (16MB), but
it can be configured.

Unfortunately there exist firmwares that have reserved regions at this
address, so Xen fails to load the dom0 kernel since it's not RAM.

The other issue is that the Linux PVH entry point is not
position-independent.  It expects to run at the compiled
CONFIG_PHYSICAL_ADDRESS.

This patch set expands the PVH dom0 builder to try to relocate the
kernel if needed and possible.  XEN_ELFNOTE_PHYS32_RELOC is added for
kernels to indicate they are relocatable and their acceptable address
range and alignment.

v6:
Choose the alignment from the Note if specified, or the Maximum PHDR
p_align value if greater than PAGE_SIZE.  Otherwise, it falls back to
the default 2MB.

Patches from v5 commited:
853c71dfbf xen/elfnote: Specify ELF Notes are x86-specific
7d8c9b4e8d libelf: Expand ELF note printing
8802230bfa Revert "xen/x86: bzImage parse kernel_alignment"

The first and second patches move MB/GB() to common-macros.h.

The third patch stores the maximum p_align value from the ELF PHDRs.

The fourth patch expands the pvh dom0 kernel placement code.

I'll post an additional patch showing the Linux changes to make PVH
relocatable.

Jason Andryuk (4):
  tools/init-xenstore-domain: Replace variable MB() usage
  tools: Move MB/GB() to common-macros.h
  libelf: Store maximum PHDR p_align
  x86/PVH: Support relocatable dom0 kernels

 tools/firmware/hvmloader/util.h         |   3 -
 tools/helpers/init-xenstore-domain.c    |  11 ++-
 tools/include/xen-tools/common-macros.h |   4 +
 tools/libs/light/libxl_internal.h       |   4 -
 xen/arch/x86/hvm/dom0_build.c           | 108 ++++++++++++++++++++++++
 xen/common/libelf/libelf-dominfo.c      |  35 ++++++++
 xen/common/libelf/libelf-loader.c       |  15 +++-
 xen/common/libelf/libelf-private.h      |   1 +
 xen/include/public/elfnote.h            |  16 +++-
 xen/include/xen/libelf.h                |   5 ++
 10 files changed, 186 insertions(+), 16 deletions(-)

-- 
2.44.0

[PATCH v6 1/4] tools/init-xenstore-domain: Replace variable MB() usage

[Jason Andryuk]

The local MB() & GB() macros will be replaced with a common
implementation, but those only work with constant values.  Introduce a
static inline mb_to_bytes() in place of the MB() macro to convert the
variable values.

Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
---
v4:
New
---
 tools/helpers/init-xenstore-domain.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/tools/helpers/init-xenstore-domain.c b/tools/helpers/init-xenstore-domain.c
index 1683438c5c..5405842dfe 100644
--- a/tools/helpers/init-xenstore-domain.c
+++ b/tools/helpers/init-xenstore-domain.c
@@ -20,7 +20,6 @@
 #include "init-dom-json.h"
 
 #define LAPIC_BASE_ADDRESS  0xfee00000UL
-#define MB(x)               ((uint64_t)x << 20)
 #define GB(x)               ((uint64_t)x << 30)
 
 static uint32_t domid = ~0;
@@ -36,6 +35,11 @@ static xc_evtchn_port_or_error_t console_evtchn;
 static xentoollog_level minmsglevel = XTL_PROGRESS;
 static void *logger;
 
+static inline uint64_t mb_to_bytes(int mem)
+{
+	return (uint64_t)mem << 20;
+}
+
 static struct option options[] = {
     { "kernel", 1, NULL, 'k' },
     { "memory", 1, NULL, 'm' },
@@ -76,8 +80,8 @@ static int build(xc_interface *xch)
     int rv, xs_fd;
     struct xc_dom_image *dom = NULL;
     int limit_kb = (maxmem ? : memory) * 1024 + X86_HVM_NR_SPECIAL_PAGES * 4;
-    uint64_t mem_size = MB(memory);
-    uint64_t max_size = MB(maxmem ? : memory);
+    uint64_t mem_size = mb_to_bytes(memory);
+    uint64_t max_size = mb_to_bytes(maxmem ? : memory);
     struct e820entry e820[3];
     struct xen_domctl_createdomain config = {
         .ssidref = SECINITSID_DOMU,
-- 
2.44.0

Re: [PATCH v6 1/4] tools/init-xenstore-domain: Replace variable MB() usage

[Anthony PERARD]

On Wed, Mar 27, 2024 at 05:50:59PM -0400, Jason Andryuk wrote:
> The local MB() & GB() macros will be replaced with a common
> implementation, but those only work with constant values.  Introduce a

By the way, this is not true, the macro introduce in the following patch
("tools: Move MB/GB() to common-macros.h") works (compiler doesn't
complain) if you do something like MB(maxmem+0) ;-).

> static inline mb_to_bytes() in place of the MB() macro to convert the
> variable values.
> 
> diff --git a/tools/helpers/init-xenstore-domain.c b/tools/helpers/init-xenstore-domain.c
> index 1683438c5c..5405842dfe 100644
> --- a/tools/helpers/init-xenstore-domain.c
> +++ b/tools/helpers/init-xenstore-domain.c
> @@ -20,7 +20,6 @@
>  #include "init-dom-json.h"
>  
>  #define LAPIC_BASE_ADDRESS  0xfee00000UL
> -#define MB(x)               ((uint64_t)x << 20)
>  #define GB(x)               ((uint64_t)x << 30)
>  
>  static uint32_t domid = ~0;
> @@ -36,6 +35,11 @@ static xc_evtchn_port_or_error_t console_evtchn;
>  static xentoollog_level minmsglevel = XTL_PROGRESS;
>  static void *logger;
>  
> +static inline uint64_t mb_to_bytes(int mem)
> +{
> +	return (uint64_t)mem << 20;
> +}
> +
>  static struct option options[] = {
>      { "kernel", 1, NULL, 'k' },
>      { "memory", 1, NULL, 'm' },
> @@ -76,8 +80,8 @@ static int build(xc_interface *xch)
>      int rv, xs_fd;
>      struct xc_dom_image *dom = NULL;
>      int limit_kb = (maxmem ? : memory) * 1024 + X86_HVM_NR_SPECIAL_PAGES * 4;
> -    uint64_t mem_size = MB(memory);
> -    uint64_t max_size = MB(maxmem ? : memory);
> +    uint64_t mem_size = mb_to_bytes(memory);
> +    uint64_t max_size = mb_to_bytes(maxmem ? : memory);
>      struct e820entry e820[3];
>      struct xen_domctl_createdomain config = {
>          .ssidref = SECINITSID_DOMU,
 
Sorry, just notice they were more versions of the series, so repeating
here:

Looks like this change actually fix a bug. When `maxmem` is set, we
would have "max_size = maxmem", otherwise, it would be
"max_size = memory << 20"

The macro should have been written as
 #define MB(x)               ((uint64_t)(x) << 20)

This patch could be backported to 4.17 and later, with:
    Fixes: 134d53f57707 ("tools/init-xenstore-domain: fix memory map for PVH stubdom")

Anyway, this patch on it's own looks fine:
Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>



-- 
Anthony PERARD

Re: [PATCH v6 1/4] tools/init-xenstore-domain: Replace variable MB() usage

[Jason Andryuk]

On 2024-04-03 08:41, Anthony PERARD wrote:
> On Wed, Mar 27, 2024 at 05:50:59PM -0400, Jason Andryuk wrote:
>> The local MB() & GB() macros will be replaced with a common
>> implementation, but those only work with constant values.  Introduce a
> 
> By the way, this is not true, the macro introduce in the following patch
> ("tools: Move MB/GB() to common-macros.h") works (compiler doesn't
> complain) if you do something like MB(maxmem+0) ;-).

Well, I think I wrote the wrong word for the description.  The old Macro 
might have been okay, but the new one as (_AC(_mb, ULL) << 20) fails for 
MB(memory):

init-xenstore-domain.c: In function ‘build’:
init-xenstore-domain.c:82:28: error: ‘memoryULL’ undeclared (first use 
in this function); did you mean ‘memory’?
    82 |     uint64_t mem_size = MB(memory);
       |                            ^~~~~~

I should have written "only work for numeric values."

