powerpc: enable the relocatable support for fsl booke 32bit kernel
Subsystems (2)
v4 cover powerpc: enable the relocatable support for fsl booke 32bit kernel
v4:
- Fix the bug when booting above 64M.
- Rebase onto v3.13-rc5
- Pass the following test on a p5020ds board:
boot kernel at 0x5000000 and 0x9000000 kdump test with kernel option "crashkernel=64M@80M"
v3:
The main changes include:
* Drop the patch 5 in v2 (memblock: introduce the memblock_reinit function)
* Change to use the 64M boot init tlb.
Please refer to the comment section of each patch for more detail.
This patch series passed the kdump test with kernel option "crashkernel=64M@32M" and "crashkernel=64M@80M" on a p2020rdb board.
v2:
These patches are based on the Ben's next branch. In this version we choose to do a second relocation if the PAGE_OFFSET is not mapped to the memstart_addr and we also choose to set the tlb1 entries for the kernel space in address space 1. With this implementation:
* We can load the kernel at any place between
memstart_addr ~ memstart_addr + 768M
* We can reserve any memory between memstart_addr ~ memstart_addr + 768M for a kdump kernel.
I have done a kdump boot on a p2020rdb kernel with the memory reserved by 'crashkernel=32M@320M'.
v1:
Currently the fsl booke 32bit kernel is using the DYNAMIC_MEMSTART relocation method. But the RELOCATABLE method is more flexible and has less alignment restriction. So enable this feature on this platform and use it by default for the kdump kernel.
These patches have passed the kdump boot test on a p2020rdb board.
---
Kevin Hao (10):
powerpc/fsl_booke: protect the access to MAS7 powerpc/fsl_booke: introduce get_phys_addr function
powerpc: introduce macro LOAD_REG_ADDR_PIC
powerpc: enable the relocatable support for the fsl booke 32bit kernel powerpc/fsl_booke: set the tlb entry for the kernel address in AS1
powerpc: introduce early_get_first_memblock_info powerpc/fsl_booke: introduce map_mem_in_cams_addr powerpc/fsl_booke: make sure PAGE_OFFSET map to memstart_addr for
relocatable kernel
powerpc/fsl_booke: smp support for booting a relocatable kernel above
64M
powerpc/fsl_booke: enable the relocatable for the kdump kernel
arch/powerpc/Kconfig | 5 +- arch/powerpc/include/asm/ppc_asm.h | 13 ++ arch/powerpc/kernel/fsl_booke_entry_mapping.S | 2 + arch/powerpc/kernel/head_fsl_booke.S | 266 +++++++++++++++++++++++--- arch/powerpc/kernel/prom.c | 41 +++- arch/powerpc/mm/fsl_booke_mmu.c | 72 ++++++- arch/powerpc/mm/hugetlbpage-book3e.c | 3 +- arch/powerpc/mm/mmu_decl.h | 2 + arch/powerpc/mm/tlb_nohash_low.S | 4 +- include/linux/of_fdt.h | 1 + 10 files changed, 370 insertions(+), 39 deletions(-)
Signature · 1 line
1.8.3.1
PATCH v4 1/10 powerpc/fsl_booke: protect the access to MAS7
The e500v1 doesn't implement the MAS7, so we should avoid to access this register on that implementations. In the current kernel, the access to MAS7 are protected by either CONFIG_PHYS_64BIT or MMU_FTR_BIG_PHYS. Since some code are executed before the code patching, we have to use CONFIG_PHYS_64BIT in these cases.
---
v4: No change.
v3: Use ifdef CONFIG_PHYS_64BIT for the code running before code patching.
v2: A new patch in v2.
arch/powerpc/kernel/head_fsl_booke.S | 2 ++ arch/powerpc/mm/hugetlbpage-book3e.c | 3 ++- 2 files changed, 4 insertions(+), 1 deletion(-)
arch/powerpc/kernel/head_fsl_booke.S +2 -0
@@ -82,7 +82,9 @@ _ENTRY(_start); and r19,r3,r18 /* r19 = page offset */ andc r31,r20,r18 /* r31 = page base */ or r31,r31,r19 /* r31 = devtree phys addr */ +#ifdef CONFIG_PHYS_64BIT mfspr r30,SPRN_MAS7 +#endif li r25,0 /* phys kernel start (low) */ li r24,0 /* CPU number */
arch/powerpc/mm/hugetlbpage-book3e.c +2 -1
@@ -103,7 +103,8 @@ void book3e_hugetlb_preload(struct vm_area_struct *vma, unsigned long ea,if (mmu_has_feature (MMU_FTR_USE_PAIRED_MAS )) {mtspr (SPRN_MAS7_MAS3 , mas7_3); }else { -mtspr (SPRN_MAS7 ,upper_32_bits (mas7_3)); +if (mmu_has_feature (MMU_FTR_BIG_PHYS )) +mtspr (SPRN_MAS7 ,upper_32_bits (mas7_3));mtspr (SPRN_MAS3 ,lower_32_bits (mas7_3)); }
Signature · 1 line
1.8.3.1
PATCH v4 2/10 powerpc/fsl_booke: introduce get_phys_addr function
In reply to Kevin HaoMove the codes which translate a effective address to physical address to a separate function. So it can be reused by other code.
