The IPI buffer location is read from the firmware that we load to the
System Companion Processor, and it's not granted that both the SRAM
(L2TCM) size that is defined in the devicetree node is large enough
for that, and while this is especially true for multi-core SCP, it's
still useful to check on single-core variants as well.
Failing to perform this check may make this driver perform R/W
oeprations out of the L2TCM boundary, resulting (at best) in a
kernel panic.
To fix that, check that the IPI buffer fits, otherwise return a
failure and refuse to boot the relevant SCP core (or the SCP at
all, if this is single core).
Fixes: 3efa0ea743b7 ("remoteproc/mediatek: read IPI buffer offset from FW")
Signed-off-by: AngeloGioacchino Del Regno <[email protected]>
---
drivers/remoteproc/mtk_scp.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
index a35409eda0cf..67518291a8ad 100644
--- a/drivers/remoteproc/mtk_scp.c
+++ b/drivers/remoteproc/mtk_scp.c
@@ -132,7 +132,7 @@ static int scp_elf_read_ipi_buf_addr(struct mtk_scp *scp,
static int scp_ipi_init(struct mtk_scp *scp, const struct firmware *fw)
{
int ret;
- size_t offset;
+ size_t buf_sz, offset;
/* read the ipi buf addr from FW itself first */
ret = scp_elf_read_ipi_buf_addr(scp, fw, &offset);
@@ -144,6 +144,14 @@ static int scp_ipi_init(struct mtk_scp *scp, const struct firmware *fw)
}
dev_info(scp->dev, "IPI buf addr %#010zx\n", offset);
+ /* Make sure IPI buffer fits in the L2TCM range assigned to this core */
+ buf_sz = sizeof(*scp->recv_buf) + sizeof(*scp->send_buf);
+
+ if (scp->sram_size < buf_sz + offset) {
+ dev_err(scp->dev, "IPI buffer does not fit in SRAM.\n");
+ return -EOVERFLOW;
+ }
+
scp->recv_buf = (struct mtk_share_obj __iomem *)
(scp->sram_base + offset);
scp->send_buf = (struct mtk_share_obj __iomem *)
--
2.44.0
Good day,
On Thu, Mar 21, 2024 at 09:46:13AM +0100, AngeloGioacchino Del Regno wrote:
> The IPI buffer location is read from the firmware that we load to the
> System Companion Processor, and it's not granted that both the SRAM
> (L2TCM) size that is defined in the devicetree node is large enough
> for that, and while this is especially true for multi-core SCP, it's
> still useful to check on single-core variants as well.
>
> Failing to perform this check may make this driver perform R/W
> oeprations out of the L2TCM boundary, resulting (at best) in a
s/oeprations/operations
I will fix that when I apply the patch.
> kernel panic.
>
> To fix that, check that the IPI buffer fits, otherwise return a
> failure and refuse to boot the relevant SCP core (or the SCP at
> all, if this is single core).
>
> Fixes: 3efa0ea743b7 ("remoteproc/mediatek: read IPI buffer offset from FW")
> Signed-off-by: AngeloGioacchino Del Regno <[email protected]>
> ---
> drivers/remoteproc/mtk_scp.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/remoteproc/mtk_scp.c b/drivers/remoteproc/mtk_scp.c
> index a35409eda0cf..67518291a8ad 100644
> --- a/drivers/remoteproc/mtk_scp.c
> +++ b/drivers/remoteproc/mtk_scp.c
> @@ -132,7 +132,7 @@ static int scp_elf_read_ipi_buf_addr(struct mtk_scp *scp,
> static int scp_ipi_init(struct mtk_scp *scp, const struct firmware *fw)
> {
> int ret;
> - size_t offset;
> + size_t buf_sz, offset;
>
> /* read the ipi buf addr from FW itself first */
> ret = scp_elf_read_ipi_buf_addr(scp, fw, &offset);
> @@ -144,6 +144,14 @@ static int scp_ipi_init(struct mtk_scp *scp, const struct firmware *fw)
> }
> dev_info(scp->dev, "IPI buf addr %#010zx\n", offset);
>
> + /* Make sure IPI buffer fits in the L2TCM range assigned to this core */
> + buf_sz = sizeof(*scp->recv_buf) + sizeof(*scp->send_buf);
> +
> + if (scp->sram_size < buf_sz + offset) {
> + dev_err(scp->dev, "IPI buffer does not fit in SRAM.\n");
> + return -EOVERFLOW;
> + }
> +
> scp->recv_buf = (struct mtk_share_obj __iomem *)
> (scp->sram_base + offset);
> scp->send_buf = (struct mtk_share_obj __iomem *)
> --
> 2.44.0
>
On Thu, Mar 21, 2024 at 09:46:13AM +0100, AngeloGioacchino Del Regno wrote:
> The IPI buffer location is read from the firmware that we load to the
> System Companion Processor, and it's not granted that both the SRAM
> (L2TCM) size that is defined in the devicetree node is large enough
> for that, and while this is especially true for multi-core SCP, it's
> still useful to check on single-core variants as well.
>
> Failing to perform this check may make this driver perform R/W
> oeprations out of the L2TCM boundary, resulting (at best) in a
> kernel panic.
>
> To fix that, check that the IPI buffer fits, otherwise return a
> failure and refuse to boot the relevant SCP core (or the SCP at
> all, if this is single core).
>
> Fixes: 3efa0ea743b7 ("remoteproc/mediatek: read IPI buffer offset from FW")
> Signed-off-by: AngeloGioacchino Del Regno <[email protected]>
Reviewed-by: Tzung-Bi Shih <[email protected]>
Il 21/03/24 16:25, Mathieu Poirier ha scritto:
> Good day,
>
> On Thu, Mar 21, 2024 at 09:46:13AM +0100, AngeloGioacchino Del Regno wrote:
>> The IPI buffer location is read from the firmware that we load to the
>> System Companion Processor, and it's not granted that both the SRAM
>> (L2TCM) size that is defined in the devicetree node is large enough
>> for that, and while this is especially true for multi-core SCP, it's
>> still useful to check on single-core variants as well.
>>
>> Failing to perform this check may make this driver perform R/W
>> oeprations out of the L2TCM boundary, resulting (at best) in a
>
> s/oeprations/operations
>
> I will fix that when I apply the patch.
>
Thanks for that.
Cheers,
Angelo