>> static inline mb_to_bytes() in place of the MB() macro to convert the
>> variable values.
>>
>> diff --git a/tools/helpers/init-xenstore-domain.c b/tools/helpers/init-xenstore-domain.c
>> index 1683438c5c..5405842dfe 100644
>> --- a/tools/helpers/init-xenstore-domain.c
>> +++ b/tools/helpers/init-xenstore-domain.c
>> @@ -20,7 +20,6 @@
>>   #include "init-dom-json.h"
>>   
>>   #define LAPIC_BASE_ADDRESS  0xfee00000UL
>> -#define MB(x)               ((uint64_t)x << 20)
>>   #define GB(x)               ((uint64_t)x << 30)
>>   
>>   static uint32_t domid = ~0;
>> @@ -36,6 +35,11 @@ static xc_evtchn_port_or_error_t console_evtchn;
>>   static xentoollog_level minmsglevel = XTL_PROGRESS;
>>   static void *logger;
>>   
>> +static inline uint64_t mb_to_bytes(int mem)
>> +{
>> +	return (uint64_t)mem << 20;
>> +}
>> +
>>   static struct option options[] = {
>>       { "kernel", 1, NULL, 'k' },
>>       { "memory", 1, NULL, 'm' },
>> @@ -76,8 +80,8 @@ static int build(xc_interface *xch)
>>       int rv, xs_fd;
>>       struct xc_dom_image *dom = NULL;
>>       int limit_kb = (maxmem ? : memory) * 1024 + X86_HVM_NR_SPECIAL_PAGES * 4;
>> -    uint64_t mem_size = MB(memory);
>> -    uint64_t max_size = MB(maxmem ? : memory);
>> +    uint64_t mem_size = mb_to_bytes(memory);
>> +    uint64_t max_size = mb_to_bytes(maxmem ? : memory);
>>       struct e820entry e820[3];
>>       struct xen_domctl_createdomain config = {
>>           .ssidref = SECINITSID_DOMU,
>   
> Sorry, just notice they were more versions of the series, so repeating
> here:
> 
> Looks like this change actually fix a bug. When `maxmem` is set, we
> would have "max_size = maxmem", otherwise, it would be
> "max_size = memory << 20"
> 
> The macro should have been written as
>   #define MB(x)               ((uint64_t)(x) << 20)
> 
> This patch could be backported to 4.17 and later, with:
>      Fixes: 134d53f57707 ("tools/init-xenstore-domain: fix memory map for PVH stubdom")

Nice catch!

> Anyway, this patch on it's own looks fine:
> Reviewed-by: Anthony PERARD <anthony.perard@citrix.com>

Thank you.

Regards,
Jason

Re: [PATCH v6 1/4] tools/init-xenstore-domain: Replace variable MB() usage

[Jan Beulich]

On 27.03.2024 22:50, Jason Andryuk wrote:
> @@ -36,6 +35,11 @@ static xc_evtchn_port_or_error_t console_evtchn;
>  static xentoollog_level minmsglevel = XTL_PROGRESS;
>  static void *logger;
>  
> +static inline uint64_t mb_to_bytes(int mem)
> +{
> +	return (uint64_t)mem << 20;
> +}

While committing I noticed tab indentation here, which looked to be in
conflict with ...

>  static struct option options[] = {
>      { "kernel", 1, NULL, 'k' },
>      { "memory", 1, NULL, 'm' },
> @@ -76,8 +80,8 @@ static int build(xc_interface *xch)
>      int rv, xs_fd;
>      struct xc_dom_image *dom = NULL;
>      int limit_kb = (maxmem ? : memory) * 1024 + X86_HVM_NR_SPECIAL_PAGES * 4;
> -    uint64_t mem_size = MB(memory);
> -    uint64_t max_size = MB(maxmem ? : memory);
> +    uint64_t mem_size = mb_to_bytes(memory);
> +    uint64_t max_size = mb_to_bytes(maxmem ? : memory);
>      struct e820entry e820[3];
>      struct xen_domctl_createdomain config = {
>          .ssidref = SECINITSID_DOMU,

... everything else in the file. Hence I took the liberty to adjust this.

Jan

Re: [PATCH v6 1/4] tools/init-xenstore-domain: Replace variable MB() usage

[Jason Andryuk]

On 2024-04-04 06:09, Jan Beulich wrote:
> On 27.03.2024 22:50, Jason Andryuk wrote:
>> @@ -36,6 +35,11 @@ static xc_evtchn_port_or_error_t console_evtchn;
>>   static xentoollog_level minmsglevel = XTL_PROGRESS;
>>   static void *logger;
>>   
>> +static inline uint64_t mb_to_bytes(int mem)
>> +{
>> +	return (uint64_t)mem << 20;
>> +}
> 
> While committing I noticed tab indentation here, which looked to be in
> conflict with ...
> 
>>   static struct option options[] = {
>>       { "kernel", 1, NULL, 'k' },
>>       { "memory", 1, NULL, 'm' },
>> @@ -76,8 +80,8 @@ static int build(xc_interface *xch)
>>       int rv, xs_fd;
>>       struct xc_dom_image *dom = NULL;
>>       int limit_kb = (maxmem ? : memory) * 1024 + X86_HVM_NR_SPECIAL_PAGES * 4;
>> -    uint64_t mem_size = MB(memory);
>> -    uint64_t max_size = MB(maxmem ? : memory);
>> +    uint64_t mem_size = mb_to_bytes(memory);
>> +    uint64_t max_size = mb_to_bytes(maxmem ? : memory);
>>       struct e820entry e820[3];
>>       struct xen_domctl_createdomain config = {
>>           .ssidref = SECINITSID_DOMU,
> 
> ... everything else in the file. Hence I took the liberty to adjust this.

Thank you for catching that and fixing it up.

-Jason

[PATCH v6 2/4] tools: Move MB/GB() to common-macros.h

[Jason Andryuk]

Consolidate to a single set of common macros for tools.

MB() will gain another use in libelf, so this movement makes it
available.

Requested-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
---
v4:
New

v5:
Add Jan's R-b & Req-b
---
 tools/firmware/hvmloader/util.h         | 3 ---
 tools/helpers/init-xenstore-domain.c    | 1 -
 tools/include/xen-tools/common-macros.h | 4 ++++
 tools/libs/light/libxl_internal.h       | 4 ----
 4 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/tools/firmware/hvmloader/util.h b/tools/firmware/hvmloader/util.h
index 87be213dec..14078bde1e 100644
--- a/tools/firmware/hvmloader/util.h
+++ b/tools/firmware/hvmloader/util.h
@@ -38,9 +38,6 @@ void __bug(const char *file, int line) __attribute__((noreturn));
 #define BUG() __bug(__FILE__, __LINE__)
 #define BUG_ON(p) do { if (p) BUG(); } while (0)
 
-#define MB(mb) (mb##ULL << 20)
-#define GB(gb) (gb##ULL << 30)
-
 static inline int test_bit(unsigned int b, const void *p)
 {
     return !!(((const uint8_t *)p)[b>>3] & (1u<<(b&7)));
diff --git a/tools/helpers/init-xenstore-domain.c b/tools/helpers/init-xenstore-domain.c
index 5405842dfe..f38ba8d6b5 100644
--- a/tools/helpers/init-xenstore-domain.c
+++ b/tools/helpers/init-xenstore-domain.c
@@ -20,7 +20,6 @@
 #include "init-dom-json.h"
 
 #define LAPIC_BASE_ADDRESS  0xfee00000UL
-#define GB(x)               ((uint64_t)x << 30)
 
 static uint32_t domid = ~0;
 static char *kernel;
diff --git a/tools/include/xen-tools/common-macros.h b/tools/include/xen-tools/common-macros.h
index 81fba2e9f5..07aed92684 100644
--- a/tools/include/xen-tools/common-macros.h
+++ b/tools/include/xen-tools/common-macros.h
@@ -91,6 +91,10 @@
 #define __AC(X, Y)   (X ## Y)
 #define _AC(X, Y)    __AC(X, Y)
 
+/* Size macros. */
+#define MB(_mb)     (_AC(_mb, ULL) << 20)
+#define GB(_gb)     (_AC(_gb, ULL) << 30)
+
 #define get_unaligned_t(type, ptr) ({                               \
     const struct { type x; } __packed *ptr_ = (typeof(ptr_))(ptr);  \
     ptr_->x;                                                        \
diff --git a/tools/libs/light/libxl_internal.h b/tools/libs/light/libxl_internal.h
index 094d0df9b1..803dbc1a03 100644
--- a/tools/libs/light/libxl_internal.h
+++ b/tools/libs/light/libxl_internal.h
@@ -126,10 +126,6 @@
 #define PVSHIM_BASENAME "xen-shim"
 #define PVSHIM_CMDLINE "pv-shim console=xen,pv"
 
-/* Size macros. */
-#define MB(_mb)     (_AC(_mb, ULL) << 20)
-#define GB(_gb)     (_AC(_gb, ULL) << 30)
-
 #define DIV_ROUNDUP(n, d) (((n) + (d) - 1) / (d))
 
 #define LIBXL__LOGGING_ENABLED
-- 
2.44.0

Re: [PATCH v6 2/4] tools: Move MB/GB() to common-macros.h

[Anthony PERARD]

On Wed, Mar 27, 2024 at 05:51:00PM -0400, Jason Andryuk wrote:
> Consolidate to a single set of common macros for tools.
> 
> MB() will gain another use in libelf, so this movement makes it
> available.
> 
> Requested-by: Jan Beulich <jbeulich@suse.com>
> Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
> Reviewed-by: Jan Beulich <jbeulich@suse.com>
> ---

So, this patch fixes potential use issues with the macros in hvmloader
and init-xenstore-domain. While it's not perfect, it still better.

