diff mbox series

[for-4.13] xen/arm: fix duplicate memory node in DT

Message ID alpine.DEB.2.21.1910041601170.30844@sstabellini-ThinkPad-T480s (mailing list archive)
State New, archived
Headers show
Series [for-4.13] xen/arm: fix duplicate memory node in DT | expand

Commit Message

Stefano Stabellini Oct. 4, 2019, 11:09 p.m. UTC
When reserved-memory regions are present in the host device tree, dom0
is started with multiple memory nodes. Each memory node should have a
unique name, but today they are all called "memory" leading to Linux
printing the following warning at boot:

  OF: Duplicate name in base, renamed to "memory#1"

This patch fixes the problem by appending a "@<unit-address>" to the
name, as per the Device Tree specification, where <unit-address> matches
the base of address of the first region.

Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>

---

Comments

Jürgen Groß Oct. 7, 2019, 6:51 a.m. UTC | #1
On 05.10.19 01:09, Stefano Stabellini wrote:
> When reserved-memory regions are present in the host device tree, dom0
> is started with multiple memory nodes. Each memory node should have a
> unique name, but today they are all called "memory" leading to Linux
> printing the following warning at boot:
> 
>    OF: Duplicate name in base, renamed to "memory#1"
> 
> This patch fixes the problem by appending a "@<unit-address>" to the
> name, as per the Device Tree specification, where <unit-address> matches
> the base of address of the first region.
> 
> Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
> 
> ---
> 
> diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
> index 921b054520..a4c07db383 100644
> --- a/xen/arch/arm/domain_build.c
> +++ b/xen/arch/arm/domain_build.c
> @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain *d,
>       int res, i;
>       int reg_size = addrcells + sizecells;
>       int nr_cells = reg_size * mem->nr_banks;
> +    /* Placeholder for memory@ + a 32-bit number + \0 */
> +    char buf[18];

You are using PRIx64 for printing the number, so I guess you should
enlarge buf by 8 bytes and adjust the comment (s/32/64/).

>       __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
>       __be32 *cells;
>   
>       BUG_ON(nr_cells >= ARRAY_SIZE(reg));
> +    /* Nothing to do */
> +    if ( mem->nr_banks == 0 )
> +        return 0;
>   
>       dt_dprintk("Create memory node (reg size %d, nr cells %d)\n",
>                  reg_size, nr_cells);
>   
>       /* ePAPR 3.4 */
> -    res = fdt_begin_node(fdt, "memory");
> +    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[0].start);
> +    res = fdt_begin_node(fdt, buf);
>       if ( res )
>           return res;


Juergen
Julien Grall Oct. 7, 2019, 10:01 a.m. UTC | #2
Hi,

On 05/10/2019 00:09, Stefano Stabellini wrote:
> When reserved-memory regions are present in the host device tree, dom0
> is started with multiple memory nodes. Each memory node should have a
> unique name, but today they are all called "memory" leading to Linux
> printing the following warning at boot:
> 
>    OF: Duplicate name in base, renamed to "memory#1"
> 
> This patch fixes the problem by appending a "@<unit-address>" to the
> name, as per the Device Tree specification, where <unit-address> matches
> the base of address of the first region.
> 
> Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
> 
> ---
> 
> diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
> index 921b054520..a4c07db383 100644
> --- a/xen/arch/arm/domain_build.c
> +++ b/xen/arch/arm/domain_build.c
> @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain *d,
>       int res, i;
>       int reg_size = addrcells + sizecells;
>       int nr_cells = reg_size * mem->nr_banks;
> +    /* Placeholder for memory@ + a 32-bit number + \0 */
> +    char buf[18];
>       __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
>       __be32 *cells;
>   
>       BUG_ON(nr_cells >= ARRAY_SIZE(reg));
> +    /* Nothing to do */

This a departure from the current solution where a node will be created with no 
"reg" property. I think this change of behavior should at least be described in 
the commit message if not implemented in a separate patch. But...

> +    if ( mem->nr_banks == 0 )
> +        return 0;

... I don't think we want to ignore it. The caller most likely messed up the 
banks and we should instead report an error.

