arm: cache: cp15: don't map reserved region with no-map property
Subsystems (4)
v3 cover arm: cache: cp15: don't map reserved region with no-map property
Hi,
It it the v3 serie of [1].
This v3 serie is rebased on top of v2021.07-rc1 with integrated previous series:
- [2] for stm32mp parts and added dram_bank_mmu_setup
- [3] for LMB impacts
On STM32MP15x platform we can use OP-TEE, loaded in DDR in a region protected by a firewall. This region is reserved in device with "no-map" property.
Sometime the platform boot failed in U-Boot on a Cortex A7 access to this region (depending of the binary and the issue can change with compiler version or with code alignment), then the firewall raise an error, for example:
E/TC:0 tzc_it_handler:19 TZC permission failure E/TC:0 dump_fail_filter:420 Permission violation on filter 0 E/TC:0 dump_fail_filter:425 Violation @0xde5c6bf0, non-secure privileged read,
AXI ID 5c0
E/TC:0 Panic
After investigation, the forbidden access is a speculative request performed by the Cortex A7 because all the DDR is mapped as MEMORY with CACHEABLE property.
The issue is solved only when the region reserved by OP-TEE is no more mapped in U-Boot as it is already done in Linux kernel.
Tested on DK2 board with OP-TEE 3.12 / TF-A 2.4:
With hard-coded address for OP-TEE reserved memory, the error doesn't occur.
void dram_bank_mmu_setup(int bank)
{
....
for (i = start >> MMU_SECTION_SHIFT;
i < (start >> MMU_SECTION_SHIFT) + (size >> MMU_SECTION_SHIFT); i++) {
option = DCACHE_DEFAULT_OPTION;
if (i >= 0xde0)
option = INVALID_ENTRY;
set_section_dcache(i, option);
}
}
Just by modifying the test on 0xde0 to 0xdf0, the OP-TEE memory protected by firewall is mapped cacheable and the error occurs.
I think that can be a general issue for ARM architecture: the no-map tag in device should be respected by U-Boot.
But I don't propose a generic
solution in arm/lib/cache-cp15.c:dram_bank_mmu_setup() because the device tree parsing done in lmb_init_and_reserve() take a long time when it is executed without data cache.
=> the previous path 7/7 of v2 series is dropped to avoid
performance issue on other ARM target.
To avoid this issue on stm32mp32mp platform, this V3 series moves the lmb initialization in enable_caches() and the lmb variable becomes a static struct.
This v3 series is composed by 7 patches
- 1..3/7: preliminary steps to support flags in library in lmb (as it is done in memblock.c in Linux)
- 4/7: unitary test on the added feature in lmb lib
- 5/7: save the no-map flags in lmb when the device tree is parsed
- 6/7: solve issue for the size of cacheable area in pre-reloc case
- 7/7: update the stm32mp mmu support
See also [4] which handle same speculative access on armv8 for area with Executable attribute.
[1] http://patchwork.ozlabs.org/project/uboot/list/?series=228543&state=*
[2] http://patchwork.ozlabs.org/project/uboot/list/?series=228202&state=*
[3] http://patchwork.ozlabs.org/project/uboot/list/?series=227570&state=*
[4] http://patchwork.ozlabs.org/project/uboot/patch/20200903000106.5016-1-marek.bykowski at gmail.com/
Regards
Patrick
Changes in v3:
- NEW: solve performance issue as relocated DT is not marked cacheable
- call lmb_init_and_reserve when data cache is activated in enable_caches()
- drop v2 patch "arm: cache: cp15: don't map the reserved region with no-map property"
Changes in v2:
- remove unnecessary comments in lmb.h
- rebase on latest lmb patches
- NEW: update in stm32mp specific MMU setup functions
Patrick Delaunay (7):
lmb: Add support of flags for no-map properties
lmb: add lmb_is_reserved_flags
lmb: add lmb_dump_region() function
test: lmb: add test for lmb_reserve_flags
image-fdt: save no-map parameter of reserve-memory
stm32mp: Increase the reserved memory in board_get_usable_ram_top
stm32mp: don't map the reserved region with no-map property
arch/arm/mach-stm32mp/cpu.c | 17 +++++- arch/arm/mach-stm32mp/dram_init.c | 3 +- common/image-fdt.c | 23 +++++--- include/lmb.h | 21 +++++++ lib/lmb.c | 94 ++++++++++++++++++++++--------- test/lib/lmb.c | 89 +++++++++++++++++++++++++++++ 6 files changed, 209 insertions(+), 38 deletions(-)
Signature · 1 line
2.17.1
PATCH v3 1/7 lmb: Add support of flags for no-map properties
Add "flags" in lmb_property to save the "no-map" property of reserved region and a new function lmb_reserve_flags() to check this flag.