I'll try "MB(memory + 0)" with the different macros:


> diff --git a/tools/firmware/hvmloader/util.h b/tools/firmware/hvmloader/util.h
> index 87be213dec..14078bde1e 100644
> --- a/tools/firmware/hvmloader/util.h
> +++ b/tools/firmware/hvmloader/util.h
> @@ -38,9 +38,6 @@ void __bug(const char *file, int line) __attribute__((noreturn));
>  #define BUG() __bug(__FILE__, __LINE__)
>  #define BUG_ON(p) do { if (p) BUG(); } while (0)
>  
> -#define MB(mb) (mb##ULL << 20)
> -#define GB(gb) (gb##ULL << 30)

With this change we have this change for MB(memory + 0) when applied:
-  (memory + 0ULL << 20)
+  ((memory + 0ULL) << 20)

>  static inline int test_bit(unsigned int b, const void *p)
>  {
>      return !!(((const uint8_t *)p)[b>>3] & (1u<<(b&7)));
> diff --git a/tools/helpers/init-xenstore-domain.c b/tools/helpers/init-xenstore-domain.c
> index 5405842dfe..f38ba8d6b5 100644
> --- a/tools/helpers/init-xenstore-domain.c
> +++ b/tools/helpers/init-xenstore-domain.c
> @@ -20,7 +20,6 @@
>  #include "init-dom-json.h"
>  
>  #define LAPIC_BASE_ADDRESS  0xfee00000UL
> -#define GB(x)               ((uint64_t)x << 30)

With this change we have this change for GB(memory + 0) when applied:
-  ((uint64_t)memory + 0 << 30)
+  ((memory + 0ULL) << 30)


So overall, patch makes things better and less duplication:
Acked-by: Anthony PERARD <anthony.perard@citrix.com>

Thanks,

-- 
Anthony PERARD

[PATCH v6 3/4] libelf: Store maximum PHDR p_align

[Jason Andryuk]

While parsing the PHDRs, store the maximum p_align value.  This may be
consulted for moving a PVH image's load address.

Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
---
v6:
New
---
 xen/common/libelf/libelf-loader.c | 15 +++++++++++----
 xen/include/xen/libelf.h          |  1 +
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/xen/common/libelf/libelf-loader.c b/xen/common/libelf/libelf-loader.c
index 629cc0d3e6..a5f6389f82 100644
--- a/xen/common/libelf/libelf-loader.c
+++ b/xen/common/libelf/libelf-loader.c
@@ -468,6 +468,7 @@ void elf_parse_binary(struct elf_binary *elf)
 {
     ELF_HANDLE_DECL(elf_phdr) phdr;
     uint64_t low = -1, high = 0, paddr, memsz;
+    uint64_t max_align = 0, palign;
     unsigned i, count;
 
     count = elf_phdr_count(elf);
@@ -481,17 +482,23 @@ void elf_parse_binary(struct elf_binary *elf)
             continue;
         paddr = elf_uval(elf, phdr, p_paddr);
         memsz = elf_uval(elf, phdr, p_memsz);
-        elf_msg(elf, "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 "\n",
-                paddr, memsz);
+        palign = elf_uval(elf, phdr, p_align);
+        elf_msg(elf,
+                "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 " palign=%#" PRIx64 "\n",
+                paddr, memsz, palign);
         if ( low > paddr )
             low = paddr;
         if ( high < paddr + memsz )
             high = paddr + memsz;
+        if ( max_align < palign )
+            max_align = palign;
     }
     elf->pstart = low;
     elf->pend = high;
-    elf_msg(elf, "ELF: memory: %#" PRIx64 " -> %#" PRIx64 "\n",
-            elf->pstart, elf->pend);
+    elf->palign = max_align;
+    elf_msg(elf,
+            "ELF: memory: %#" PRIx64 " -> %#" PRIx64 " align:%#" PRIx64 "\n",
+            elf->pstart, elf->pend, elf->palign);
 }
 
 elf_errorstatus elf_load_binary(struct elf_binary *elf)
diff --git a/xen/include/xen/libelf.h b/xen/include/xen/libelf.h
index 1c77e3df31..2d971f958e 100644
--- a/xen/include/xen/libelf.h
+++ b/xen/include/xen/libelf.h
@@ -196,6 +196,7 @@ struct elf_binary {
     size_t dest_size;
     uint64_t pstart;
     uint64_t pend;
+    uint64_t palign;
     uint64_t reloc_offset;
 
     uint64_t bsd_symtab_pstart;
-- 
2.44.0

Re: [PATCH v6 3/4] libelf: Store maximum PHDR p_align

[Jan Beulich]

On 27.03.2024 22:51, Jason Andryuk wrote:
> --- a/xen/common/libelf/libelf-loader.c
> +++ b/xen/common/libelf/libelf-loader.c
> @@ -468,6 +468,7 @@ void elf_parse_binary(struct elf_binary *elf)
>  {
>      ELF_HANDLE_DECL(elf_phdr) phdr;
>      uint64_t low = -1, high = 0, paddr, memsz;
> +    uint64_t max_align = 0, palign;
>      unsigned i, count;
>  
>      count = elf_phdr_count(elf);
> @@ -481,17 +482,23 @@ void elf_parse_binary(struct elf_binary *elf)
>              continue;
>          paddr = elf_uval(elf, phdr, p_paddr);
>          memsz = elf_uval(elf, phdr, p_memsz);
> -        elf_msg(elf, "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 "\n",
> -                paddr, memsz);
> +        palign = elf_uval(elf, phdr, p_align);
> +        elf_msg(elf,
> +                "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 " palign=%#" PRIx64 "\n",
> +                paddr, memsz, palign);
>          if ( low > paddr )
>              low = paddr;
>          if ( high < paddr + memsz )
>              high = paddr + memsz;
> +        if ( max_align < palign )
> +            max_align = palign;
>      }
>      elf->pstart = low;
>      elf->pend = high;
> -    elf_msg(elf, "ELF: memory: %#" PRIx64 " -> %#" PRIx64 "\n",
> -            elf->pstart, elf->pend);
> +    elf->palign = max_align;
> +    elf_msg(elf,
> +            "ELF: memory: %#" PRIx64 " -> %#" PRIx64 " align:%#" PRIx64 "\n",
> +            elf->pstart, elf->pend, elf->palign);
>  }

Hmm, it's just this final logging change which I'm a little concerned by:
Having looked at Linux'es phdr, I noticed that the addresses there aren't
necessarily matching the corresponding alignment. Therefore I'm a little
concerned that the output here might raise questions when people see
seemingly inconsistent values in the log. Could you/we at least make it
read like e.g. "align (max): ..."?