>   
>       dt_dprintk("Create memory node (reg size %d, nr cells %d)\n",
>                  reg_size, nr_cells);
>   
>       /* ePAPR 3.4 */
> -    res = fdt_begin_node(fdt, "memory");
> +    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[0].start);
> +    res = fdt_begin_node(fdt, buf);
>       if ( res )
>           return res;
>   
> 

Cheers,
Stefano Stabellini Oct. 7, 2019, 9:30 p.m. UTC | #3
On Mon, 7 Oct 2019, Julien Grall wrote:
> On 05/10/2019 00:09, Stefano Stabellini wrote:
> > When reserved-memory regions are present in the host device tree, dom0
> > is started with multiple memory nodes. Each memory node should have a
> > unique name, but today they are all called "memory" leading to Linux
> > printing the following warning at boot:
> > 
> >    OF: Duplicate name in base, renamed to "memory#1"
> > 
> > This patch fixes the problem by appending a "@<unit-address>" to the
> > name, as per the Device Tree specification, where <unit-address> matches
> > the base of address of the first region.
> > 
> > Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> > Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
> > 
> > ---
> > 
> > diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
> > index 921b054520..a4c07db383 100644
> > --- a/xen/arch/arm/domain_build.c
> > +++ b/xen/arch/arm/domain_build.c
> > @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain
> > *d,
> >       int res, i;
> >       int reg_size = addrcells + sizecells;
> >       int nr_cells = reg_size * mem->nr_banks;
> > +    /* Placeholder for memory@ + a 32-bit number + \0 */
> > +    char buf[18];
> >       __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
> >       __be32 *cells;
> >         BUG_ON(nr_cells >= ARRAY_SIZE(reg));
> > +    /* Nothing to do */
> 
> This a departure from the current solution where a node will be created with
> no "reg" property. I think this change of behavior should at least be
> described in the commit message if not implemented in a separate patch. But...
> 
> > +    if ( mem->nr_banks == 0 )
> > +        return 0;
> 
> ... I don't think we want to ignore it. The caller most likely messed up the
> banks and we should instead report an error.

I admit it wasn't my intention to change the current behavior. As I was
looking through the code I noticed that we call make_memory_node for
both normal memory and reserved_memory. Of course, reserved_memory could
have no banks. So I thought it would be good to check whether there are
any banks before continuing because now we are going to access
mem->bank[0].start, which would be a mistake if there are no banks.

In regards to your comment about returning error, we could return ENOENT,
however we would also have to handle ENOENT especially at the caller
side (handle_node). Or we would have to add a check if ( mem->nr_banks >
0) to avoid calling make_memory_node when nr_banks is zero.


> >         dt_dprintk("Create memory node (reg size %d, nr cells %d)\n",
> >                  reg_size, nr_cells);
> >         /* ePAPR 3.4 */
> > -    res = fdt_begin_node(fdt, "memory");
> > +    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[0].start);
> > +    res = fdt_begin_node(fdt, buf);
> >       if ( res )
> >           return res;
> >   
> 
> Cheers,
> 
> -- 
> Julien Grall
>
Stefano Stabellini Oct. 7, 2019, 9:30 p.m. UTC | #4
On Mon, 7 Oct 2019, Jürgen Groß wrote:
> On 05.10.19 01:09, Stefano Stabellini wrote:
> > When reserved-memory regions are present in the host device tree, dom0
> > is started with multiple memory nodes. Each memory node should have a
> > unique name, but today they are all called "memory" leading to Linux
> > printing the following warning at boot:
> > 
> >    OF: Duplicate name in base, renamed to "memory#1"
> > 
> > This patch fixes the problem by appending a "@<unit-address>" to the
> > name, as per the Device Tree specification, where <unit-address> matches
> > the base of address of the first region.
> > 
> > Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> > Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
> > 
> > ---
> > 
> > diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
> > index 921b054520..a4c07db383 100644
> > --- a/xen/arch/arm/domain_build.c
> > +++ b/xen/arch/arm/domain_build.c
> > @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain
> > *d,
> >       int res, i;
> >       int reg_size = addrcells + sizecells;
> >       int nr_cells = reg_size * mem->nr_banks;
> > +    /* Placeholder for memory@ + a 32-bit number + \0 */
> > +    char buf[18];
> 
> You are using PRIx64 for printing the number, so I guess you should
> enlarge buf by 8 bytes and adjust the comment (s/32/64/).