The default allocation use flags = LMB_NONE.
The adjacent reserved memory region are merged only when they have the same flags value.
This patch is partially based on flags support done in Linux kernel mm/memblock .c (previously lmb.c); it is why LMB_NOMAP = 0x4, it is aligned with MEMBLOCK_NOMAP value.
---
(no changes since v2)
Changes in v2:
- remove unnecessary comments in lmb.h
- rebase on latest lmb patches
include/lmb.h | 20 ++++++++++++++++++++ lib/lmb.c | 52 ++++++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 63 insertions(+), 9 deletions(-)
include/lmb.h +20 -0
@@ -12,6 +12,16 @@* Copyright (C) 2001 Peter Bergner, IBM Corp. */ +/** +* enum lmb_flags - definition of memory region attributes +* @LMB_NONE: no special request +* @LMB_NOMAP: don't add to mmu configuration +*/ +enum lmb_flags { +LMB_NONE =0x0 , +LMB_NOMAP =0x4 , +}; +/** * struct lmb_property - Description of one region. *
@@ -21,6 +31,7 @@struct lmb_property {phys_addr_t base;phys_size_t size; +enum lmb_flags flags; };/**
@@ -69,6 +80,8 @@ extern void lmb_init_and_reserve_range(struct lmb *lmb, phys_addr_t base, phys_size_t size, void *fdt_blob);extern long lmb_add (struct lmb *lmb,phys_addr_t base,phys_size_t size);extern long lmb_reserve (struct lmb *lmb,phys_addr_t base,phys_size_t size); +extern long lmb_reserve_flags (struct lmb *lmb,phys_addr_t base, +phys_size_t size,enum lmb_flags flags);extern phys_addr_t lmb_alloc (struct lmb *lmb,phys_size_t size,ulong align);extern phys_addr_t lmb_alloc_base (struct lmb *lmb,phys_size_t size,ulong align,phys_addr_t max_addr);
@@ -92,6 +105,13 @@ lmb_size_bytes(struct lmb_region *type, unsigned long region_nr)void board_lmb_reserve (struct lmb *lmb);void arch_lmb_reserve (struct lmb *lmb); +/* Low level functions */ + +static inline bool lmb_is_nomap (struct lmb_property *m) +{ +return !!(m->flags &LMB_NOMAP ); +} +#endif /* __KERNEL__ */ #endif /* _LINUX_LMB_H */
lib/lmb.c +43 -9
135 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH v3 1/7 lmb: Add support of flags for no-map properties
Hi Patrick,
Quoted 19 lines · Patrick Delaunay
Add "flags" in
lmb_propertyto save the "no-map" property of reserved region and a new functionlmb_reserve_flags()to check this flag.The default allocation use flags =
LMB_NONE.The adjacent reserved memory region are merged only when they have the same flags value.
This patch is partially based on flags support done in Linux kernel mm/memblock .c (previously
lmb.c); it is whyLMB_NOMAP=0x4, it is aligned withMEMBLOCK_NOMAPvalue.Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v2)
Quoted 7 lines
Changes in v2:
- remove unnecessary comments inlmb.h
- rebase on latest lmb patchesinclude/lmb.h | 20 ++++++++++++++++++++ lib/lmb.c | 52 ++++++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 63 insertions(+), 9 deletions(-)
Quoted 31 lines
include/lmb.h +11
-0@@ -12,6 +12,16 @@* Copyright (C) 2001 Peter Bergner, IBM Corp. */ +/** +* enum lmb_flags - definition of memory region attributes +* @LMB_NONE: no special request +* @LMB_NOMAP: don't add to mmu configuration +*/ +enum lmb_flags { +LMB_NONE =0x0 , +LMB_NOMAP =0x4 , +}; +/** * struct lmb_property - Description of one region. * @@ -21,6 +31,7 @@struct lmb_property {phys_addr_t base;phys_size_t size; +enum lmb_flags flags; };/**
include/lmb.h extern void lmb_init_and_reserve_range(struct lmb *lmb, phys_addr_t base, · line 84
Quoted 5 lines
include/lmb.h +2
-0@@ -69,6 +80,8 @@ extern void lmb_init_and_reserve_range(struct lmb *lmb, phys_addr_t base, phys_size_t size, void *fdt_blob);extern long lmb_add (struct lmb *lmb,phys_addr_t base,phys_size_t size);extern long lmb_reserve (struct lmb *lmb,phys_addr_t base,phys_size_t size); +extern long lmb_reserve_flags (struct lmb *lmb,phys_addr_t base, +phys_size_t size,enum lmb_flags flags);
Needs a comment
[..]