---
v4: No change.
v3: Use ifdef CONFIG_PHYS_64BIT to protect the access to MAS7
v2: A new patch in v2.
arch/powerpc/kernel/head_fsl_booke.S | 50 +++++++++++++++++++++--------------- 1 file changed, 30 insertions(+), 20 deletions(-)
arch/powerpc/kernel/head_fsl_booke.S +30 -20
@@ -65,26 +65,9 @@ _ENTRY(_start); nop /* Translate device tree address to physical, save in r30/r31 */ - mfmsr r16 - mfspr r17,SPRN_PID - rlwinm r17,r17,16,0x3fff0000 /* turn PID into MAS6[SPID] */ - rlwimi r17,r16,28,0x00000001 /* turn MSR[DS] into MAS6[SAS] */ - mtspr SPRN_MAS6,r17 - - tlbsx 0,r3 /* must succeed */ - - mfspr r16,SPRN_MAS1 - mfspr r20,SPRN_MAS3 - rlwinm r17,r16,25,0x1f /* r17 = log2(page size) */ - li r18,1024 - slw r18,r18,r17 /* r18 = page size */ - addi r18,r18,-1 - and r19,r3,r18 /* r19 = page offset */ - andc r31,r20,r18 /* r31 = page base */ - or r31,r31,r19 /* r31 = devtree phys addr */ -#ifdef CONFIG_PHYS_64BIT - mfspr r30,SPRN_MAS7 -#endif + bl get_phys_addr + mr r30,r3 + mr r31,r4 li r25,0 /* phys kernel start (low) */ li r24,0 /* CPU number */
@@ -858,6 +841,33 @@ KernelSPE: #endif /* CONFIG_SPE */ /* + * Translate the effec addr in r3 to phys addr. The phys addr will be put + * into r3(higher 32bit) and r4(lower 32bit) + */ +get_phys_addr: + mfmsr r8 + mfspr r9,SPRN_PID + rlwinm r9,r9,16,0x3fff0000 /* turn PID into MAS6[SPID] */ + rlwimi r9,r8,28,0x00000001 /* turn MSR[DS] into MAS6[SAS] */ + mtspr SPRN_MAS6,r9 + + tlbsx 0,r3 /* must succeed */ + + mfspr r8,SPRN_MAS1 + mfspr r12,SPRN_MAS3 + rlwinm r9,r8,25,0x1f /* r9 = log2(page size) */ + li r10,1024 + slw r10,r10,r9 /* r10 = page size */ + addi r10,r10,-1 + and r11,r3,r10 /* r11 = page offset */ + andc r4,r12,r10 /* r4 = page base */ + or r4,r4,r11 /* r4 = devtree phys addr */ +#ifdef CONFIG_PHYS_64BIT + mfspr r3,SPRN_MAS7 +#endif + blr + +/* * Global functions */
Signature · 1 line
1.8.3.1
PATCH v4 3/10 powerpc: introduce macro LOAD_REG_ADDR_PIC
In reply to Kevin HaoThis is used to get the address of a variable when the kernel is not running at the linked or relocated address.
---
v4: A new patch in v4.
arch/powerpc/include/asm/ppc_asm.h | 13 +++++++++++++ 1 file changed, 13 insertions(+)
arch/powerpc/include/asm/ppc_asm.h +13 -0
@@ -295,6 +295,11 @@ n:* you want to access various offsets within it). On ppc32 this is * identical to LOAD_REG_IMMEDIATE. * +* LOAD_REG_ADDR_PIC(rn, name) +* Loads the address of label 'name' into register 'run'. Use this when +* the kernel doesn't run at the linked or relocated address. Please +* note that this macro will clobber the lr register. +* * LOAD_REG_ADDRBASE(rn, name) * ADDROFF(name) * LOAD_REG_ADDRBASE loads part of the address of label 'name' into
@@ -305,6 +310,14 @@ n:* LOAD_REG_ADDRBASE(rX, name) * ld rY,ADDROFF(name)(rX) */ + +/* Be careful, this will clobber the lr register. */ +#define LOAD_REG_ADDR_PIC (reg, name) \ + bl 0f; \ +0: mflr reg; \ + addis reg,reg,(name - 0b)@ha; \ + addi reg,reg,(name - 0b)@l; +#ifdef __powerpc64__#define LOAD_REG_IMMEDIATE (reg,expr) \ lis reg,(expr)@highest; \
Signature · 1 line
1.8.3.1
PATCH v4 4/10 powerpc: enable the relocatable support for the fsl booke 32bit kernel
In reply to Kevin HaoThis is based on the codes in the head_44x.S. The difference is that the init tlb size we used is 64M. With this patch we can only load the kernel at address between memstart_addr ~ memstart_addr + 64M. We will fix this restriction in the following patches.