Well spotted! In fact, there is a similar error in make_cpus_node. I'll
fix that one too in a separate patch.


> >       __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
> >       __be32 *cells;
> >         BUG_ON(nr_cells >= ARRAY_SIZE(reg));
> > +    /* Nothing to do */
> > +    if ( mem->nr_banks == 0 )
> > +        return 0;
> >         dt_dprintk("Create memory node (reg size %d, nr cells %d)\n",
> >                  reg_size, nr_cells);
> >         /* ePAPR 3.4 */
> > -    res = fdt_begin_node(fdt, "memory");
> > +    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[0].start);
> > +    res = fdt_begin_node(fdt, buf);
> >       if ( res )
> >           return res;
> 
> 
> Juergen
>
Julien Grall Oct. 7, 2019, 9:47 p.m. UTC | #5
Hi,

On 07/10/2019 22:30, Stefano Stabellini wrote:
> On Mon, 7 Oct 2019, Julien Grall wrote:
>> On 05/10/2019 00:09, Stefano Stabellini wrote:
>>> When reserved-memory regions are present in the host device tree, dom0
>>> is started with multiple memory nodes. Each memory node should have a
>>> unique name, but today they are all called "memory" leading to Linux
>>> printing the following warning at boot:
>>>
>>>     OF: Duplicate name in base, renamed to "memory#1"
>>>
>>> This patch fixes the problem by appending a "@<unit-address>" to the
>>> name, as per the Device Tree specification, where <unit-address> matches
>>> the base of address of the first region.
>>>
>>> Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
>>> Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
>>>
>>> ---
>>>
>>> diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
>>> index 921b054520..a4c07db383 100644
>>> --- a/xen/arch/arm/domain_build.c
>>> +++ b/xen/arch/arm/domain_build.c
>>> @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain
>>> *d,
>>>        int res, i;
>>>        int reg_size = addrcells + sizecells;
>>>        int nr_cells = reg_size * mem->nr_banks;
>>> +    /* Placeholder for memory@ + a 32-bit number + \0 */
>>> +    char buf[18];
>>>        __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
>>>        __be32 *cells;
>>>          BUG_ON(nr_cells >= ARRAY_SIZE(reg));
>>> +    /* Nothing to do */
>>
>> This a departure from the current solution where a node will be created with
>> no "reg" property. I think this change of behavior should at least be
>> described in the commit message if not implemented in a separate patch. But...
>>
>>> +    if ( mem->nr_banks == 0 )
>>> +        return 0;
>>
>> ... I don't think we want to ignore it. The caller most likely messed up the
>> banks and we should instead report an error.
> 
> I admit it wasn't my intention to change the current behavior. As I was
> looking through the code I noticed that we call make_memory_node for
> both normal memory and reserved_memory. Of course, reserved_memory could
> have no banks. So I thought it would be good to check whether there are
> any banks before continuing because now we are going to access
> mem->bank[0].start, which would be a mistake if there are no banks.

Ok, so this not theoritical bug as I first thought but a real bug on 
platform where DT does not have reserved-regions node.

In this case, this should be in a separate patch as this is now 2 
different bugs solved in one patch.

> 
> In regards to your comment about returning error, we could return ENOENT,
> however we would also have to handle ENOENT especially at the caller
> side (handle_node). Or we would have to add a check if ( mem->nr_banks >
> 0) to avoid calling make_memory_node when nr_banks is zero.

I would much prefer if we check mem->nr_banks > 0 for reserved-regions 
before hand.

Both will need a "Fixes:" to keep track of the original patch.