Jan

Re: [PATCH v6 3/4] libelf: Store maximum PHDR p_align

[Jason Andryuk]

On 2024-03-28 12:47, Jan Beulich wrote:
> On 27.03.2024 22:51, Jason Andryuk wrote:
>> --- a/xen/common/libelf/libelf-loader.c
>> +++ b/xen/common/libelf/libelf-loader.c
>> @@ -468,6 +468,7 @@ void elf_parse_binary(struct elf_binary *elf)
>>   {
>>       ELF_HANDLE_DECL(elf_phdr) phdr;
>>       uint64_t low = -1, high = 0, paddr, memsz;
>> +    uint64_t max_align = 0, palign;
>>       unsigned i, count;
>>   
>>       count = elf_phdr_count(elf);
>> @@ -481,17 +482,23 @@ void elf_parse_binary(struct elf_binary *elf)
>>               continue;
>>           paddr = elf_uval(elf, phdr, p_paddr);
>>           memsz = elf_uval(elf, phdr, p_memsz);
>> -        elf_msg(elf, "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 "\n",
>> -                paddr, memsz);
>> +        palign = elf_uval(elf, phdr, p_align);
>> +        elf_msg(elf,
>> +                "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 " palign=%#" PRIx64 "\n",
>> +                paddr, memsz, palign);
>>           if ( low > paddr )
>>               low = paddr;
>>           if ( high < paddr + memsz )
>>               high = paddr + memsz;
>> +        if ( max_align < palign )
>> +            max_align = palign;
>>       }
>>       elf->pstart = low;
>>       elf->pend = high;
>> -    elf_msg(elf, "ELF: memory: %#" PRIx64 " -> %#" PRIx64 "\n",
>> -            elf->pstart, elf->pend);
>> +    elf->palign = max_align;
>> +    elf_msg(elf,
>> +            "ELF: memory: %#" PRIx64 " -> %#" PRIx64 " align:%#" PRIx64 "\n",
>> +            elf->pstart, elf->pend, elf->palign);
>>   }
> 
> Hmm, it's just this final logging change which I'm a little concerned by:
> Having looked at Linux'es phdr, I noticed that the addresses there aren't
> necessarily matching the corresponding alignment. Therefore I'm a little
> concerned that the output here might raise questions when people see
> seemingly inconsistent values in the log. Could you/we at least make it
> read like e.g. "align (max): ..."?

max_align?

Looking at my test vmlinux, it looks like PHDR 0 (.text) and PHDR 1 
(.data) are aligned.  It's the PHDR 2 (.init) that isn't aligned.  Is 
that what you see?

This line is already printing the min and max across all the PHDRs, so 
it would only look confusing if the start didn't match the align.

I'm not sure how useful it is to print the alignment, and I considered 
not printing it.  It's only used with PVH Dom0 right now, so it's not 
relevant in most cases.

Regards,
Jason

Re: [PATCH v6 3/4] libelf: Store maximum PHDR p_align

[Jan Beulich]

On 29.03.2024 15:41, Jason Andryuk wrote:
> On 2024-03-28 12:47, Jan Beulich wrote:
>> On 27.03.2024 22:51, Jason Andryuk wrote:
>>> --- a/xen/common/libelf/libelf-loader.c
>>> +++ b/xen/common/libelf/libelf-loader.c
>>> @@ -468,6 +468,7 @@ void elf_parse_binary(struct elf_binary *elf)
>>>   {
>>>       ELF_HANDLE_DECL(elf_phdr) phdr;
>>>       uint64_t low = -1, high = 0, paddr, memsz;
>>> +    uint64_t max_align = 0, palign;
>>>       unsigned i, count;
>>>   
>>>       count = elf_phdr_count(elf);
>>> @@ -481,17 +482,23 @@ void elf_parse_binary(struct elf_binary *elf)
>>>               continue;
>>>           paddr = elf_uval(elf, phdr, p_paddr);
>>>           memsz = elf_uval(elf, phdr, p_memsz);
>>> -        elf_msg(elf, "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 "\n",
>>> -                paddr, memsz);
>>> +        palign = elf_uval(elf, phdr, p_align);
>>> +        elf_msg(elf,
>>> +                "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 " palign=%#" PRIx64 "\n",
>>> +                paddr, memsz, palign);
>>>           if ( low > paddr )
>>>               low = paddr;
>>>           if ( high < paddr + memsz )
>>>               high = paddr + memsz;
>>> +        if ( max_align < palign )
>>> +            max_align = palign;
>>>       }
>>>       elf->pstart = low;
>>>       elf->pend = high;
>>> -    elf_msg(elf, "ELF: memory: %#" PRIx64 " -> %#" PRIx64 "\n",
>>> -            elf->pstart, elf->pend);
>>> +    elf->palign = max_align;
>>> +    elf_msg(elf,
>>> +            "ELF: memory: %#" PRIx64 " -> %#" PRIx64 " align:%#" PRIx64 "\n",
>>> +            elf->pstart, elf->pend, elf->palign);
>>>   }
>>
>> Hmm, it's just this final logging change which I'm a little concerned by:
>> Having looked at Linux'es phdr, I noticed that the addresses there aren't
>> necessarily matching the corresponding alignment. Therefore I'm a little
>> concerned that the output here might raise questions when people see
>> seemingly inconsistent values in the log. Could you/we at least make it
>> read like e.g. "align (max): ..."?
> 
> max_align?
> 
> Looking at my test vmlinux, it looks like PHDR 0 (.text) and PHDR 1 
> (.data) are aligned.  It's the PHDR 2 (.init) that isn't aligned.  Is 
> that what you see?
> 
> This line is already printing the min and max across all the PHDRs, so 
> it would only look confusing if the start didn't match the align.
> 
> I'm not sure how useful it is to print the alignment, and I considered 
> not printing it.  It's only used with PVH Dom0 right now, so it's not 
> relevant in most cases.

Well, yes, not printing the alignment would be fine with me. I just didn't
suggest that as an alternative since I was assuming you put its printing
there for a reason.

Jan

Re: [PATCH v6 3/4] libelf: Store maximum PHDR p_align

[Jason Andryuk]

On 2024-04-02 02:44, Jan Beulich wrote:
> On 29.03.2024 15:41, Jason Andryuk wrote:
>> On 2024-03-28 12:47, Jan Beulich wrote:
>>> On 27.03.2024 22:51, Jason Andryuk wrote:
>>>> --- a/xen/common/libelf/libelf-loader.c
>>>> +++ b/xen/common/libelf/libelf-loader.c
>>>> @@ -468,6 +468,7 @@ void elf_parse_binary(struct elf_binary *elf)
>>>>    {
>>>>        ELF_HANDLE_DECL(elf_phdr) phdr;
>>>>        uint64_t low = -1, high = 0, paddr, memsz;
>>>> +    uint64_t max_align = 0, palign;
>>>>        unsigned i, count;
>>>>    
>>>>        count = elf_phdr_count(elf);
>>>> @@ -481,17 +482,23 @@ void elf_parse_binary(struct elf_binary *elf)
>>>>                continue;
>>>>            paddr = elf_uval(elf, phdr, p_paddr);
>>>>            memsz = elf_uval(elf, phdr, p_memsz);
>>>> -        elf_msg(elf, "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 "\n",
>>>> -                paddr, memsz);
>>>> +        palign = elf_uval(elf, phdr, p_align);
>>>> +        elf_msg(elf,
>>>> +                "ELF: phdr: paddr=%#" PRIx64 " memsz=%#" PRIx64 " palign=%#" PRIx64 "\n",
>>>> +                paddr, memsz, palign);
>>>>            if ( low > paddr )
>>>>                low = paddr;
>>>>            if ( high < paddr + memsz )
>>>>                high = paddr + memsz;
>>>> +        if ( max_align < palign )
>>>> +            max_align = palign;
>>>>        }
>>>>        elf->pstart = low;
>>>>        elf->pend = high;
>>>> -    elf_msg(elf, "ELF: memory: %#" PRIx64 " -> %#" PRIx64 "\n",
>>>> -            elf->pstart, elf->pend);
>>>> +    elf->palign = max_align;
>>>> +    elf_msg(elf,
>>>> +            "ELF: memory: %#" PRIx64 " -> %#" PRIx64 " align:%#" PRIx64 "\n",
>>>> +            elf->pstart, elf->pend, elf->palign);
>>>>    }
>>>
>>> Hmm, it's just this final logging change which I'm a little concerned by:
>>> Having looked at Linux'es phdr, I noticed that the addresses there aren't
>>> necessarily matching the corresponding alignment. Therefore I'm a little
>>> concerned that the output here might raise questions when people see
>>> seemingly inconsistent values in the log. Could you/we at least make it
>>> read like e.g. "align (max): ..."?
>>
>> max_align?
>>
>> Looking at my test vmlinux, it looks like PHDR 0 (.text) and PHDR 1
>> (.data) are aligned.  It's the PHDR 2 (.init) that isn't aligned.  Is
>> that what you see?
>>
>> This line is already printing the min and max across all the PHDRs, so
>> it would only look confusing if the start didn't match the align.
>>
>> I'm not sure how useful it is to print the alignment, and I considered
>> not printing it.  It's only used with PVH Dom0 right now, so it's not
>> relevant in most cases.
> 
> Well, yes, not printing the alignment would be fine with me. I just didn't
> suggest that as an alternative since I was assuming you put its printing
> there for a reason.