---
v4: Use macro LOAD_REG_ADDR_PIC.
v3:
* Use the 64M align.
* typo fix.
v2: Move the code to set kernstart_addr and virt_phys_offset to a c function.
So we can expand it easily later.
arch/powerpc/Kconfig | 2 +- arch/powerpc/kernel/fsl_booke_entry_mapping.S | 2 ++ arch/powerpc/kernel/head_fsl_booke.S | 34 +++++++++++++++++++++++++++ arch/powerpc/mm/fsl_booke_mmu.c | 28 ++++++++++++++++++++++ 4 files changed, 65 insertions(+), 1 deletion(-)
arch/powerpc/Kconfig +1 -1
@@ -881,7 +881,7 @@ config DYNAMIC_MEMSTART config RELOCATABLE bool "Build a relocatable kernel" - depends on ADVANCED_OPTIONS && FLATMEM && 44x + depends on ADVANCED_OPTIONS && FLATMEM && (44x || FSL_BOOKE) select NONSTATIC_KERNEL help This builds a kernel image that is capable of running at the
arch/powerpc/kernel/fsl_booke_entry_mapping.S +2 -0
@@ -176,6 +176,8 @@ skpinv: addi r6,r6,1 /* Increment */ /* 7. Jump to KERNELBASE mapping */ lis r6,(KERNELBASE & ~0xfff)@h ori r6,r6,(KERNELBASE & ~0xfff)@l + rlwinm r7,r25,0,0x03ffffff + add r6,r7,r6 #elif defined(ENTRY_MAPPING_KEXEC_SETUP) /*
arch/powerpc/kernel/head_fsl_booke.S +34 -0
@@ -73,6 +73,30 @@ _ENTRY(_start); li r24,0 /* CPU number */ li r23,0 /* phys kernel start (high) */ +#ifdef CONFIG_RELOCATABLE + LOAD_REG_ADDR_PIC(r3, _stext) /* Get our current runtime base */ + + /* Translate _stext address to physical, save in r23/r25 */ + bl get_phys_addr + mr r23,r3 + mr r25,r4 + + /* + * We have the runtime (virutal) address of our base. + * We calculate our shift of offset from a 64M page. + * We could map the 64M page we belong to at PAGE_OFFSET and + * get going from there. + */ + lis r4,KERNELBASE@h + ori r4,r4,KERNELBASE@l + rlwinm r6,r25,0,0x3ffffff /* r6 = PHYS_START % 64M */ + rlwinm r5,r4,0,0x3ffffff /* r5 = KERNELBASE % 64M */ + subf r3,r5,r6 /* r3 = r6 - r5 */ + add r3,r4,r3 /* Required Virtual Address */ + + bl relocate +#endif + /* We try to not make any assumptions about how the boot loader * setup or used the TLBs. We invalidate all mappings from the * boot loader and load a single entry in TLB1[0] to map the
@@ -182,6 +206,16 @@ _ENTRY(__early_start) bl early_init +#ifdef CONFIG_RELOCATABLE +#ifdef CONFIG_PHYS_64BIT + mr r3,r23 + mr r4,r25 +#else + mr r3,r25 +#endif + bl relocate_init +#endif + #ifdef CONFIG_DYNAMIC_MEMSTART lis r3,kernstart_addr@ha la r3,kernstart_addr@l(r3)
arch/powerpc/mm/fsl_booke_mmu.c +28 -0
@@ -241,4 +241,32 @@ void setup_initial_memory_limit(phys_addr_t first_memblock_base,/* 64M mapped initially according to head_fsl_booke.S */ memblock_set_current_limit (min_t (u64, limit,0x04000000 )); } + +#ifdef CONFIG_RELOCATABLE +notrace void __init relocate_init(phys_addr_t start) +{ +unsigned long base =KERNELBASE ; + +/* +* Relocatable kernel support based on processing of dynamic +* relocation entries. +* Compute the virt_phys_offset : +* virt_phys_offset = stext.run - kernstart_addr +* +* stext.run = (KERNELBASE & ~0x3ffffff) + (kernstart_addr & 0x3ffffff) +* When we relocate, we have : +* +* (kernstart_addr & 0x3ffffff) = (stext.run & 0x3ffffff) +* +* hence: +* virt_phys_offset = (KERNELBASE & ~0x3ffffff) - +* (kernstart_addr & ~0x3ffffff) +* +*/ + kernstart_addr = start; + start &= ~0x3ffffff ; + base &= ~0x3ffffff ; + virt_phys_offset = base - start; +} +#endif #endif
Signature · 1 line
1.8.3.1
PATCH v4 5/10 powerpc/fsl_booke: set the tlb entry for the kernel address in AS1
In reply to Kevin HaoWe use the tlb1 entries to map low mem to the kernel space. In the current code, it assumes that the first tlb entry would cover the kernel image. But this is not true for some special cases, such as when we run a relocatable kernel above the 64M or set CONFIG_KERNEL_START above 64M. So we choose to switch to address space 1 before setting these tlb entries.
