[PATCH for-8.1] linux-user: Define real MAP_FIXED_NOREPLACE value

[PATCH for-8.1] linux-user: Define real MAP_FIXED_NOREPLACE value

[Akihiko Odaki]

do_brk() assumes target_mmap() emulates MAP_FIXED_NOREPLACE even when
the host does not support it. However, such emulation is not possible
if MAP_FIXED_NOREPLACE is defined as zero.

Define MAP_FIXED_NOREPLACE with the real value instead of zero if it is
not defined.

Fixes: e69e032d1a ("linux-user: Use MAP_FIXED_NOREPLACE for do_brk()")
Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
---
 include/qemu/osdep.h | 8 ++++++--
 linux-user/elfload.c | 1 -
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/include/qemu/osdep.h b/include/qemu/osdep.h
index cc61b00ba9..1aac17ec2f 100644
--- a/include/qemu/osdep.h
+++ b/include/qemu/osdep.h
@@ -289,8 +289,12 @@ void QEMU_ERROR("code path is reachable")
 #ifndef MAP_ANONYMOUS
 #define MAP_ANONYMOUS MAP_ANON
 #endif
-#ifndef MAP_FIXED_NOREPLACE
-#define MAP_FIXED_NOREPLACE 0
+#if defined(__linux__) && !defined(MAP_FIXED_NOREPLACE)
+#if HOST_ALPHA
+#define MAP_FIXED_NOREPLACE 0x200000
+#else
+#define MAP_FIXED_NOREPLACE 0x100000
+#endif
 #endif
 #ifndef MAP_NORESERVE
 #define MAP_NORESERVE 0
diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 36e4026f05..9d9c79a653 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -2807,7 +2807,6 @@ static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr,
     /* Widen the "image" to the entire reserved address space. */
     pgb_static(image_name, 0, reserved_va, align);
 
-    /* osdep.h defines this as 0 if it's missing */
     flags |= MAP_FIXED_NOREPLACE;
 
     /* Reserve the memory on the host. */
-- 
2.41.0

Re: [PATCH for-8.1] linux-user: Define real MAP_FIXED_NOREPLACE value

[Richard Henderson]

On 8/8/23 04:52, Akihiko Odaki wrote:
> do_brk() assumes target_mmap() emulates MAP_FIXED_NOREPLACE even when
> the host does not support it. However, such emulation is not possible
> if MAP_FIXED_NOREPLACE is defined as zero.
> 
> Define MAP_FIXED_NOREPLACE with the real value instead of zero if it is
> not defined.
> 
> Fixes: e69e032d1a ("linux-user: Use MAP_FIXED_NOREPLACE for do_brk()")
> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
> ---
>   include/qemu/osdep.h | 8 ++++++--
>   linux-user/elfload.c | 1 -
>   2 files changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/include/qemu/osdep.h b/include/qemu/osdep.h
> index cc61b00ba9..1aac17ec2f 100644
> --- a/include/qemu/osdep.h
> +++ b/include/qemu/osdep.h
> @@ -289,8 +289,12 @@ void QEMU_ERROR("code path is reachable")
>   #ifndef MAP_ANONYMOUS
>   #define MAP_ANONYMOUS MAP_ANON
>   #endif
> -#ifndef MAP_FIXED_NOREPLACE
> -#define MAP_FIXED_NOREPLACE 0
> +#if defined(__linux__) && !defined(MAP_FIXED_NOREPLACE)
> +#if HOST_ALPHA

HOST_ALPHA is not a thing.  Also, see

https://lore.kernel.org/qemu-devel/20230808120303.585509-3-pbonzini@redhat.com/

where host support for Alpha is completely removed.


> +#define MAP_FIXED_NOREPLACE 0x200000
> +#else
> +#define MAP_FIXED_NOREPLACE 0x100000
> +#endif

Which supported hosts do not define this value?  Can we simply remove the fallback?


r~

Re: [PATCH for-8.1] linux-user: Define real MAP_FIXED_NOREPLACE value

[Akihiko Odaki]

On 2023/08/08 23:42, Richard Henderson wrote:
> On 8/8/23 04:52, Akihiko Odaki wrote:
>> do_brk() assumes target_mmap() emulates MAP_FIXED_NOREPLACE even when
>> the host does not support it. However, such emulation is not possible
>> if MAP_FIXED_NOREPLACE is defined as zero.
>>
>> Define MAP_FIXED_NOREPLACE with the real value instead of zero if it is
>> not defined.
>>
>> Fixes: e69e032d1a ("linux-user: Use MAP_FIXED_NOREPLACE for do_brk()")
>> Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
>> ---
>>   include/qemu/osdep.h | 8 ++++++--
>>   linux-user/elfload.c | 1 -
>>   2 files changed, 6 insertions(+), 3 deletions(-)
>>
>> diff --git a/include/qemu/osdep.h b/include/qemu/osdep.h
>> index cc61b00ba9..1aac17ec2f 100644
>> --- a/include/qemu/osdep.h
>> +++ b/include/qemu/osdep.h
>> @@ -289,8 +289,12 @@ void QEMU_ERROR("code path is reachable")
>>   #ifndef MAP_ANONYMOUS
>>   #define MAP_ANONYMOUS MAP_ANON
>>   #endif
>> -#ifndef MAP_FIXED_NOREPLACE
>> -#define MAP_FIXED_NOREPLACE 0
>> +#if defined(__linux__) && !defined(MAP_FIXED_NOREPLACE)
>> +#if HOST_ALPHA
> 
> HOST_ALPHA is not a thing.  Also, see
> 
> https://lore.kernel.org/qemu-devel/20230808120303.585509-3-pbonzini@redhat.com/
> 
> where host support for Alpha is completely removed.

I sent v2 with the condition HOST_ALPHA replaced with
MAP_HUGETLB == 0x100000.

While Alpha is no longer supported, and linux-user will not work on 
Alpha, code outside linux-user also refers to osdep.h and the build 
script does not actively block building QEMU on an unsupported host so I 
left the definition for Alpha just in case.

> 
> 
>> +#define MAP_FIXED_NOREPLACE 0x200000
>> +#else
>> +#define MAP_FIXED_NOREPLACE 0x100000
>> +#endif
> 
> Which supported hosts do not define this value?  Can we simply remove 
> the fallback?

glibc didn't have this defined until 2.28. The older releases still 
maintained are 2.26, and 2.27, according to:
https://sourceware.org/glibc/wiki/Release

The page says ALT Linux p9 and Ubuntu 18.04 LTS (Bionic Beaver) has 
glibc 2.27.

> 
> 
> r~

Re: [PATCH for-8.1] linux-user: Define real MAP_FIXED_NOREPLACE value

[Richard Henderson]

On 8/8/23 08:16, Akihiko Odaki wrote:
>> Which supported hosts do not define this value?  Can we simply remove the fallback?
> 
> glibc didn't have this defined until 2.28. The older releases still maintained are 2.26, 
> and 2.27, according to:
> https://sourceware.org/glibc/wiki/Release

Thanks for digging into glibc versions.

> The page says ALT Linux p9 and Ubuntu 18.04 LTS (Bionic Beaver) has glibc 2.27.

QEMU says

# Support for the previous major version will be dropped 2 years
# after the new major version is released or when the vendor itself
# drops support, whichever comes first.

Ubuntu 22.04 is the current major version, so 20.04 is still supported until 2024, but 
18.04 is unsupported.

Similarly, ALT 10.0 was released in 2021, so ALT 9 is now unsupported.

I have just run a patch to remove the fallback through gitlab CI and it has passed:

https://gitlab.com/rth7680/qemu/-/pipelines/959680899/


r~