[PATCH] imx8m, imx9: Fix DRAM size calculation in SPL/dram_init_banksize()
If dram_init_banksize() is called from SPL, the rom_pointer, at that point, is not correctly initialized. This causes wrong calculation of DRAM start and size in dram_init_banksize(). The issue became apparent only in Falcon Mode. Added an extra condition to prevent using rom_pointer in SPL.
Signed-off-by: Elena Popa elena.popa@nxp.com --- arch/arm/mach-imx/imx8m/soc.c | 2 +- arch/arm/mach-imx/imx9/soc.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm/mach-imx/imx8m/soc.c b/arch/arm/mach-imx/imx8m/soc.c index d5254886be..dbb2154b08 100644 --- a/arch/arm/mach-imx/imx8m/soc.c +++ b/arch/arm/mach-imx/imx8m/soc.c @@ -273,7 +273,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM; - if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { + if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && !IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) { phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];
diff --git a/arch/arm/mach-imx/imx9/soc.c b/arch/arm/mach-imx/imx9/soc.c index f43b73a6c2..ca91f20fe7 100644 --- a/arch/arm/mach-imx/imx9/soc.c +++ b/arch/arm/mach-imx/imx9/soc.c @@ -390,7 +390,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM; - if (rom_pointer[1]) { + if (!IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) { phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];
On 8/7/23 12:56, Elena Popa wrote:
Subject tags should be (per git log arch/arm/mach-imx/imx8m/soc.c):
arm: imx: imx8m: imx9: Fix DRAM .....
If dram_init_banksize() is called from SPL, the rom_pointer, at that point, is not correctly initialized. This causes wrong calculation of DRAM start and size in dram_init_banksize(). The issue became apparent only in Falcon Mode. Added an extra condition to prevent using rom_pointer in SPL.
+CC Stefano, Fabio, Peng.
Signed-off-by: Elena Popa elena.popa@nxp.com
arch/arm/mach-imx/imx8m/soc.c | 2 +- arch/arm/mach-imx/imx9/soc.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm/mach-imx/imx8m/soc.c b/arch/arm/mach-imx/imx8m/soc.c index d5254886be..dbb2154b08 100644 --- a/arch/arm/mach-imx/imx8m/soc.c +++ b/arch/arm/mach-imx/imx8m/soc.c @@ -273,7 +273,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM;
- if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) {
- if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && !IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) {
Use ./scripts/checkpatch.pl , it should indicate line too long here.
There seem to be other places which likely also need similar fix:
$ git grep CONFIG_ARMV8_PSCI.*rom_pointer arch/arm/mach-imx/ arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[0] &&
phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];diff --git a/arch/arm/mach-imx/imx9/soc.c b/arch/arm/mach-imx/imx9/soc.c index f43b73a6c2..ca91f20fe7 100644 --- a/arch/arm/mach-imx/imx9/soc.c +++ b/arch/arm/mach-imx/imx9/soc.c @@ -390,7 +390,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM;
- if (rom_pointer[1]) {
- if (!IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) { phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];
[...]
Subject tags should be (per git log arch/arm/mach-imx/imx8m/soc.c):
arm: imx: imx8m: imx9: Fix DRAM .....
Will do!
If dram_init_banksize() is called from SPL, the rom_pointer, at that point, is not correctly initialized. This causes wrong calculation of DRAM start and size in dram_init_banksize(). The issue became apparent only in Falcon Mode. Added an extra condition to prevent using rom_pointer in SPL.
+CC Stefano, Fabio, Peng.
Signed-off-by: Elena Popa elena.popa@nxp.com
arch/arm/mach-imx/imx8m/soc.c | 2 +- arch/arm/mach-imx/imx9/soc.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm/mach-imx/imx8m/soc.c b/arch/arm/mach-imx/imx8m/soc.c index d5254886be..dbb2154b08
100644
--- a/arch/arm/mach-imx/imx8m/soc.c +++ b/arch/arm/mach-imx/imx8m/soc.c @@ -273,7 +273,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM;
if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) {
if (!IS_ENABLED(CONFIG_ARMV8_PSCI) &&- !IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) {
Use ./scripts/checkpatch.pl , it should indicate line too long here.
I did that, but got no warning. For text, the script limit is 75 and code limit is 100. Should I break the line?
There seem to be other places which likely also need similar fix:
$ git grep CONFIG_ARMV8_PSCI.*rom_pointer arch/arm/mach-imx/ arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[0] &&
The other 2 places with similar conditions are in get_effective_memsize() and dram_init(). At least for i.MX8M and i.MX93 are not called from within SPL. Should we add nonetheless the extra condition in those places as well? (for some of the other architectures, dram_init() is called in SPL).
phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];diff --git a/arch/arm/mach-imx/imx9/soc.c b/arch/arm/mach-imx/imx9/soc.c index f43b73a6c2..ca91f20fe7 100644 --- a/arch/arm/mach-imx/imx9/soc.c +++ b/arch/arm/mach-imx/imx9/soc.c @@ -390,7 +390,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM;
if (rom_pointer[1]) {
if (!IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) { phys_addr_t optee_start = (phys_addr_t)rom_pointer[0]; phys_size_t optee_size = (size_t)rom_pointer[1];[...]
On 8/7/23 18:43, Elena Popa wrote:
Subject tags should be (per git log arch/arm/mach-imx/imx8m/soc.c):
arm: imx: imx8m: imx9: Fix DRAM .....
Will do!
If dram_init_banksize() is called from SPL, the rom_pointer, at that point, is not correctly initialized. This causes wrong calculation of DRAM start and size in dram_init_banksize(). The issue became apparent only in Falcon Mode. Added an extra condition to prevent using rom_pointer in SPL.
+CC Stefano, Fabio, Peng.
Signed-off-by: Elena Popa elena.popa@nxp.com
arch/arm/mach-imx/imx8m/soc.c | 2 +- arch/arm/mach-imx/imx9/soc.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm/mach-imx/imx8m/soc.c b/arch/arm/mach-imx/imx8m/soc.c index d5254886be..dbb2154b08
100644
--- a/arch/arm/mach-imx/imx8m/soc.c +++ b/arch/arm/mach-imx/imx8m/soc.c @@ -273,7 +273,7 @@ int dram_init_banksize(void) }
gd->bd->bi_dram[bank].start = PHYS_SDRAM;
if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) {
if (!IS_ENABLED(CONFIG_ARMV8_PSCI) &&- !IS_ENABLED(CONFIG_SPL_BUILD) && rom_pointer[1]) {
Use ./scripts/checkpatch.pl , it should indicate line too long here.
I did that, but got no warning. For text, the script limit is 75 and code limit is 100.
I wasn't aware of that, thanks for pointing it out.
Should I break the line?
No
There seem to be other places which likely also need similar fix:
$ git grep CONFIG_ARMV8_PSCI.*rom_pointer arch/arm/mach-imx/ arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[1]) { arch/arm/mach-imx/imx8m/soc.c: if (!IS_ENABLED(CONFIG_ARMV8_PSCI) && rom_pointer[0] &&
The other 2 places with similar conditions are in get_effective_memsize() and dram_init(). At least for i.MX8M and i.MX93 are not called from within SPL. Should we add nonetheless the extra condition in those places as well? (for some of the other architectures, dram_init() is called in SPL).
Better be consistent and fix them all, yes.
participants (2)
-
Elena Popa -
Marek Vasut