---
v4: No change.
v3: Typo fix.
v2: A new patch in v2.
arch/powerpc/kernel/head_fsl_booke.S | 81 ++++++++++++++++++++++++++++++++++++ arch/powerpc/mm/fsl_booke_mmu.c | 2 + arch/powerpc/mm/mmu_decl.h | 2 + 3 files changed, 85 insertions(+)
arch/powerpc/kernel/head_fsl_booke.S +81 -0
@@ -1157,6 +1157,87 @@ __secondary_hold_acknowledge: #endif /* + * Create a tlb entry with the same effective and physical address as + * the tlb entry used by the current running code. But set the TS to 1. + * Then switch to the address space 1. It will return with the r3 set to + * the ESEL of the new created tlb. + */ +_GLOBAL(switch_to_as1) + mflr r5 + + /* Find a entry not used */ + mfspr r3,SPRN_TLB1CFG + andi. r3,r3,0xfff + mfspr r4,SPRN_PID + rlwinm r4,r4,16,0x3fff0000 /* turn PID into MAS6[SPID] */ + mtspr SPRN_MAS6,r4 +1: lis r4,0x1000 /* Set MAS0(TLBSEL) = 1 */ + addi r3,r3,-1 + rlwimi r4,r3,16,4,15 /* Setup MAS0 = TLBSEL | ESEL(r3) */ + mtspr SPRN_MAS0,r4 + tlbre + mfspr r4,SPRN_MAS1 + andis. r4,r4,MAS1_VALID@h + bne 1b + + /* Get the tlb entry used by the current running code */ + bl 0f +0: mflr r4 + tlbsx 0,r4 + + mfspr r4,SPRN_MAS1 + ori r4,r4,MAS1_TS /* Set the TS = 1 */ + mtspr SPRN_MAS1,r4 + + mfspr r4,SPRN_MAS0 + rlwinm r4,r4,0,~MAS0_ESEL_MASK + rlwimi r4,r3,16,4,15 /* Setup MAS0 = TLBSEL | ESEL(r3) */ + mtspr SPRN_MAS0,r4 + tlbwe + isync + sync + + mfmsr r4 + ori r4,r4,MSR_IS | MSR_DS + mtspr SPRN_SRR0,r5 + mtspr SPRN_SRR1,r4 + sync + rfi + +/* + * Restore to the address space 0 and also invalidate the tlb entry created + * by switch_to_as1. +*/ +_GLOBAL(restore_to_as0) + mflr r0 + + bl 0f +0: mflr r9 + addi r9,r9,1f - 0b + + mfmsr r7 + li r8,(MSR_IS | MSR_DS) + andc r7,r7,r8 + + mtspr SPRN_SRR0,r9 + mtspr SPRN_SRR1,r7 + sync + rfi + + /* Invalidate the temporary tlb entry for AS1 */ +1: lis r9,0x1000 /* Set MAS0(TLBSEL) = 1 */ + rlwimi r9,r3,16,4,15 /* Setup MAS0 = TLBSEL | ESEL(r3) */ + mtspr SPRN_MAS0,r9 + tlbre + mfspr r9,SPRN_MAS1 + rlwinm r9,r9,0,2,31 /* Clear MAS1 Valid and IPPROT */ + mtspr SPRN_MAS1,r9 + tlbwe + isync + mtlr r0 + blr + +/* * We put a few things here that have to be page-aligned. This stuff * goes at the beginning of the data segment, which is page-aligned. */
arch/powerpc/mm/fsl_booke_mmu.c +2 -0
@@ -222,7 +222,9 @@ void __init adjust_total_lowmem(void)/* adjust lowmem size to __max_low_memory */ ram =min ((phys_addr_t )__max_low_memory, (phys_addr_t )total_lowmem); + i =switch_to_as1 (); __max_low_memory =map_mem_in_cams (ram,CONFIG_LOWMEM_CAM_NUM ); +restore_to_as0 (i);pr_info ("Memory CAM mapping: " );for (i =0 ; i < tlbcam_index -1 ; i++)
arch/powerpc/mm/mmu_decl.h +2 -0
@@ -148,6 +148,8 @@ extern unsigned long calc_cam_sz(unsigned long ram, unsigned long virt,extern void MMU_init_hw (void );extern unsigned long mmu_mapin_ram (unsigned long top);extern void adjust_total_lowmem (void ); +extern int switch_to_as1 (void ); +extern void restore_to_as0 (int esel);#endif extern void loadcam_entry (unsigned int index);
Signature · 1 line
1.8.3.1
PATCH v4 6/10 powerpc: introduce early_get_first_memblock_info
In reply to Kevin HaoFor a relocatable kernel since it can be loaded at any place, there is no any relation between the kernel start addr and the memstart_addr. So we can't calculate the memstart_addr from kernel start addr. And also we can't wait to do the relocation after we get the real memstart_addr from device tree because it is so late. So introduce a new function we can use to get the first memblock address and size in a very early stage (before machine_init).