I added it to make the information available, which might be useful if 
there was a problem finding a load location.  Since it's usually not 
useful information, I'll just drop it.

Thanks,
Jason

[PATCH v6 4/4] x86/PVH: Support relocatable dom0 kernels

[Jason Andryuk]

Xen tries to load a PVH dom0 kernel at the fixed guest physical address
from the elf headers.  For Linux, this defaults to 0x1000000 (16MB), but
it can be configured.

Unfortunately there exist firmwares that have reserved regions at this
address, so Xen fails to load the dom0 kernel since it's not RAM.

The PVH entry code is not relocatable - it loads from absolute
addresses, which fail when the kernel is loaded at a different address.
With a suitably modified kernel, a reloctable entry point is possible.

Add XEN_ELFNOTE_PHYS32_RELOC which specifies optional alignment,
minimum, and maximum addresses needed for the kernel.  The presence of
the NOTE indicates the kernel supports a relocatable entry path.

Change the loading to check for an acceptable load address.  If the
kernel is relocatable, support finding an alternate load address.

The primary motivation for an explicit align field is that Linux has a
configurable CONFIG_PHYSICAL_ALIGN field.  This value is present in the
bzImage setup header, but not the ELF program headers p_align, which
report 2MB even when CONFIG_PHYSICAL_ALIGN is greater.  Since a kernel
is only considered relocatable if the PHYS32_RELOC elf note is present,
the alignment contraints can just be specified within the note instead
of searching for an alignment value via a heuristic.

Load alignment uses the PHYS32_RELOC note value if specified.
Otherwise, the maxmum ELF PHDR p_align value is selected if greater than
or equal to PAGE_SIZE.  Finally, the fallback default is 2MB.

libelf-private.h includes common-macros.h to satisfy the fuzzer build.

Link: https://gitlab.com/xen-project/xen/-/issues/180
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
---
ELF Note printing looks like:
(XEN) ELF: note: PHYS32_RELOC align: 0x200000 min: 0x1000000 max: 0x3fffffff

v2:
Use elfnote for min, max & align - use 64bit values.
Print original and relocated memory addresses
Use check_and_adjust_load_address() name
Return relocated base instead of offset
Use PAGE_ALIGN
Don't load above max_phys (expected to be 4GB in kernel elf note)
Use single line comments
Exit check_load_address loop earlier
Add __init to find_kernel_memory()

v3:
Remove kernel_start/end page rounding
Change loop comment to rely on a sorted memory map.
Reorder E820_RAM check first
Use %p for dest_base
Use PRIpaddr
Use 32bit phys_min/max/align
Consolidate to if ( x || y ) continue
Use max_t
Add parms->phys_reloc
Use common "%pd kernel: " prefix for messages
Re-order phys_entry assignment
Print range ends inclusively
Remove extra "Unable to load kernel" message
s/PVH_RELOCATION/PHYS32_RELOC/
Make PHYS32_RELOC contents optional
Use 2MB default alignment
Update ELF Note comment
Update XEN_ELFNOTE_MAX

v4:
Cast dest_base to uintptr_t
Use local start variable
Mention e820 requiring adjacent entries merged
Remove extra whitespace
Re-word elfnote comment & mention x86
Use ELFNOTE_NAME
Return -ENOSPC
Use ! instead of == 0
Check kend - 1 to avoid off by one
libelf: Use MB/GB() to define the size.
Use ARCH_PHYS_* defines

v5:
Place kernel in higher memory addresses
Remove stray semicolons
ELFNOTE_NAME comment about newline
Make PHYS32_RELOC element order align, min, max
Re-word PHYS32_RELOC comment
Move phys_align next to other bool variables

v6:
Select alignment from, in order, Note, PHDRs, then default
---
 xen/arch/x86/hvm/dom0_build.c      | 108 +++++++++++++++++++++++++++++
 xen/common/libelf/libelf-dominfo.c |  35 ++++++++++
 xen/common/libelf/libelf-private.h |   1 +
 xen/include/public/elfnote.h       |  16 ++++-
 xen/include/xen/libelf.h           |   4 ++
 5 files changed, 163 insertions(+), 1 deletion(-)

diff --git a/xen/arch/x86/hvm/dom0_build.c b/xen/arch/x86/hvm/dom0_build.c
index 0ceda4140b..01f39d650e 100644
--- a/xen/arch/x86/hvm/dom0_build.c
+++ b/xen/arch/x86/hvm/dom0_build.c
@@ -537,6 +537,111 @@ static paddr_t __init find_memory(
     return INVALID_PADDR;
 }
 
+static bool __init check_load_address(
+    const struct domain *d, const struct elf_binary *elf)
+{
+    paddr_t kernel_start = (uintptr_t)elf->dest_base;
+    paddr_t kernel_end = kernel_start + elf->dest_size;
+    unsigned int i;
+
+    /* Relies on a sorted memory map with adjacent entries merged. */
+    for ( i = 0; i < d->arch.nr_e820; i++ )
+    {
+        paddr_t start = d->arch.e820[i].addr;
+        paddr_t end = start + d->arch.e820[i].size;
+
+        if ( start >= kernel_end )
+            return false;
+
+        if ( d->arch.e820[i].type == E820_RAM &&
+             start <= kernel_start &&
+             end >= kernel_end )
+            return true;
+    }
+
+    return false;
+}
+
+/* Find an e820 RAM region that fits the kernel at a suitable alignment. */
+static paddr_t __init find_kernel_memory(
+    const struct domain *d, struct elf_binary *elf,
+    const struct elf_dom_parms *parms)
+{
+    paddr_t kernel_size = elf->dest_size;
+    unsigned int align;
+    int i;
+
+    if ( parms->phys_align != UNSET_ADDR32 )
+        align = parms->phys_align;
+    else if ( elf->palign > PAGE_SIZE )
+        align = elf->palign;
+    else
+        align = MB(2);
+
+    /* Search backwards to find the highest address. */
+    for ( i = d->arch.nr_e820 - 1; i >= 0 ; i-- )
+    {
+        paddr_t start = d->arch.e820[i].addr;
+        paddr_t end = start + d->arch.e820[i].size;
+        paddr_t kstart, kend;
+
+        if ( d->arch.e820[i].type != E820_RAM ||
+             d->arch.e820[i].size < kernel_size )
+            continue;
+
+        if ( start > parms->phys_max )
+            continue;
+
+        if ( end - 1 > parms->phys_max )
+            end = parms->phys_max + 1;
+
+        kstart = (end - kernel_size) & ~(align - 1);
+        kend = kstart + kernel_size;
+
+        if ( kstart < parms->phys_min )
+            return 0;
+
+        if ( kstart >= start && kend <= end )
+            return kstart;
+    }
+
+    return 0;
+}
+
+/* Check the kernel load address, and adjust if necessary and possible. */
+static bool __init check_and_adjust_load_address(
+    const struct domain *d, struct elf_binary *elf, struct elf_dom_parms *parms)
+{
+    paddr_t reloc_base;
+
+    if ( check_load_address(d, elf) )
+        return true;
+
+    if ( !parms->phys_reloc )
+    {
+        printk("%pd kernel: Address conflict and not relocatable\n", d);
+        return false;
+    }
+
+    reloc_base = find_kernel_memory(d, elf, parms);
+    if ( !reloc_base )
+    {
+        printk("%pd kernel: Failed find a load address\n", d);
+        return false;
+    }
+
+    if ( opt_dom0_verbose )
+        printk("%pd kernel: Moving [%p, %p] -> [%"PRIpaddr", %"PRIpaddr"]\n", d,
+               elf->dest_base, elf->dest_base + elf->dest_size - 1,
+               reloc_base, reloc_base + elf->dest_size - 1);
+
+    parms->phys_entry = reloc_base +
+                            (parms->phys_entry - (uintptr_t)elf->dest_base);
+    elf->dest_base = (char *)reloc_base;
+
+    return true;
+}
+
 static int __init pvh_load_kernel(struct domain *d, const module_t *image,
                                   unsigned long image_headroom,
                                   module_t *initrd, void *image_base,
@@ -585,6 +690,9 @@ static int __init pvh_load_kernel(struct domain *d, const module_t *image,
     elf.dest_base = (void *)(parms.virt_kstart - parms.virt_base);
     elf.dest_size = parms.virt_kend - parms.virt_kstart;
 