Regards,
Simon
PATCH v3 2/7 lmb: add lmb_is_reserved_flags
In reply to Patrick DelaunayAdd a new function lmb_is_reserved_flags to check is a address is reserved with a specific flags.
This function can be used to check if an address was reserved with no-map flags with:
lmb_is_reserved_flags(lmb, addr, LMB_NOMAP);
---
(no changes since v1)
include/lmb.h | 1 + lib/lmb.c | 10 ++++++++-- 2 files changed, 9 insertions(+), 2 deletions(-)
include/lmb.h +1 -0
@@ -91,6 +91,7 @@ extern phys_addr_t lmb_alloc_addr(struct lmb *lmb, phys_addr_t base, phys_size_t size);extern phys_size_t lmb_get_free_size (struct lmb *lmb,phys_addr_t addr);extern int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr); +extern int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags);extern long lmb_free (struct lmb *lmb,phys_addr_t base,phys_size_t size);extern void lmb_dump_all (struct lmb *lmb);
lib/lmb.c +8 -2
@@ -443,7 +443,7 @@ phys_size_t lmb_get_free_size(struct lmb *lmb, phys_addr_t addr)return 0 ; } -int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags) {int i;
@@ -451,11 +451,17 @@ int lmb_is_reserved(struct lmb *lmb, phys_addr_t addr)phys_addr_t upper = lmb->reserved.region[i].base + lmb->reserved.region[i].size -1 ;if ((addr >= lmb->reserved.region[i].base) && (addr <= upper)) -return 1 ; +return !!((lmb->reserved.region[i].flags & flags) + == flags); }return 0 ; } +int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +{ +return lmb_is_reserved_flags (lmb, addr,LMB_NONE ); +} +__weak void board_lmb_reserve (struct lmb *lmb) {/* please define platform specific board_lmb_reserve() */
Signature · 1 line
2.17.1
PATCH v3 2/7 lmb: add lmb_is_reserved_flags
Hi Patrick,
Quoted 23 lines · Patrick Delaunay
Add a new function
lmb_is_reserved_flagsto check is a address is reserved with a specific flags.This function can be used to check if an address was reserved with no-map flags with:
lmb_is_reserved_flags(lmb, addr, LMB_NOMAP);Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
include/lmb.h | 1 + lib/lmb.c | 10 ++++++++-- 2 files changed, 9 insertions(+), 2 deletions(-)include/lmb.h +0
-0
include/lmb.h extern phys_addr_t lmb_alloc_addr(struct lmb *lmb, phys_addr_t base, · line 94
Quoted 4 lines
include/lmb.h +1
-0@@ -91,6 +91,7 @@ extern phys_addr_t lmb_alloc_addr(struct lmb *lmb, phys_addr_t base, phys_size_t size);extern phys_size_t lmb_get_free_size (struct lmb *lmb,phys_addr_t addr);extern int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr); +extern int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags);
drop extern and please add a function comment
Quoted 17 lines
include/lmb.h +0
-0extern long lmb_free (struct lmb *lmb,phys_addr_t base,phys_size_t size);extern void lmb_dump_all (struct lmb *lmb);lib/lmb.c +1
-1@@ -443,7 +443,7 @@ phys_size_t lmb_get_free_size(struct lmb *lmb, phys_addr_t addr)return 0 ; } -int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags) {int i;
lib/lmb.c int lmb_is_reserved(struct lmb *lmb, phys_addr_t addr) · line 455
Quoted 6 lines
lib/lmb.c +2
-1@@ -451,11 +451,17 @@ int lmb_is_reserved(struct lmb *lmb, phys_addr_t addr)phys_addr_t upper = lmb->reserved.region[i].base + lmb->reserved.region[i].size -1 ;if ((addr >= lmb->reserved.region[i].base) && (addr <= upper)) -return 1 ; +return !!((lmb->reserved.region[i].flags & flags) + == flags);
I don't know what flags is since there is no comment, but it seems that you should drop the !!()
Quoted 13 lines
lib/lmb.c +5
-1}return 0 ; } +int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +{ +return lmb_is_reserved_flags (lmb, addr,LMB_NONE ); +} +__weak void board_lmb_reserve (struct lmb *lmb) {/* please define platform specific board_lmb_reserve() */ --
Quoted 2 lines
2.17.1
Regards,
Simon
PATCH v3 2/7 lmb: add lmb_is_reserved_flags
Quoted 21 lines · Simon Glass
Hi Patrick,
Quoted 17 lines · Patrick Delaunay
Add a new function
lmb_is_reserved_flagsto check is a address is reserved with a specific flags.This function can be used to check if an address was reserved with no-map flags with:
lmb_is_reserved_flags(lmb, addr, LMB_NOMAP);Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
include/lmb.h | 1 + lib/lmb.c | 10 ++++++++-- 2 files changed, 9 insertions(+), 2 deletions(-)
Quoted 10 lines
include/lmb.h extern phys_addr_t lmb_alloc_addr(struct lmb *lmb, phys_addr_t base, · line 94
Quoted 9 lines
include/lmb.h +1
-0@@ -91,6 +91,7 @@ extern phys_addr_t lmb_alloc_addr(struct lmb *lmb, phys_addr_t base, phys_size_t size);extern phys_size_t lmb_get_free_size (struct lmb *lmb,phys_addr_t addr);extern int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr); +extern int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags);drop extern and please add a function comment
I chosed to use extern to by aligned with other function and the linux memblock library.