---
v4: No change.
v3: Introduce a variable to avoid to mess the memblock.
v2: A new patch in v2.
arch/powerpc/kernel/prom.c | 41 ++++++++++++++++++++++++++++++++++++++++- include/linux/of_fdt.h | 1 + 2 files changed, 41 insertions(+), 1 deletion(-)
arch/powerpc/kernel/prom.c +40 -1
@@ -523,6 +523,20 @@ static int __init early_init_dt_scan_memory_ppc(unsigned long node,return early_init_dt_scan_memory (node, uname, depth, data); } +/* +* For a relocatable kernel, we need to get the memstart_addr first, +* then use it to calculate the virtual kernel start address. This has +* to happen at a very early stage (before machine_init). In this case, +* we just want to get the memstart_address and would not like to mess the +* memblock at this stage. So introduce a variable to skip the memblock_add() +* for this reason. +*/ +#ifdef CONFIG_RELOCATABLE +static int add_mem_to_memblock =1 ; +#else +#define add_mem_to_memblock 1 +#endif +void __init early_init_dt_add_memory_arch (u64 base,u64 size) {#ifdef CONFIG_PPC64
@@ -543,7 +557,8 @@ void __init early_init_dt_add_memory_arch(u64 base, u64 size) }/* Add the chunk to the MEMBLOCK list */ -memblock_add (base, size); +if (add_mem_to_memblock) +memblock_add (base, size); }static void __init early_reserve_mem_dt (void )
@@ -740,6 +755,30 @@ void __init early_init_devtree(void *params)DBG (" <- early_init_devtree()\n" ); } +#ifdef CONFIG_RELOCATABLE +/* +* This function run before early_init_devtree, so we have to init +* initial_boot_params. +*/ +void __init early_get_first_memblock_info (void *params,phys_addr_t *size) +{ +/* Setup flat device-tree pointer */ + initial_boot_params = params; + +/* +* Scan the memory nodes and set add_mem_to_memblock to 0 to avoid +* mess the memblock. +*/ + add_mem_to_memblock =0 ; +of_scan_flat_dt (early_init_dt_scan_root,NULL ); +of_scan_flat_dt (early_init_dt_scan_memory_ppc,NULL ); + add_mem_to_memblock =1 ; + +if (size) + *size = first_memblock_size; +} +#endif +/******* * * New implementation of the OF "find" APIs, return a refcounted
include/linux/of_fdt.h +1 -0
@@ -116,6 +116,7 @@ extern const void *of_flat_dt_match_machine(const void *default_match,extern void unflatten_device_tree (void );extern void unflatten_and_copy_device_tree (void );extern void early_init_devtree (void *); +extern void early_get_first_memblock_info (void *,phys_addr_t *);#else /* CONFIG_OF_FLATTREE */ static inline const char *of_flat_dt_get_machine_name (void ) {return NULL ; }static inline void unflatten_device_tree (void ) {}
Signature · 1 line
1.8.3.1
PATCH v4 7/10 powerpc/fsl_booke: introduce map_mem_in_cams_addr
In reply to Kevin HaoIntroduce this function so we can set both the physical and virtual address for the map in cams. This will be used by the relocation code.