+    if ( !check_and_adjust_load_address(d, &elf, &parms) )
+        return -ENOSPC;
+
     elf_set_vcpu(&elf, v);
     rc = elf_load_binary(&elf);
     if ( rc < 0 )
diff --git a/xen/common/libelf/libelf-dominfo.c b/xen/common/libelf/libelf-dominfo.c
index 0cdb419b8a..260294d0ed 100644
--- a/xen/common/libelf/libelf-dominfo.c
+++ b/xen/common/libelf/libelf-dominfo.c
@@ -17,6 +17,14 @@
 
 #include "libelf-private.h"
 
+#if defined(__i386__) || defined(__x86_64__)
+#define ARCH_PHYS_MIN_DEFAULT   0
+#define ARCH_PHYS_MAX_DEFAULT   (GB(4) - 1)
+#else
+#define ARCH_PHYS_MIN_DEFAULT   0
+#define ARCH_PHYS_MAX_DEFAULT   0
+#endif
+
 /* ------------------------------------------------------------------------ */
 /* xen features                                                             */
 
@@ -125,6 +133,7 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf,
         [XEN_ELFNOTE_SUSPEND_CANCEL] = { "SUSPEND_CANCEL", ELFNOTE_INT },
         [XEN_ELFNOTE_MOD_START_PFN] = { "MOD_START_PFN", ELFNOTE_INT },
         [XEN_ELFNOTE_PHYS32_ENTRY] = { "PHYS32_ENTRY", ELFNOTE_INT },
+        [XEN_ELFNOTE_PHYS32_RELOC] = { "PHYS32_RELOC", ELFNOTE_NAME },
     };
 /* *INDENT-ON* */
 
@@ -132,6 +141,7 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf,
     uint64_t val = 0;
     unsigned int i;
     unsigned type = elf_uval(elf, note, type);
+    unsigned descsz = elf_uval(elf, note, descsz);
 
     if ( (type >= sizeof(note_desc) / sizeof(note_desc[0])) ||
          (note_desc[type].name == NULL) )
@@ -228,6 +238,27 @@ elf_errorstatus elf_xen_parse_note(struct elf_binary *elf,
     case XEN_ELFNOTE_PHYS32_ENTRY:
         parms->phys_entry = val;
         break;
+
+    case XEN_ELFNOTE_PHYS32_RELOC:
+        parms->phys_reloc = true;
+
+        if ( descsz >= 4 )
+        {
+            parms->phys_align = elf_note_numeric_array(elf, note, 4, 0);
+            elf_msg(elf, " align: %#"PRIx32, parms->phys_align);
+        }
+        if ( descsz >= 8 )
+        {
+            parms->phys_min = elf_note_numeric_array(elf, note, 4, 1);
+            elf_msg(elf, " min: %#"PRIx32, parms->phys_min);
+        }
+        if ( descsz >= 12 )
+        {
+            parms->phys_max = elf_note_numeric_array(elf, note, 4, 2);
+            elf_msg(elf, " max: %#"PRIx32, parms->phys_max);
+        }
+
+        break;
     }
 
     if ( note_desc[type].type == ELFNOTE_NAME)
@@ -543,6 +574,10 @@ elf_errorstatus elf_xen_parse(struct elf_binary *elf,
     parms->p2m_base = UNSET_ADDR;
     parms->elf_paddr_offset = UNSET_ADDR;
     parms->phys_entry = UNSET_ADDR32;
+    parms->phys_align = UNSET_ADDR32;
+    parms->phys_min = ARCH_PHYS_MIN_DEFAULT;
+    parms->phys_max = ARCH_PHYS_MAX_DEFAULT;
+    parms->phys_reloc = false;
 
     /* Find and parse elf notes. */
     count = elf_phdr_count(elf);
diff --git a/xen/common/libelf/libelf-private.h b/xen/common/libelf/libelf-private.h
index 47db679966..98cac65bc5 100644
--- a/xen/common/libelf/libelf-private.h
+++ b/xen/common/libelf/libelf-private.h
@@ -71,6 +71,7 @@
 #endif
 #include <xen/elfnote.h>
 #include <xen/libelf/libelf.h>
+#include <xen-tools/common-macros.h>
 
 #ifndef FUZZ_NO_LIBXC
 #include "xenctrl.h"
diff --git a/xen/include/public/elfnote.h b/xen/include/public/elfnote.h
index 1d84b05f44..a2f89bdc40 100644
--- a/xen/include/public/elfnote.h
+++ b/xen/include/public/elfnote.h
@@ -196,10 +196,24 @@
  */
 #define XEN_ELFNOTE_PHYS32_ENTRY 18
 
+/*
+ * Physical loading constraints for PVH kernels
+ *
+ * The presence of this note indicates the kernel supports relocating itself.
+ *
+ * The note may include up to three 32bit values to place constraints on the
+ * guest physical loading addresses and alignment for a PVH kernel.  Values
+ * are read in the following order:
+ *  - a required start alignment (default 0x200000)
+ *  - a minimum address for the start of the image (default 0)
+ *  - a maximum address for the last byte of the image (default 0xffffffff)
+ */
+#define XEN_ELFNOTE_PHYS32_RELOC 19
+
 /*
  * The number of the highest elfnote defined.
  */
-#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_ENTRY
+#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_RELOC
 
 /*
  * System information exported through crash notes.
diff --git a/xen/include/xen/libelf.h b/xen/include/xen/libelf.h
index 2d971f958e..9ac530acc2 100644
--- a/xen/include/xen/libelf.h
+++ b/xen/include/xen/libelf.h
@@ -426,6 +426,7 @@ struct elf_dom_parms {
     enum xen_pae_type pae;
     bool bsd_symtab;
     bool unmapped_initrd;
+    bool phys_reloc;
     uint64_t virt_base;
     uint64_t virt_entry;
     uint64_t virt_hypercall;
@@ -435,6 +436,9 @@ struct elf_dom_parms {
     uint32_t f_supported[XENFEAT_NR_SUBMAPS];
     uint32_t f_required[XENFEAT_NR_SUBMAPS];
     uint32_t phys_entry;
+    uint32_t phys_align;
+    uint32_t phys_min;
+    uint32_t phys_max;
 
     /* calculated */
     uint64_t virt_kstart;
-- 
2.44.0

Re: [PATCH v6 4/4] x86/PVH: Support relocatable dom0 kernels

[Jan Beulich]

On 27.03.2024 22:51, Jason Andryuk wrote:
> v6:
> Select alignment from, in order, Note, PHDRs, then default

The comment in the public header also needs to reflect this change.

> +static bool __init check_and_adjust_load_address(
> +    const struct domain *d, struct elf_binary *elf, struct elf_dom_parms *parms)
> +{
> +    paddr_t reloc_base;
> +
> +    if ( check_load_address(d, elf) )
> +        return true;
> +
> +    if ( !parms->phys_reloc )
> +    {
> +        printk("%pd kernel: Address conflict and not relocatable\n", d);
> +        return false;
> +    }
> +
> +    reloc_base = find_kernel_memory(d, elf, parms);
> +    if ( !reloc_base )
> +    {
> +        printk("%pd kernel: Failed find a load address\n", d);
> +        return false;
> +    }
> +
> +    if ( opt_dom0_verbose )
> +        printk("%pd kernel: Moving [%p, %p] -> [%"PRIpaddr", %"PRIpaddr"]\n", d,
> +               elf->dest_base, elf->dest_base + elf->dest_size - 1,
> +               reloc_base, reloc_base + elf->dest_size - 1);
> +
> +    parms->phys_entry = reloc_base +
> +                            (parms->phys_entry - (uintptr_t)elf->dest_base);

I think this would be easier to read as

    parms->phys_entry =
        reloc_base + (parms->phys_entry - (uintptr_t)elf->dest_base);

Everything else looks good to me now.

Jan

Re: [PATCH v6 4/4] x86/PVH: Support relocatable dom0 kernels

[Jason Andryuk]

On 2024-04-02 10:34, Jan Beulich wrote:
> On 27.03.2024 22:51, Jason Andryuk wrote:
>> v6:
>> Select alignment from, in order, Note, PHDRs, then default
> 
> The comment in the public header also needs to reflect this change.

