linux-kernel - Re: [PATCH 2/2] uio:powerpc:mpc85xx: l2-cache-sram uio driver implementation

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 14 Jun 2022 14:59:55 +0000
From:   Christophe Leroy <christophe.leroy@...roup.eu>
To:     Wenhu Wang <wenhu.wang@...mail.com>,
        Greg KH <gregkh@...uxfoundation.org>
CC:     "mpe@...erman.id.au" <mpe@...erman.id.au>,
        "linuxppc-dev@...ts.ozlabs.org" <linuxppc-dev@...ts.ozlabs.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/2] uio:powerpc:mpc85xx: l2-cache-sram uio driver
 implementation



Le 14/06/2022 à 16:40, Wenhu Wang a écrit :
>>>>
>>>> I looked at that patch.
>>>>
>>>> I don't think you can just drop the #ifdef in function
>>>> __access_remote_vm() in mm/memory.c
>>>>
>>>> You have to replace it with something like:
>>>>
>>>>        if (!IS_ENABLED(CONFIG_HAVE_IOREMAP_PROT))
>>>>            break;
>>>>
>>>
>>>
>>> Another thing in that patch:
>>>
>>> By making generic_access_phys() a static inline, it means that everytime
>>> you refer to the address of that function in a vm_operations_struct
>>> struct, the compiler has to provide an outlined instance of the
>>> function. It means you'll likely have several instances of a
>>> generic_access_phys().
>>>
>>> What you could do instead is to add the following at the start of
>>> generic_access_phys() in mm/memory.c :
>>>
>>>          if (!IS_ENABLED(CONFIG_HAVE_IOREMAP_PROT))
>>>                  return 0;
>>>
>>
>> It is really a better chmoce, thanks for the advice.
>> Multiple instances exist as you mentioned, the block returns 0 with no-op
>> instance which makes no difference with the function return value.
>>
>> I will update the patch after a re-confirming.
>>
> 
> I tried as adviced but when not defined, error happens on archectures such
> as arm64. Actually the function generic_access_phys calls a lot of functions
> that become undefined if we compile it with CONFIG_HAVE_IOREMAP_PROT disabled.
> The archectures that support CONFIG_HAVE_IOREMAP_PROT are mips, x86, sh, arc,
> s390, loongarch and powerpc.
> 
> So we may just define the function with static inline and add IS_ENABLED
> condition branch in function __access_remote_vm in mm/memory.c. The executing
> path breaks if CONFIG_HAVE_IOREMAP_PROT is disabled, and never goes into the
> static no-op function.
> 
> In short, the static inline no-op function would never be executed, the only
> difference is that there would be a lot of function code in compiled target.
> 

In that case all you have to do is:

diff --git a/mm/memory.c b/mm/memory.c
index 7a089145cad4..39b369fc77f6 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -5413,6 +5413,13 @@ int generic_access_phys(struct vm_area_struct 
*vma, unsigned long addr,
  	return ret;
  }
  EXPORT_SYMBOL_GPL(generic_access_phys);
+#else
+int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
+			void *buf, int len, int write)
+{
+	return 0;
+}
+EXPORT_SYMBOL_GPL(generic_access_phys);
  #endif

  /*


Christophe