Cheers,
Stefano Stabellini Oct. 7, 2019, 10:15 p.m. UTC | #6
On Mon, 7 Oct 2019, Julien Grall wrote:
> Hi,
> 
> On 07/10/2019 22:30, Stefano Stabellini wrote:
> > On Mon, 7 Oct 2019, Julien Grall wrote:
> >> On 05/10/2019 00:09, Stefano Stabellini wrote:
> >>> When reserved-memory regions are present in the host device tree, dom0
> >>> is started with multiple memory nodes. Each memory node should have a
> >>> unique name, but today they are all called "memory" leading to Linux
> >>> printing the following warning at boot:
> >>>
> >>>     OF: Duplicate name in base, renamed to "memory#1"
> >>>
> >>> This patch fixes the problem by appending a "@<unit-address>" to the
> >>> name, as per the Device Tree specification, where <unit-address> matches
> >>> the base of address of the first region.
> >>>
> >>> Reported-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> >>> Signed-off-by: Stefano Stabellini <stefano.stabellini@xilinx.com>
> >>>
> >>> ---
> >>>
> >>> diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
> >>> index 921b054520..a4c07db383 100644
> >>> --- a/xen/arch/arm/domain_build.c
> >>> +++ b/xen/arch/arm/domain_build.c
> >>> @@ -646,16 +646,22 @@ static int __init make_memory_node(const struct domain
> >>> *d,
> >>>        int res, i;
> >>>        int reg_size = addrcells + sizecells;
> >>>        int nr_cells = reg_size * mem->nr_banks;
> >>> +    /* Placeholder for memory@ + a 32-bit number + \0 */
> >>> +    char buf[18];
> >>>        __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
> >>>        __be32 *cells;
> >>>          BUG_ON(nr_cells >= ARRAY_SIZE(reg));
> >>> +    /* Nothing to do */
> >>
> >> This a departure from the current solution where a node will be created with
> >> no "reg" property. I think this change of behavior should at least be
> >> described in the commit message if not implemented in a separate patch. But...
> >>
> >>> +    if ( mem->nr_banks == 0 )
> >>> +        return 0;
> >>
> >> ... I don't think we want to ignore it. The caller most likely messed up the
> >> banks and we should instead report an error.
> > 
> > I admit it wasn't my intention to change the current behavior. As I was
> > looking through the code I noticed that we call make_memory_node for
> > both normal memory and reserved_memory. Of course, reserved_memory could
> > have no banks. So I thought it would be good to check whether there are
> > any banks before continuing because now we are going to access
> > mem->bank[0].start, which would be a mistake if there are no banks.
> 
> Ok, so this not theoritical bug as I first thought but a real bug on 
> platform where DT does not have reserved-regions node.
> 
> In this case, this should be in a separate patch as this is now 2 
> different bugs solved in one patch.

OK


> > In regards to your comment about returning error, we could return ENOENT,
> > however we would also have to handle ENOENT especially at the caller
> > side (handle_node). Or we would have to add a check if ( mem->nr_banks >
> > 0) to avoid calling make_memory_node when nr_banks is zero.
> 
> I would much prefer if we check mem->nr_banks > 0 for reserved-regions 
> before hand.

All right


> Both will need a "Fixes:" to keep track of the original patch.
diff mbox series

Patch

diff --git a/xen/arch/arm/domain_build.c b/xen/arch/arm/domain_build.c
index 921b054520..a4c07db383 100644
--- a/xen/arch/arm/domain_build.c
+++ b/xen/arch/arm/domain_build.c
@@ -646,16 +646,22 @@  static int __init make_memory_node(const struct domain *d,
     int res, i;
     int reg_size = addrcells + sizecells;
     int nr_cells = reg_size * mem->nr_banks;
+    /* Placeholder for memory@ + a 32-bit number + \0 */
+    char buf[18];
     __be32 reg[NR_MEM_BANKS * 4 /* Worst case addrcells + sizecells */];
     __be32 *cells;
 
     BUG_ON(nr_cells >= ARRAY_SIZE(reg));
+    /* Nothing to do */
+    if ( mem->nr_banks == 0 )
+        return 0;
 
     dt_dprintk("Create memory node (reg size %d, nr cells %d)\n",
                reg_size, nr_cells);
 
     /* ePAPR 3.4 */
-    res = fdt_begin_node(fdt, "memory");
+    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[0].start);
+    res = fdt_begin_node(fdt, buf);
     if ( res )
         return res;