Yes.  I'll update.

I'm going to tweak the code from elf->palign > PAGE_SIZE to >=, since 
PAGE_SIZE is a reasonable value to use.

>> +static bool __init check_and_adjust_load_address(
>> +    const struct domain *d, struct elf_binary *elf, struct elf_dom_parms *parms)
>> +{
>> +    paddr_t reloc_base;
>> +
>> +    if ( check_load_address(d, elf) )
>> +        return true;
>> +
>> +    if ( !parms->phys_reloc )
>> +    {
>> +        printk("%pd kernel: Address conflict and not relocatable\n", d);
>> +        return false;
>> +    }
>> +
>> +    reloc_base = find_kernel_memory(d, elf, parms);
>> +    if ( !reloc_base )
>> +    {
>> +        printk("%pd kernel: Failed find a load address\n", d);
>> +        return false;
>> +    }
>> +
>> +    if ( opt_dom0_verbose )
>> +        printk("%pd kernel: Moving [%p, %p] -> [%"PRIpaddr", %"PRIpaddr"]\n", d,
>> +               elf->dest_base, elf->dest_base + elf->dest_size - 1,
>> +               reloc_base, reloc_base + elf->dest_size - 1);
>> +
>> +    parms->phys_entry = reloc_base +
>> +                            (parms->phys_entry - (uintptr_t)elf->dest_base);
> 
> I think this would be easier to read as
> 
>      parms->phys_entry =
>          reloc_base + (parms->phys_entry - (uintptr_t)elf->dest_base);
> 
> Everything else looks good to me now.

Sounds good to me.

Thanks,
Jason

[PATCH v6] RFC: x86/pvh: Make Xen PVH entrypoint PIC for x86-64

[Jason Andryuk]

The Xen PVH entrypoint is 32bit non-PIC code running at a default load
address of 0x1000000 (16MB) (CONFIG_PHYSICAL_START).  Xen loads the
kernel at that physical address inside the PVH container.

When running a PVH Dom0, the system reserved addresses are mapped 1-1
into the PVH container.  There exist system firmwares (Coreboot/EDK2)
with reserved memory at 16MB.  This creates a conflict where the PVH
kernel cannot be loaded at that address.

Modify the PVH entrypoint to be position-indepedent to allow flexibility
in load address.  Only the 64bit entry path is converted.  A 32bit
kernel is not PIC, so calling into other parts of the kernel, like
xen_prepare_pvh() and mk_pgtable_32(), don't work properly when
relocated.

Initial PVH entry runs at the physical addresses and then transitions to
the identity mapped address.  While executing xen_prepare_pvh() calls
through pv_ops function pointers transition to the high mapped
addresses.  Additionally, __va() is called on some hvm_start_info
physical addresses, we need the directmap address range is used.  So we
need to run page tables with all of those ranges mapped.

Modifying init_top_pgt tables ran into issue since
startup_64/__startup_64() will modify those page tables again.  Use a
dedicated set of page tables - pvh_init_top_pgt  - for the PVH entry to
avoid unwanted interactions.

In xen_pvh_init(), __pa() is called to find the physical address of the
hypercall page.  Set phys_base temporarily before calling into
xen_prepare_pvh(), which calls xen_pvh_init(), and clear it afterwards.
__startup_64() assumes phys_base is zero and adds load_delta to it.  If
phys_base is already set, the calculation results in an incorrect cr3.

TODO: Sync elfnote.h from xen.git commit xxxxxxxxxx when it is
commited.

Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
---
Put this out as an example for the Xen modifications

Instead of setting and clearing phys_base, add a dedicated variable?
Clearing phys_base is a little weird, but it leaves the kernel more
consistent when running non-entry code.

Make __startup_64() exit if phys_base is already set to allow calling
multiple times, and use that and init_top_pgt instead of adding
additional page tables?  That won't work.  __startup_64 is 64bit code,
and pvh needs to create page tables in 32bit code before it can
transition to 64bit long mode.  Hence it can't be re-used to relocate
page tables.
---
 arch/x86/platform/pvh/head.S    | 182 +++++++++++++++++++++++++++++---
 include/xen/interface/elfnote.h |  16 ++-
 2 files changed, 185 insertions(+), 13 deletions(-)

diff --git a/arch/x86/platform/pvh/head.S b/arch/x86/platform/pvh/head.S
index f7235ef87bc3..ce2b201210e3 100644
--- a/arch/x86/platform/pvh/head.S
+++ b/arch/x86/platform/pvh/head.S
@@ -50,11 +50,32 @@
 #define PVH_CS_SEL		(PVH_GDT_ENTRY_CS * 8)
 #define PVH_DS_SEL		(PVH_GDT_ENTRY_DS * 8)
 
+#define rva(x) ((x) - pvh_start_xen)
+
 SYM_CODE_START_LOCAL(pvh_start_xen)
 	UNWIND_HINT_END_OF_STACK
 	cld
 
-	lgdt (_pa(gdt))
+	/*
+	 * See the comment for startup_32 for more details.  We need to
+	 * execute a call to get the execution address to be position
+	 * independent, but we don't have a stack.  Save and restore the
+	 * magic field of start_info in ebx, and use that as the stack.
+	 */
+	mov	(%ebx), %eax
+	leal	4(%ebx), %esp
+	ANNOTATE_INTRA_FUNCTION_CALL
+	call	1f
+1:	popl	%ebp
+	mov	%eax, (%ebx)
+	subl	$ rva(1b), %ebp
+	movl	$0, %esp
+
+	leal	rva(gdt)(%ebp), %eax
+	movl	%eax, %ecx
+	leal	rva(gdt_start)(%ebp), %ecx
+	movl	%ecx, 2(%eax)
+	lgdt	(%eax)
 
 	mov $PVH_DS_SEL,%eax
 	mov %eax,%ds
@@ -62,14 +83,14 @@ SYM_CODE_START_LOCAL(pvh_start_xen)
 	mov %eax,%ss
 
 	/* Stash hvm_start_info. */
-	mov $_pa(pvh_start_info), %edi
+	leal rva(pvh_start_info)(%ebp), %edi
 	mov %ebx, %esi
-	mov _pa(pvh_start_info_sz), %ecx
+	movl rva(pvh_start_info_sz)(%ebp), %ecx
 	shr $2,%ecx
 	rep
 	movsl
 
-	mov $_pa(early_stack_end), %esp
+	leal rva(early_stack_end)(%ebp), %esp
 
 	/* Enable PAE mode. */
 	mov %cr4, %eax
@@ -83,29 +104,81 @@ SYM_CODE_START_LOCAL(pvh_start_xen)
 	btsl $_EFER_LME, %eax
 	wrmsr
 
+	mov %ebp, %ebx
+	subl $LOAD_PHYSICAL_ADDR, %ebx /* offset */
+	jz .Lpagetable_done
+
+	/* Fixup page-tables for relocation. */
+	leal rva(pvh_init_top_pgt)(%ebp), %edi
+	movl $512, %ecx
+2:
+	testl $_PAGE_PRESENT, 0x00(%edi)
+	jz 1f
+	addl %ebx, 0x00(%edi)
+1:
+	addl $8, %edi
+	decl %ecx
+	jnz 2b
+
+	/* L3 ident has a single entry. */
+	leal rva(pvh_level3_ident_pgt)(%ebp), %edi
+	addl %ebx, 0x00(%edi)
+
+	leal rva(pvh_level3_kernel_pgt)(%ebp), %edi
+	addl %ebx, (4096 - 16)(%edi)
+	addl %ebx, (4096 - 8)(%edi)
+
+	/* pvh_level2_ident_pgt is fine - large pages */
+
+	/* pvh_level2_kernel_pgt needs adjustment - large pages */
+	leal rva(pvh_level2_kernel_pgt)(%ebp), %edi
+	movl $512, %ecx
+2:
+	testl $_PAGE_PRESENT, 0x00(%edi)
+	jz 1f
+	addl %ebx, 0x00(%edi)
+1:
+	addl $8, %edi
+	decl %ecx
+	jnz 2b
+
+.Lpagetable_done:
 	/* Enable pre-constructed page tables. */
-	mov $_pa(init_top_pgt), %eax
+	leal rva(pvh_init_top_pgt)(%ebp), %eax
 	mov %eax, %cr3
 	mov $(X86_CR0_PG | X86_CR0_PE), %eax
 	mov %eax, %cr0
 