---
v4: A new patch in v4.
arch/powerpc/mm/fsl_booke_mmu.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-)
arch/powerpc/mm/fsl_booke_mmu.c +10 -3
@@ -171,11 +171,10 @@ unsigned long calc_cam_sz(unsigned long ram, unsigned long virt,return 1UL << camsize; } -unsigned long map_mem_in_cams (unsigned long ram,int max_cam_idx) +static unsigned long map_mem_in_cams_addr (phys_addr_t phys,unsigned long virt, +unsigned long ram,int max_cam_idx) {int i; -unsigned long virt =PAGE_OFFSET ; -phys_addr_t phys = memstart_addr;unsigned long amount_mapped =0 ;/* Calculate CAM values */
@@ -195,6 +194,14 @@ unsigned long map_mem_in_cams(unsigned long ram, int max_cam_idx)return amount_mapped; } +unsigned long map_mem_in_cams (unsigned long ram,int max_cam_idx) +{ +unsigned long virt =PAGE_OFFSET ; +phys_addr_t phys = memstart_addr; + +return map_mem_in_cams_addr (phys, virt, ram, max_cam_idx); +} +#ifdef CONFIG_PPC32 #if defined(CONFIG_LOWMEM_CAM_NUM_BOOL ) && (CONFIG_LOWMEM_CAM_NUM >=NUM_TLBCAMS )
Signature · 1 line
1.8.3.1
PATCH v4 8/10 powerpc/fsl_booke: make sure PAGE_OFFSET map to memstart_addr for relocatable kernel
In reply to Kevin HaoThis is always true for a non-relocatable kernel. Otherwise the kernel would get stuck. But for a relocatable kernel, it seems a little complicated. When booting a relocatable kernel, we just align the kernel start addr to 64M and map the PAGE_OFFSET from there. The relocation will base on this virtual address. But if this address is not the same as the memstart_addr, we will have to change the map of PAGE_OFFSET to the real memstart_addr and do another relocation again.
---
v4:
* Create the correct mem map in cams when booting above 64M.
* Don't skip the init tlb mapping for the second cpu.
v3:
* Typo fix.
* Refactor relocate_init, no function change.
* Map only 64M memory before the second relocation.
* Comments update.
v2: A new patch in v2.
arch/powerpc/kernel/head_fsl_booke.S | 74 +++++++++++++++++++++++++++++++++--- arch/powerpc/mm/fsl_booke_mmu.c | 41 +++++++++++++++++--- arch/powerpc/mm/mmu_decl.h | 2 +- 3 files changed, 105 insertions(+), 12 deletions(-)
arch/powerpc/kernel/head_fsl_booke.S +69 -5
@@ -81,6 +81,39 @@ _ENTRY(_start); mr r23,r3 mr r25,r4 + bl 0f +0: mflr r8 + addis r3,r8,(is_second_reloc - 0b)@ha + lwz r19,(is_second_reloc - 0b)@l(r3) + + /* Check if this is the second relocation. */ + cmpwi r19,1 + bne 1f + + /* + * For the second relocation, we already get the real memstart_addr + * from device tree. So we will map PAGE_OFFSET to memstart_addr, + * then the virtual address of start kernel should be: + * PAGE_OFFSET + (kernstart_addr - memstart_addr) + * Since the offset between kernstart_addr and memstart_addr should + * never be beyond 1G, so we can just use the lower 32bit of them + * for the calculation. + */ + lis r3,PAGE_OFFSET@h + + addis r4,r8,(kernstart_addr - 0b)@ha + addi r4,r4,(kernstart_addr - 0b)@l + lwz r5,4(r4) + + addis r6,r8,(memstart_addr - 0b)@ha + addi r6,r6,(memstart_addr - 0b)@l + lwz r7,4(r6) + + subf r5,r7,r5 + add r3,r3,r5 + b 2f + +1: /* * We have the runtime (virutal) address of our base. * We calculate our shift of offset from a 64M page.