But I will drop them in v4
Quoted 16 lines
Quoted 16 lines
include/lmb.h +0
-0extern long lmb_free (struct lmb *lmb,phys_addr_t base,phys_size_t size);extern void lmb_dump_all (struct lmb *lmb);lib/lmb.c +1
-1@@ -443,7 +443,7 @@ phys_size_t lmb_get_free_size(struct lmb *lmb, phys_addr_t addr)return 0 ; } -int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +int lmb_is_reserved_flags (struct lmb *lmb,phys_addr_t addr,int flags) {int i;
Quoted 9 lines
lib/lmb.c int lmb_is_reserved(struct lmb *lmb, phys_addr_t addr) · line 455
Quoted 7 lines
lib/lmb.c +2
-1@@ -451,11 +451,17 @@ int lmb_is_reserved(struct lmb *lmb, phys_addr_t addr)phys_addr_t upper = lmb->reserved.region[i].base + lmb->reserved.region[i].size -1 ;if ((addr >= lmb->reserved.region[i].base) && (addr <= upper)) -return 1 ; +return !!((lmb->reserved.region[i].flags & flags) + == flags);I don't know what flags is since there is no comment, but it seems that you should drop the !!()
Yes, it is a old C-coding habit.
Quoted 17 lines
Quoted 13 lines
lib/lmb.c +5
-1}return 0 ; } +int lmb_is_reserved (struct lmb *lmb,phys_addr_t addr) +{ +return lmb_is_reserved_flags (lmb, addr,LMB_NONE ); +} +__weak void board_lmb_reserve (struct lmb *lmb) {/* please define platform specific board_lmb_reserve() */ --Quoted 2 lines
2.17.1
Regards,
Simon
PATCH v3 3/7 lmb: add lmb_dump_region() function
In reply to Patrick DelaunayAdd lmb_dump_region() function, to simplify lmb_dump_all_force(). This patch is based on Linux memblock dump function.
An example of bdinfo output is:
.....fdt_size = 0x000146a0
FB base = 0xfdd00000lmb_dump_all:
memory.cnt = 0x1
memory[0] [0xc0000000-0xffffffff], 0x40000000 bytes flags: 0
reserved.cnt = 0x6
reserved[0] [0x10000000-0x10045fff], 0x00046000 bytes flags: 4
reserved[1] [0x30000000-0x3003ffff], 0x00040000 bytes flags: 4
reserved[2] [0x38000000-0x3800ffff], 0x00010000 bytes flags: 4
reserved[3] [0xe8000000-0xefffffff], 0x08000000 bytes flags: 4
reserved[4] [0xfbaea344-0xfdffffff], 0x02515cbc bytes flags: 0
reserved[5] [0xfe000000-0xffffffff], 0x02000000 bytes flags: 4arch_number = 0x00000000
TLB addr = 0xfdff0000
....
