Date:   Wed, 20 Apr 2022 16:07:36 +0200
From:   "Fabio M. De Francesco" <fmdefrancesco@...il.com>
To:     Julia Lawall <julia.lawall@...ia.fr>
Cc:     Julia Lawall <julia.lawall@...ia.fr>, ira.weiny@...el.com,
        Alaa Mohamed <eng.alaamohamedsoliman.am@...il.com>,
        outreachy@...ts.linux.dev, boris.ostrovsky@...cle.com,
        jgross@...e.com, sstabellini@...nel.org,
        xen-devel@...ts.xenproject.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2] xen:  Convert kmap() to kmap_local_page()

On mercoledì 20 aprile 2022 15:57:14 CEST Julia Lawall wrote:
> 
> On Wed, 20 Apr 2022, Fabio M. De Francesco wrote:
> 
> > On mercoledì 20 aprile 2022 15:40:10 CEST Julia Lawall wrote:
> > >
> > > On Wed, 20 Apr 2022, Fabio M. De Francesco wrote:
> > >
> > > > On mercoledì 20 aprile 2022 08:03:05 CEST Julia Lawall wrote:
> > > > >
> > > > > On Wed, 20 Apr 2022, Alaa Mohamed wrote:
> > > > >
> > > > > > kmap() is being deprecated and these usages are all local to 
the
> > thread
> > > > > > so there is no reason kmap_local_page() can't be used.
> > > > > >
> > > > > > Replace kmap() calls with kmap_local_page().
> > > > >
> > > > > OK, so from a Coccinelle point of view, could we do
> > > > >
> > > > > @@
> > > > > expression e1,e2,x,f;
> > > > > @@
> > > > >
> > > > > e1 =
> > > > > - kmap
> > > > > + kmap_local_page
> > > > >     (e2)
> > > > > ... when != x = e1 // not stored in any location and not passed 
to
> > > > another function
> > > > >     when != f(...,e1,...)
> > > > >     when != x = e2
> > > > >     when != f(...,e2,...)
> > > > > -kunmap(e2)
> > > > > +kunmap_local(e1)
> > > > >
> > > > > julia
> > > > >
> > > >
> > > > I've never spent sufficient time to understand properly the syntax 
and
> > > > semantics of expressions of Coccinelle. However, thanks Julia, this
> > code
> > > > looks good and can be very helpful.
> > > >
> > > > Only a minor objection... it doesn't tell when 'e2' has been 
allocated
> > > > within the same function where the kmap() call is.
> > > >
> > > > In the particular case that I cite above, I'd prefer to remove the
> > > > allocation of the page (say with alloc_page()) and convert kmap() /
> > kunmap()
> > > > to use kmalloc() / kfree().
> > > >
> > > > Fox example, this is done in the following patch:
> > > >
> > > > commit 633b0616cfe0 ("x86/sgx: Remove unnecessary kmap() from
> > > > sgx_ioc_enclave_init()") from Ira Weiny.
> > > >
> > > > Can Coccinelle catch also those special cases where a page that is
> > passed
> > > > to kmap() is allocated within that same function (vs. being passed 
as
> > > > argument to this function) and, if so, propose a replacement with
> > > > kmalloc()?
> > >
> > > It looks complex in this case, because the allocation is in another
> > > function, and it is passed to another function.
> >
> > This is not the special case I was talking about. In this case your 
code
> > for Coccinelle tells the right proposal and it is exactly what Alaa did 
in
> > her patch (which is good!).
> >
> > I'm talking about other special cases like the one I pointed to with 
the
> > link I provided. I'm sorry if my bad English made you think that Alaa's
> > patch was one of those cases where the page is allocated within the 
same
> > function where kmap() is.
> >
> > I hope that now I've been clearer :)
> 
> Ah, sorry for the misunderstanding.  If you have an example, I can take a
> look and propose something for this special case.
> 
> julia

Yes, I have the example that you are asking for. It's that commit 
633b0616cfe0 from Ira Weiny.

Let me copy and paste it here for your convenience...

diff --git a/arch/x86/kernel/cpu/sgx/ioctl.c b/arch/x86/kernel/cpu/sgx/
ioctl.c
index 90a5caf76939..2e10367ea66c 100644
--- a/arch/x86/kernel/cpu/sgx/ioctl.c
+++ b/arch/x86/kernel/cpu/sgx/ioctl.c
@@ -604,7 +604,6 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, 
void __user *arg)
 {
        struct sgx_sigstruct *sigstruct;
        struct sgx_enclave_init init_arg;
-       struct page *initp_page;
        void *token;
        int ret;
 
@@ -615,11 +614,15 @@ static long sgx_ioc_enclave_init(struct sgx_encl 
*encl, void __user *arg)
        if (copy_from_user(&init_arg, arg, sizeof(init_arg)))
                return -EFAULT;
 
-       initp_page = alloc_page(GFP_KERNEL);
-       if (!initp_page)
+       /*
+        * 'sigstruct' must be on a page boundary and 'token' on a 512 byte
+        * boundary.  kmalloc() will give this alignment when allocating
+        * PAGE_SIZE bytes.
+        */
+       sigstruct = kmalloc(PAGE_SIZE, GFP_KERNEL);
+       if (!sigstruct)
                return -ENOMEM;
 
-       sigstruct = kmap(initp_page);
        token = (void *)((unsigned long)sigstruct + PAGE_SIZE / 2);
        memset(token, 0, SGX_LAUNCH_TOKEN_SIZE);
 
@@ -645,8 +648,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, 
void __user *arg)
        ret = sgx_encl_init(encl, sigstruct, token);
 
 out:
-       kunmap(initp_page);
-       __free_page(initp_page);
+       kfree(sigstruct);
        return ret;
 }

I think that Coccinelle might understand that "initp_page" is allocated in 
the same function where later it is kmap()'ed. But I'm not able to write a 
Coccinelle check to find out these kinds of special cases. In these cases 
the correct solution is not to use kmap_local_page(). Instead delete the 
alloc_page() and use kmalloc().

Thanks,

Fabio

Powered by blists - more mailing lists