@@ -94,7 +127,14 @@ _ENTRY(_start); subf r3,r5,r6 /* r3 = r6 - r5 */ add r3,r4,r3 /* Required Virtual Address */ - bl relocate +2: bl relocate + + /* + * For the second relocation, we already set the right tlb entries + * for the kernel space, so skip the code in fsl_booke_entry_mapping.S + */ + cmpwi r19,1 + beq set_ivor #endif /* We try to not make any assumptions about how the boot loader
@@ -122,6 +162,7 @@ _ENTRY(__early_start) #include "fsl_booke_entry_mapping.S" #undef ENTRY_MAPPING_BOOT_SETUP +set_ivor: /* Establish the interrupt vector offsets */ SET_IVOR(0, CriticalInput); SET_IVOR(1, MachineCheck);
@@ -207,11 +248,13 @@ _ENTRY(__early_start) bl early_init #ifdef CONFIG_RELOCATABLE + mr r3,r30 + mr r4,r31 #ifdef CONFIG_PHYS_64BIT - mr r3,r23 - mr r4,r25 + mr r5,r23 + mr r6,r25 #else - mr r3,r25 + mr r5,r25 #endif bl relocate_init #endif
@@ -1207,6 +1250,9 @@ _GLOBAL(switch_to_as1) /* * Restore to the address space 0 and also invalidate the tlb entry created * by switch_to_as1. + * r3 - the tlb entry which should be invalidated + * r4 - __pa(PAGE_OFFSET in AS0) - __pa(PAGE_OFFSET in AS1) + * r5 - device tree virtual address. If r4 is 0, r5 is ignored. */ _GLOBAL(restore_to_as0) mflr r0
@@ -1215,7 +1261,15 @@ _GLOBAL(restore_to_as0) 0: mflr r9 addi r9,r9,1f - 0b - mfmsr r7 + /* + * We may map the PAGE_OFFSET in AS0 to a different physical address, + * so we need calculate the right jump and device tree address based + * on the offset passed by r4. + */ + subf r9,r4,r9 + subf r5,r4,r5 + +2: mfmsr r7 li r8,(MSR_IS | MSR_DS) andc r7,r7,r8
@@ -1234,9 +1288,19 @@ _GLOBAL(restore_to_as0) mtspr SPRN_MAS1,r9 tlbwe isync + + cmpwi r4,0 + bne 3f mtlr r0 blr + /* + * The PAGE_OFFSET will map to a different physical address, + * jump to _start to do another relocation again. + */ +3: mr r3,r5 + bl _start + /* * We put a few things here that have to be page-aligned. This stuff * goes at the beginning of the data segment, which is page-aligned.
arch/powerpc/mm/fsl_booke_mmu.c +35 -6
69 lines not on this page · raw message
arch/powerpc/mm/mmu_decl.h +1 -1
@@ -149,7 +149,7 @@ extern void MMU_init_hw(void);extern unsigned long mmu_mapin_ram (unsigned long top);extern void adjust_total_lowmem (void );extern int switch_to_as1 (void ); -extern void restore_to_as0 (int esel); +extern void restore_to_as0 (int esel,int offset,void *dt_ptr);#endif extern void loadcam_entry (unsigned int index);
Signature · 1 line
1.8.3.1
PATCH v4 9/10 powerpc/fsl_booke: smp support for booting a relocatable kernel above 64M
In reply to Kevin HaoWhen booting above the 64M for a secondary cpu, we also face the same issue as the boot cpu that the PAGE_OFFSET map two different physical address for the init tlb and the final map. So we have to use switch_to_as1/restore_to_as0 between the conversion of these two maps. When restoring to as0 for a secondary cpu, we only need to return to the caller. So add a new parameter for function restore_to_as0 for this purpose.
Use LOAD_REG_ADDR_PIC to get the address of variables which may be used before we set the final map in cams for the secondary cpu. Move the setting of cams a bit earlier in order to avoid the unnecessary using of LOAD_REG_ADDR_PIC.
---
v4: A new patch in v4.
arch/powerpc/kernel/head_fsl_booke.S | 41 ++++++++++++++++++++++++------------ arch/powerpc/mm/fsl_booke_mmu.c | 4 ++-- arch/powerpc/mm/mmu_decl.h | 2 +- arch/powerpc/mm/tlb_nohash_low.S | 4 +++- 4 files changed, 34 insertions(+), 17 deletions(-)
arch/powerpc/kernel/head_fsl_booke.S +28 -13
@@ -216,8 +216,7 @@ set_ivor: /* Check to see if we're the second processor, and jump * to the secondary_start code if so */ - lis r24, boot_cpuid@h - ori r24, r24, boot_cpuid@l + LOAD_REG_ADDR_PIC(r24, boot_cpuid) lwz r24, 0(r24) cmpwi r24, -1 mfspr r24,SPRN_PIR