---
(no changes since v1)
lib/lmb.c | 40 ++++++++++++++++++++-------------------- 1 file changed, 20 insertions(+), 20 deletions(-)
lib/lmb.c +20 -20
@@ -14,32 +14,32 @@#define LMB_ALLOC_ANYWHERE 0 -void lmb_dump_all_force (struct lmb *lmb) +static void lmb_dump_region (struct lmb_region *rgn,char *name) { -unsigned long i; +unsigned long long base, size, end; +enum lmb_flags flags; +int i; -printf ("lmb_dump_all:\n" ); -printf (" memory.cnt = 0x%lx\n" , lmb->memory.cnt); -for (i =0 ; i < lmb->memory.cnt; i++) { -printf (" memory.reg[0x%lx].base = 0x%llx\n" , i, - (unsigned long long )lmb->memory.region[i].base); -printf (" .size = 0x%llx\n" , - (unsigned long long )lmb->memory.region[i].size); -printf (" .flags = 0x%x\n" , - lmb->memory.region[i].flags); - } +printf (" %s.cnt = 0x%lx\n" , name, rgn->cnt); -printf ("\n reserved.cnt = 0x%lx\n" , lmb->reserved.cnt); -for (i =0 ; i < lmb->reserved.cnt; i++) { -printf (" reserved.reg[0x%lx].base = 0x%llx\n" , i, - (unsigned long long )lmb->reserved.region[i].base); -printf (" .size = 0x%llx\n" , - (unsigned long long )lmb->reserved.region[i].size); -printf (" .flags = 0x%x\n" , - lmb->reserved.region[i].flags); +for (i =0 ; i < rgn->cnt; i++) { + base = rgn->region[i].base; + size = rgn->region[i].size; + end = base + size -1 ; + flags = rgn->region[i].flags; + +printf (" %s[%d]\t[0x%llx-0x%llx], 0x%08llx bytes flags: %x\n" , + name, i, base, end, size, flags); } } +void lmb_dump_all_force (struct lmb *lmb) +{ +printf ("lmb_dump_all:\n" ); +lmb_dump_region (&lmb->memory,"memory" ); +lmb_dump_region (&lmb->reserved,"reserved" ); +} +void lmb_dump_all (struct lmb *lmb) {#ifdef DEBUG
Signature · 1 line
2.17.1
PATCH v3 3/7 lmb: add lmb_dump_region() function
Quoted 23 lines · Patrick Delaunay
Add
lmb_dump_region()function, to simplifylmb_dump_all_force(). This patch is based on Linux memblock dump function.An example of bdinfo output is:
.....
fdt_size=0x000146a0
FB base =0xfdd00000lmb_dump_all:
memory.cnt =0x1
memory[0] [0xc0000000-0xffffffff],0x40000000bytes flags: 0
reserved.cnt =0x6reserved[0] [0x10000000-0x10045fff], 0x00046000 bytes flags: 4 reserved[1] [0x30000000-0x3003ffff], 0x00040000 bytes flags: 4 reserved[2] [0x38000000-0x3800ffff], 0x00010000 bytes flags: 4 reserved[3] [0xe8000000-0xefffffff], 0x08000000 bytes flags: 4 reserved[4] [0xfbaea344-0xfdffffff], 0x02515cbc bytes flags: 0 reserved[5] [0xfe000000-0xffffffff], 0x02000000 bytes flags: 4
arch_number=0x00000000
TLB addr =0xfdff0000
....
Quoted 7 lines
Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
lib/lmb.c | 40 ++++++++++++++++++++-------------------- 1 file changed, 20 insertions(+), 20 deletions(-)
PATCH v3 4/7 test: lmb: add test for lmb_reserve_flags
In reply to Patrick DelaunayAdd a test to check the management of reserved region with flags.
---
(no changes since v1)
test/lib/lmb.c | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 89 insertions(+)
test/lib/lmb.c +89 -0
@@ -723,3 +723,92 @@ static int lib_test_lmb_max_regions(struct unit_test_state *uts)DM_TEST (lib_test_lmb_max_regions,UT_TESTF_SCAN_PDATA |UT_TESTF_SCAN_FDT ); + +static int lib_test_lmb_flags (struct unit_test_state *uts) +{ +const phys_addr_t ram =0x40000000 ; +const phys_size_t ram_size =0x20000000 ; +struct lmb lmb; +long ret; + +lmb_init (&lmb); + + ret =lmb_add (&lmb, ram, ram_size); +ut_asserteq (ret,0 ); + +/* reserve, same flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +/* reserve again, same flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +/* reserve again, new flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NONE ); +ut_asserteq (ret,-1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); + +/* merge after */ + ret =lmb_reserve_flags (&lmb,0x40020000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x20000 , +0 ,0 ,0 ,0 ); + +/* merge before */ + ret =lmb_reserve_flags (&lmb,0x40000000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40000000 ,0x30000 , +0 ,0 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); + + ret =lmb_reserve_flags (&lmb,0x40030000 ,0x10000 ,LMB_NONE ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,2 ,0x40000000 ,0x30000 , +0x40030000 ,0x10000 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); + +/* test that old API use LMB_NONE */ + ret =lmb_reserve (&lmb,0x40040000 ,0x10000 ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,2 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); + + ret =lmb_reserve_flags (&lmb,0x40070000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,3 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40070000 ,0x10000 ); + + ret =lmb_reserve_flags (&lmb,0x40050000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,4 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40050000 ,0x10000 ); + +/* merge with 2 adjacent regions */ + ret =lmb_reserve_flags (&lmb,0x40060000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,2 ); +ASSERT_LMB (&lmb, ram, ram_size,3 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40050000 ,0x30000 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[2 ]),1 ); + +return 0 ; +} + +DM_TEST (lib_test_lmb_flags, +UT_TESTF_SCAN_PDATA |UT_TESTF_SCAN_FDT );
Signature · 1 line
2.17.1
PATCH v3 4/7 test: lmb: add test for lmb_reserve_flags
Quoted 10 lines · Patrick Delaunay
Add a test to check the management of reserved region with flags.
Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
test/lib/lmb.c | 89 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 89 insertions(+)
Quoted 99 lines
test/lib/lmb.c +89
-1@@ -723,3 +723,92 @@ static int lib_test_lmb_max_regions(struct unit_test_state *uts)DM_TEST (lib_test_lmb_max_regions,UT_TESTF_SCAN_PDATA |UT_TESTF_SCAN_FDT ); + +static int lib_test_lmb_flags (struct unit_test_state *uts) +{ +const phys_addr_t ram =0x40000000 ; +const phys_size_t ram_size =0x20000000 ; +struct lmb lmb; +long ret; + +lmb_init (&lmb); + + ret =lmb_add (&lmb, ram, ram_size); +ut_asserteq (ret,0 ); + +/* reserve, same flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +/* reserve again, same flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +/* reserve again, new flag */ + ret =lmb_reserve_flags (&lmb,0x40010000 ,0x10000 ,LMB_NONE ); +ut_asserteq (ret,-1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x10000 , +0 ,0 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); + +/* merge after */ + ret =lmb_reserve_flags (&lmb,0x40020000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40010000 ,0x20000 , +0 ,0 ,0 ,0 ); + +/* merge before */ + ret =lmb_reserve_flags (&lmb,0x40000000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,1 ,0x40000000 ,0x30000 , +0 ,0 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); + + ret =lmb_reserve_flags (&lmb,0x40030000 ,0x10000 ,LMB_NONE ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,2 ,0x40000000 ,0x30000 , +0x40030000 ,0x10000 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); + +/* test that old API use LMB_NONE */ + ret =lmb_reserve (&lmb,0x40040000 ,0x10000 ); +ut_asserteq (ret,1 ); +ASSERT_LMB (&lmb, ram, ram_size,2 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0 ,0 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); + + ret =lmb_reserve_flags (&lmb,0x40070000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,3 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40070000 ,0x10000 ); + + ret =lmb_reserve_flags (&lmb,0x40050000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,0 ); +ASSERT_LMB (&lmb, ram, ram_size,4 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40050000 ,0x10000 ); + +/* merge with 2 adjacent regions */ + ret =lmb_reserve_flags (&lmb,0x40060000 ,0x10000 ,LMB_NOMAP ); +ut_asserteq (ret,2 ); +ASSERT_LMB (&lmb, ram, ram_size,3 ,0x40000000 ,0x30000 , +0x40030000 ,0x20000 ,0x40050000 ,0x30000 ); + +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[0 ]),1 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[1 ]),0 ); +ut_asserteq (lmb_is_nomap (&lmb.reserved.region[2 ]),1 ); + +return 0 ; +} + +DM_TEST (lib_test_lmb_flags, +UT_TESTF_SCAN_PDATA |UT_TESTF_SCAN_FDT ); --
Quoted 2 lines
2.17.1
Regards,
Simon
PATCH v3 5/7 image-fdt: save no-map parameter of reserve-memory
In reply to Patrick DelaunaySave the no-map information present in 'reserved-memory' node to allow correct handling when the MMU is configured in board to avoid speculative access.