 	/* Jump to 64-bit mode. */
-	ljmp $PVH_CS_SEL, $_pa(1f)
+	pushl $PVH_CS_SEL
+	leal  rva(1f)(%ebp), %eax
+	pushl %eax
+	lretl
 
 	/* 64-bit entry point. */
 	.code64
 1:
 	/* Set base address in stack canary descriptor. */
 	mov $MSR_GS_BASE,%ecx
-	mov $_pa(canary), %eax
+	leal rva(canary)(%ebp), %eax
 	xor %edx, %edx
 	wrmsr
 
+	/* Calculate load offset from LOAD_PHYSICAL_ADDR and store in
+	 * phys_base.  __pa() needs phys_base set to calculate the the
+	 * hypercall page in xen_pvh_init(). */
+	movq %rbp, %rbx
+	subq $LOAD_PHYSICAL_ADDR, %rbx
+	movq %rbx, phys_base(%rip)
 	call xen_prepare_pvh
+	/* Clear phys_base.  startup_64/__startup_64 will *add* to its value,
+	   so start from 0. */
+	xor  %rbx, %rbx
+	movq %rbx, phys_base(%rip)
 
 	/* startup_64 expects boot_params in %rsi. */
-	mov $_pa(pvh_bootparams), %rsi
-	mov $_pa(startup_64), %rax
+	lea rva(pvh_bootparams)(%ebp), %rsi
+	lea rva(startup_64)(%ebp), %rax
 	ANNOTATE_RETPOLINE_SAFE
 	jmp *%rax
 
@@ -137,13 +210,14 @@ SYM_CODE_START_LOCAL(pvh_start_xen)
 
 	ljmp $PVH_CS_SEL, $_pa(startup_32)
 #endif
+
 SYM_CODE_END(pvh_start_xen)
 
 	.section ".init.data","aw"
 	.balign 8
 SYM_DATA_START_LOCAL(gdt)
-	.word gdt_end - gdt_start
-	.long _pa(gdt_start)
+	.word gdt_end - gdt_start - 1
+	.long _pa(gdt_start) /* x86-64 will overwrite if relocated. */
 	.word 0
 SYM_DATA_END(gdt)
 SYM_DATA_START_LOCAL(gdt_start)
@@ -163,5 +237,89 @@ SYM_DATA_START_LOCAL(early_stack)
 	.fill BOOT_STACK_SIZE, 1, 0
 SYM_DATA_END_LABEL(early_stack, SYM_L_LOCAL, early_stack_end)
 
+#ifdef CONFIG_X86_64
+/*
+ * We are not able to switch in one step to the final KERNEL ADDRESS SPACE
+ * because we need identity-mapped pages.
+ */
+#define l4_index(x)     (((x) >> 39) & 511)
+#define pud_index(x)    (((x) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
+
+L4_PAGE_OFFSET  = l4_index(__PAGE_OFFSET_BASE_L4)
+L4_START_KERNEL = l4_index(__START_KERNEL_map)
+L3_START_KERNEL = pud_index(__START_KERNEL_map)
+
+#define SYM_DATA_START_PAGE_ALIGNED(name)			\
+	SYM_START(name, SYM_L_GLOBAL, .balign PAGE_SIZE)
+
+/* Automate the creation of 1 to 1 mapping pmd entries */
+#define PMDS(START, PERM, COUNT)			\
+	i = 0 ;						\
+	.rept (COUNT) ;					\
+	.quad	(START) + (i << PMD_SHIFT) + (PERM) ;	\
+	i = i + 1 ;					\
+	.endr
+
+/*
+ * Xen PVH needs a set of identity mapped and kernel high mapping
+ * page tables.  pvh_start_xen starts running on the identity mapped
+ * page tables, but xen_prepare_pvh calls into the high mapping.
+ * These page tables need to be relocatable and are only used until
+ * startup_64 transitions to init_top_pgt.
+ */
+SYM_DATA_START_PAGE_ALIGNED(pvh_init_top_pgt)
+	.quad   pvh_level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
+	.org    pvh_init_top_pgt + L4_PAGE_OFFSET*8, 0
+	.quad   pvh_level3_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
+	.org    pvh_init_top_pgt + L4_START_KERNEL*8, 0
+	/* (2^48-(2*1024*1024*1024))/(2^39) = 511 */
+	.quad   pvh_level3_kernel_pgt - __START_KERNEL_map + _PAGE_TABLE_NOENC
+SYM_DATA_END(pvh_init_top_pgt)
+
+SYM_DATA_START_PAGE_ALIGNED(pvh_level3_ident_pgt)
+	.quad	pvh_level2_ident_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
+	.fill	511, 8, 0
+SYM_DATA_END(pvh_level3_ident_pgt)
+SYM_DATA_START_PAGE_ALIGNED(pvh_level2_ident_pgt)
+	/*
+	 * Since I easily can, map the first 1G.
+	 * Don't set NX because code runs from these pages.
+	 *
+	 * Note: This sets _PAGE_GLOBAL despite whether
+	 * the CPU supports it or it is enabled.  But,
+	 * the CPU should ignore the bit.
+	 */
+	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
+SYM_DATA_END(pvh_level2_ident_pgt)
+SYM_DATA_START_PAGE_ALIGNED(pvh_level3_kernel_pgt)
+	.fill	L3_START_KERNEL,8,0
+	/* (2^48-(2*1024*1024*1024)-((2^39)*511))/(2^30) = 510 */
+	.quad	pvh_level2_kernel_pgt - __START_KERNEL_map + _KERNPG_TABLE_NOENC
+	.quad	0 /* no fixmap */
+SYM_DATA_END(pvh_level3_kernel_pgt)
+
+SYM_DATA_START_PAGE_ALIGNED(pvh_level2_kernel_pgt)
+	/*
+	 * Kernel high mapping.
+	 *
+	 * The kernel code+data+bss must be located below KERNEL_IMAGE_SIZE in
+	 * virtual address space, which is 1 GiB if RANDOMIZE_BASE is enabled,
+	 * 512 MiB otherwise.
+	 *
+	 * (NOTE: after that starts the module area, see MODULES_VADDR.)
+	 *
+	 * This table is eventually used by the kernel during normal runtime.
+	 * Care must be taken to clear out undesired bits later, like _PAGE_RW
+	 * or _PAGE_GLOBAL in some cases.
+	 */
+	PMDS(0, __PAGE_KERNEL_LARGE_EXEC, KERNEL_IMAGE_SIZE/PMD_SIZE)
+SYM_DATA_END(pvh_level2_kernel_pgt)
+
+	ELFNOTE(Xen, XEN_ELFNOTE_PHYS32_RELOC,
+		     .long CONFIG_PHYSICAL_ALIGN;
+		     .long LOAD_PHYSICAL_ADDR;
+		     .long KERNEL_IMAGE_SIZE - 1)
+#endif
+
 	ELFNOTE(Xen, XEN_ELFNOTE_PHYS32_ENTRY,
-	             _ASM_PTR (pvh_start_xen - __START_KERNEL_map))
+	             .long (pvh_start_xen - __START_KERNEL_map))
diff --git a/include/xen/interface/elfnote.h b/include/xen/interface/elfnote.h
index 38deb1214613..98b3099fe1a6 100644
--- a/include/xen/interface/elfnote.h
+++ b/include/xen/interface/elfnote.h
@@ -185,9 +185,23 @@
  */
 #define XEN_ELFNOTE_PHYS32_ENTRY 18
 
+/*
+ * Physical loading constraints for PVH kernels
+ *
+ * The presence of this note indicates the kernel supports relocating itself.
+ *
+ * The note may include up to three 32bit values to place constraints on the
+ * guest physical loading addresses and alignment for a PVH kernel.  Values
+ * are read in the following order:
+ *  - a required start alignment (default 0x200000)
+ *  - a minimum address for the start of the image (default 0)
+ *  - a maximum address for the last byte of the image (default 0xffffffff)
+ */
+#define XEN_ELFNOTE_PHYS32_RELOC 19
+
 /*
  * The number of the highest elfnote defined.
  */
-#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_ENTRY
+#define XEN_ELFNOTE_MAX XEN_ELFNOTE_PHYS32_RELOC
 
 #endif /* __XEN_PUBLIC_ELFNOTE_H__ */
-- 
2.44.0