@@ -1146,24 +1145,36 @@ _GLOBAL(__flush_disable_L1) /* When we get here, r24 needs to hold the CPU # */ .globl __secondary_start __secondary_start: - lis r3,__secondary_hold_acknowledge@h - ori r3,r3,__secondary_hold_acknowledge@l - stw r24,0(r3) - - li r3,0 - mr r4,r24 /* Why? */ - bl call_setup_cpu - - lis r3,tlbcam_index@ha - lwz r3,tlbcam_index@l(r3) + LOAD_REG_ADDR_PIC(r3, tlbcam_index) + lwz r3,0(r3) mtctr r3 li r26,0 /* r26 safe? */ + bl switch_to_as1 + mr r27,r3 /* tlb entry */ /* Load each CAM entry */ 1: mr r3,r26 bl loadcam_entry addi r26,r26,1 bdnz 1b + mr r3,r27 /* tlb entry */ + LOAD_REG_ADDR_PIC(r4, memstart_addr) + lwz r4,0(r4) + mr r5,r25 /* phys kernel start */ + rlwinm r5,r5,0,~0x3ffffff /* aligned 64M */ + subf r4,r5,r4 /* memstart_addr - phys kernel start */ + li r5,0 /* no device tree */ + li r6,0 /* not boot cpu */ + bl restore_to_as0 + + + lis r3,__secondary_hold_acknowledge@h + ori r3,r3,__secondary_hold_acknowledge@l + stw r24,0(r3) + + li r3,0 + mr r4,r24 /* Why? */ + bl call_setup_cpu /* get current_thread_info and current */ lis r1,secondary_ti@ha
@@ -1253,6 +1264,7 @@ _GLOBAL(switch_to_as1) * r3 - the tlb entry which should be invalidated * r4 - __pa(PAGE_OFFSET in AS0) - __pa(PAGE_OFFSET in AS1) * r5 - device tree virtual address. If r4 is 0, r5 is ignored. + * r6 - boot cpu */ _GLOBAL(restore_to_as0) mflr r0
@@ -1268,6 +1280,7 @@ _GLOBAL(restore_to_as0) */ subf r9,r4,r9 subf r5,r4,r5 + subf r0,r4,r0 2: mfmsr r7 li r8,(MSR_IS | MSR_DS)
@@ -1290,7 +1303,9 @@ _GLOBAL(restore_to_as0) isync cmpwi r4,0 - bne 3f + cmpwi cr1,r6,0 + cror eq,4*cr1+eq,eq + bne 3f /* offset != 0 && is_boot_cpu */ mtlr r0 blr
arch/powerpc/mm/fsl_booke_mmu.c +2 -2
@@ -231,7 +231,7 @@ void __init adjust_total_lowmem(void) i =switch_to_as1 (); __max_low_memory =map_mem_in_cams (ram,CONFIG_LOWMEM_CAM_NUM ); -restore_to_as0 (i,0 ,0 ); +restore_to_as0 (i,0 ,0 ,1 );pr_info ("Memory CAM mapping: " );for (i =0 ; i < tlbcam_index -1 ; i++)
@@ -301,7 +301,7 @@ notrace void __init relocate_init(u64 dt_ptr, phys_addr_t start)else map_mem_in_cams_addr (start,PAGE_OFFSET - offset,0x4000000 ,CONFIG_LOWMEM_CAM_NUM ); -restore_to_as0 (n, offset,__va (dt_ptr)); +restore_to_as0 (n, offset,__va (dt_ptr),1 );/* We should never reach here */ panic ("Relocation error" ); }
arch/powerpc/mm/mmu_decl.h +1 -1
@@ -149,7 +149,7 @@ extern void MMU_init_hw(void);extern unsigned long mmu_mapin_ram (unsigned long top);extern void adjust_total_lowmem (void );extern int switch_to_as1 (void ); -extern void restore_to_as0 (int esel,int offset,void *dt_ptr); +extern void restore_to_as0 (int esel,int offset,void *dt_ptr,int bootcpu);#endif extern void loadcam_entry (unsigned int index);
arch/powerpc/mm/tlb_nohash_low.S +3 -1
@@ -402,7 +402,9 @@ _GLOBAL(set_context) * Load TLBCAM[index] entry in to the L2 CAM MMU */ _GLOBAL(loadcam_entry) - LOAD_REG_ADDR(r4, TLBCAM) + mflr r5 + LOAD_REG_ADDR_PIC(r4, TLBCAM) + mtlr r5 mulli r5,r3,TLBCAM_SIZE add r3,r5,r4 lwz r4,TLBCAM_MAS0(r3)
Signature · 1 line
1.8.3.1
PATCH v4 10/10 powerpc/fsl_booke: enable the relocatable for the kdump kernel
In reply to Kevin HaoThe RELOCATABLE is more flexible and without any alignment restriction. And it is a superset of DYNAMIC_MEMSTART. So use it by default for a kdump kernel.
---
v4: No change.
v3: No change.
v2: A new patch in v2.
arch/powerpc/Kconfig | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
arch/powerpc/Kconfig +1 -2
@@ -399,8 +399,7 @@ config KEXEC config CRASH_DUMP bool "Build a kdump crash kernel" depends on PPC64 || 6xx || FSL_BOOKE || (44x && !SMP) - select RELOCATABLE if PPC64 || 44x - select DYNAMIC_MEMSTART if FSL_BOOKE + select RELOCATABLE if PPC64 || 44x || FSL_BOOKE help Build a kernel suitable for use as a kdump capture kernel. The same kernel binary can be used as production kernel and dump
Signature · 1 line
1.8.3.1