---
(no changes since v1)
common/image-fdt.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)
common/image-fdt.c +15 -8
@@ -75,18 +75,20 @@ static const image_header_t *image_get_fdt(ulong fdt_addr)#endif static void boot_fdt_reserve_region (struct lmb *lmb,uint64_t addr, -uint64_t size) +uint64_t size,enum lmb_flags flags) {long ret; - ret =lmb_reserve (lmb, addr, size); + ret =lmb_reserve_flags (lmb, addr, size, flags);if (ret >=0 ) { -debug (" reserving fdt memory region: addr=%llx size=%llx\n" , - (unsigned long long )addr, (unsigned long long )size); +debug (" reserving fdt memory region: addr=%llx size=%llx flags=%x\n" , + (unsigned long long )addr, + (unsigned long long )size, flags); }else {puts ("ERROR: reserving fdt memory region failed " ); -printf ("(addr=%llx size=%llx)\n" , - (unsigned long long )addr, (unsigned long long )size); +printf ("(addr=%llx size=%llx flags=%x)\n" , + (unsigned long long )addr, + (unsigned long long )size, flags); } }
@@ -106,6 +108,7 @@ void boot_fdt_add_mem_rsv_regions(struct lmb *lmb, void *fdt_blob)int i, total, ret;int nodeoffset, subnode;struct fdt_resource res; +enum lmb_flags flags;if (fdt_check_header (fdt_blob) !=0 )return ;
@@ -115,7 +118,7 @@ void boot_fdt_add_mem_rsv_regions(struct lmb *lmb, void *fdt_blob)for (i =0 ; i < total; i++) {if (fdt_get_mem_rsv (fdt_blob, i, &addr, &size) !=0 )continue ; -boot_fdt_reserve_region (lmb, addr, size); +boot_fdt_reserve_region (lmb, addr, size,LMB_NONE ); }/* process reserved-memory */
@@ -127,9 +130,13 @@ void boot_fdt_add_mem_rsv_regions(struct lmb *lmb, void *fdt_blob) ret =fdt_get_resource (fdt_blob, subnode,"reg" ,0 , &res);if (!ret &&fdtdec_get_is_enabled (fdt_blob, subnode)) { + flags =LMB_NONE ; +if (fdtdec_get_bool (fdt_blob, subnode, +"no-map" )) + flags =LMB_NOMAP ; addr = res.start; size = res.end - res.start +1 ; -boot_fdt_reserve_region (lmb, addr, size); +boot_fdt_reserve_region (lmb, addr, size, flags); } subnode =fdt_next_subnode (fdt_blob, subnode);
Signature · 1 line
2.17.1
PATCH v3 5/7 image-fdt: save no-map parameter of reserve-memory
Hi Patrick,
Quoted 5 lines · Patrick Delaunay
Save the no-map information present in 'reserved-memory' node to allow correct handling when the MMU is configured in board to avoid speculative access.
Quoted 7 lines
Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
common/image-fdt.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)
Where is no-map documented?
PATCH v3 5/7 image-fdt: save no-map parameter of reserve-memory
Hi Simon,
Quoted 13 lines · Simon Glass
Hi Patrick,
Quoted 8 lines · Patrick Delaunay
Save the no-map information present in 'reserved-memory' node to allow correct handling when the MMU is configured in board to avoid speculative access.
Signed-off-by Patrick Delaunay <[email protected]>---
Quoted 6 lines
Quoted 4 lines
(no changes since v1)
common/image-fdt.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)Where is no-map documented?
See Linux kernel Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
Quoted 2 lines
Reviewed-by Simon Glass <[email protected]>
Regards,
Bin
PATCH v3 5/7 image-fdt: save no-map parameter of reserve-memory
In reply to Simon GlassQuoted 13 lines · Simon Glass
Hi Patrick,
Quoted 8 lines · Patrick Delaunay
Save the no-map information present in 'reserved-memory' node to allow correct handling when the MMU is configured in board to avoid speculative access.
Signed-off-by Patrick Delaunay <[email protected]>---
Quoted 8 lines
Quoted 4 lines
(no changes since v1)
common/image-fdt.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)Where is no-map documented?
Reviewed-by Simon Glass <[email protected]>
I don't remember exactly where the discussion ended up the last time, so please disregard this if we settled it, but I still don't think that secure-only memory should be described in the DT at all.
The v7 and v8 documentation are not 100% aligned on this, but the S bit is now considered a true address bit, and so secure address 0x0 and non-secure address zero could be decoded by different peripherals at the same time, even if some TZ firewall implementations don't implement it that way.
Since DT addresses don't carry the S bit at all, any address described in DT must be assumed to be a non-secure address. This means that the no-map region essentially describes a non-secure region that does not exist, in order to prevent a secure region at the same offset (which DT cannot describe in the first place) from being covered by the linear mapping.
So, apologies if we are going around in circles here, but could you please explain again why the DT is describing secure memory as non-secure memory, and then reserving it again using no-map?
PATCH v3 5/7 image-fdt: save no-map parameter of reserve-memory
In reply to Simon GlassHi Simon,
Quoted 8 lines · Simon Glass
Hi Patrick,
Quoted 4 lines · Patrick Delaunay
Save the no-map information present in 'reserved-memory' node to allow correct handling when the MMU is configured in board to avoid speculative access.
Quoted 8 lines
Quoted 7 lines
Signed-off-by Patrick Delaunay <[email protected]>---
(no changes since v1)
common/image-fdt.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)Where is no-map documented?
In linux:Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt
in U-Boot: doc/device-tree-bindings/reserved-memory/reserved-memory.txt:55
no-map (optional) - empty property
??? - Indicates the operating system must not create a virtual mapping ????? of the region as part of its standard mapping of system memory, ????? nor permit speculative access to it under any circumstances other ????? than under the control of the device driver using the region.
But I will add this indication in the commit message.
Quoted 2 lines
Reviewed-by Simon Glass <[email protected]>
Thanks,
Patrick
PATCH v3 6/7 stm32mp: Increase the reserved memory in board_get_usable_ram_top
In reply to Patrick DelaunayAdd 8M for the U-Boot reserved memory (display, fdt, gd, ...).
Without this patch the device tree, located before the MALLOC area is not tagged cacheable just after relocation, before mmu reconfiguration.
This patch reduces the duration for device tree parsing in lmb_init_and_reserve.
---
Changes in v3:
- NEW: solve performance issue as relocated DT is not marked cacheable
arch/arm/mach-stm32mp/dram_init.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
arch/arm/mach-stm32mp/dram_init.c +2 -1
@@ -50,7 +50,8 @@ ulong board_get_usable_ram_top(ulong total_size)lmb_init (&lmb);lmb_add (&lmb, gd->ram_base, gd->ram_size);boot_fdt_add_mem_rsv_regions (&lmb, (void *)gd->fdt_blob); - size =ALIGN (CONFIG_SYS_MALLOC_LEN + total_size,MMU_SECTION_SIZE ), +/* add 8M for reserved memory for display, fdt, gd,... */ + size =ALIGN (SZ_8M +CONFIG_SYS_MALLOC_LEN + total_size,MMU_SECTION_SIZE ), reg =lmb_alloc (&lmb, size,MMU_SECTION_SIZE );if (!reg)
Signature · 1 line
2.17.1
PATCH v3 7/7 stm32mp: don't map the reserved region with no-map property
In reply to Patrick DelaunayNo more map the reserved region with "no-map" property by marking the corresponding TLB entries with invalid entry (=0) to avoid speculative access.
The device tree parsing done in lmb_init_and_reserve() takes a long time when it is executed without data cache, so it is called in enable_caches() before to disable it.
This patch fixes an issue where predictive read access on secure DDR OP-TEE reserved area are caught by firewall.
---
Changes in v3:
- call lmb_init_and_reserve when data cache is activated in enable_caches()
- drop v2 patch "arm: cache: cp15: don't map the reserved region with no-map property"
Changes in v2:
- NEW: update in stm32mp specific MMU setup functions
arch/arm/mach-stm32mp/cpu.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
arch/arm/mach-stm32mp/cpu.c +15 -2
@@ -12,6 +12,7 @@#include <env.h> #include <init.h> #include <log.h> +#include <lmb.h> #include <misc.h> #include <net.h> #include <asm/io.h>
@@ -90,6 +91,8 @@*/ u8 early_tlb[PGTABLE_SIZE] __section(".data") __aligned(0x4000); +struct lmb lmb; +#if !defined(CONFIG_SPL ) || defined(CONFIG_SPL_BUILD )#ifndef CONFIG_TFABOOT static void security_init (void )
@@ -221,6 +224,8 @@ void dram_bank_mmu_setup(int bank)int i;phys_addr_t start;phys_size_t size; +bool use_lmb = false; +enum dcache_option option;if (IS_ENABLED (CONFIG_SPL_BUILD )) { start =ALIGN_DOWN (STM32_SYSRAM_BASE ,MMU_SECTION_SIZE );
@@ -229,6 +234,7 @@ void dram_bank_mmu_setup(int bank)/* bd->bi_dram is available only after relocation */ start = bd->bi_dram[bank].start; size = bd->bi_dram[bank].size; + use_lmb = true; }else {/* mark cacheable and executable the beggining of the DDR */ start =STM32_DDR_BASE ;
@@ -237,8 +243,12 @@ void dram_bank_mmu_setup(int bank)for (i = start >>MMU_SECTION_SHIFT ; i < (start >>MMU_SECTION_SHIFT ) + (size >>MMU_SECTION_SHIFT ); - i++) -set_section_dcache (i,DCACHE_DEFAULT_OPTION ); + i++) { + option =DCACHE_DEFAULT_OPTION ; +if (use_lmb &&lmb_is_reserved_flags (&lmb, i <<MMU_SECTION_SHIFT ,LMB_NOMAP )) + option =0 ;/* INVALID ENTRY in TLB */ +set_section_dcache (i, option); + } }/* * initialize the MMU and activate cache in SPL or in U-Boot pre-reloc stage
@@ -302,6 +312,9 @@ int arch_cpu_init(void)void enable_caches (void ) { +/* parse device tree when data cache is still activated */ +lmb_init_and_reserve (&lmb, gd->bd, (void *)gd->fdt_blob); +/* I-cache is already enabled in start.S: icache_enable() not needed */ /* deactivate the data cache, early enabled in arch_cpu_init() */
Signature · 1 line
2.17.1