PPC 85xx: Add support for QEMU's ppce500 PV machine
v3 cover PPC 85xx: Add support for QEMU's ppce500 PV machine
In QEMU we implement a PV machine type called "ppce500". That board is able to run any e500+ FSL cores (e500v2, e500mc, e5500, e6500).
It is heavily inspired by the MPC8544DS SoC and board combination, but implements only the bare minimum to make Linux happy enough to drive a virtual machine.
This patch set implements support for this PV machine type in U-Boot, enabling users to run their virtual machines with netboot, u-boot payload binaries or other fun things they come up with.
---
v1 -> v2:
- Write device tree offset directly into global variable
- use r4 rather than r2 for that
- access fdt directly from in-memory copy
- remove unneeded header includes
- clean up pci enumeration
- coding style fixes
- populate and only use fdt_addr_r
- remove unused exported functions
- remove unused TLB0 entries
- make TLB1 I/O maps non-executable
- remove unused defines in board header
- make -kernel boot variables more clear
- remove TLB0 invalidation
- use tlb1.14 for temporary as=1 map
- use CONFIG_SYS_MPC85XX_NO_RESETVEC
- store fdt pointer in gd through cpu_init_early_f()
- replace fixup_tlb1() with dynamic TLB creation hook
- find CCSRBAR from device tree
- find PCI controllers from device tree
- find CPU speed from device tree
v2 -> v3:
- new patch: fdt_support: split fdt_getprop_u32_default
- new patch: fdt_support: Add helper function to read "ranges" property
- new patch: PPC: 85xx: Remove IVOR reset
- removed patch: PPC 85xx: Detect e500v2 / e500mc during runtime
- merge dt patches into machine patch
- Document and rename CONFIG_SYS_USE_DYNAMIC_TLBS
- remove CONFIG_DDR_SPD
- Reduce PCI message to only base address
- Map AS=1 fdt behind CCSRBAR
- Use CONFIG_SYS_CCSR_DO_NOT_RELOCATE
- remove unnecessary ccsr map
- s/cell size/cell count/
- add temporary map define
- use fdt_get_base_address() to find region
- use new APIs
- remove myfdt helpers
Alexander Graf (6):
fdt_support: split fdt_getprop_u32_default
fdt_support: Add helper function to read "ranges" property
PPC: 85xx: Remove IVOR reset
PPC: 85xx: Generalize DDR TLB mapping function PPC 85xx: Add ELF entry point
PPC 85xx: Add qemu-ppce500 machine
README | 5 + arch/powerpc/cpu/mpc85xx/Makefile | 2 + arch/powerpc/cpu/mpc85xx/cpu.c | 3 +- arch/powerpc/cpu/mpc85xx/cpu_init.c | 4 - arch/powerpc/cpu/mpc85xx/cpu_init_early.c | 10 +- arch/powerpc/cpu/mpc85xx/fixed_ivor.S | 63 ----- arch/powerpc/cpu/mpc85xx/release.S | 3 - arch/powerpc/cpu/mpc85xx/start.S | 19 +- arch/powerpc/cpu/mpc85xx/tlb.c | 52 ++-- arch/powerpc/cpu/mpc85xx/u-boot.lds | 1 + arch/powerpc/include/asm/config_mpc85xx.h | 4 + arch/powerpc/include/asm/mmu.h | 3 + board/freescale/qemu-ppce500/Makefile | 10 + board/freescale/qemu-ppce500/qemu-ppce500.c | 334 ++++++++++++++++++++++++++ board/freescale/qemu-ppce500/tlb.c | 16 ++ boards.cfg | 1 + common/fdt_support.c | 132 +++++++++- include/configs/qemu-ppce500.h | 206 ++++++++++++++++ include/fdt_support.h | 4 + nand_spl/board/freescale/mpc8536ds/Makefile | 4 - nand_spl/board/freescale/mpc8569mds/Makefile | 4 - nand_spl/board/freescale/mpc8572ds/Makefile | 4 - nand_spl/board/freescale/p1023rds/Makefile | 4 - nand_spl/board/freescale/p1_p2_rdb/Makefile | 4 - 24 files changed, 771 insertions(+), 121 deletions(-) delete mode 100644 arch/powerpc/cpu/mpc85xx/fixed_ivor.S create mode 100644 board/freescale/qemu-ppce500/Makefile create mode 100644 board/freescale/qemu-ppce500/qemu-ppce500.c create mode 100644 board/freescale/qemu-ppce500/tlb.c create mode 100644 include/configs/qemu-ppce500.h
Signature · 1 line
1.7.10.4
PATCH v3 1/6 fdt_support: split fdt_getprop_u32_default
We already have a nice helper to give us a property cell value with default fall back from a path. Split that into two helpers - one for the old path based lookup and one to give us a value based on a node offset.
--- common/fdt_support.c | 38 ++++++++++++++++++++++++++++++++------ include/fdt_support.h | 2 ++ 2 files changed, 34 insertions(+), 6 deletions(-)
common/fdt_support.c +32 -6
@@ -50,6 +50,37 @@ static void write_cell(u8 *addr, u64 val, int size) }/** +* fdt_getprop_u32_default_node - Return a node's property or a default +* +* @fdt: ptr to device tree +* @off: offset of node +* @cell: cell offset in property +* @prop: property name +* @dflt: default value if the property isn't found +* +* Convenience function to return a node's property or a default value if +* the property doesn't exist. +*/ +u32 fdt_getprop_u32_default_node (const void *fdt,int off,int cell, +const char *prop,const u32 dflt) +{ +const fdt32_t *val; +int len; + + val =fdt_getprop (fdt, off, prop, &len); + +/* Check if property exists */ +if (!val) +return dflt; + +/* Check if property is long enough */ +if (len < ((cell +1 ) *sizeof (uint32_t ))) +return dflt; + +return fdt32_to_cpu (*val); +} + +/** * fdt_getprop_u32_default - Find a node and return it's property or a default * * @fdt: ptr to device tree
@@ -63,18 +94,13 @@ static void write_cell(u8 *addr, u64 val, int size)u32 fdt_getprop_u32_default (const void *fdt,const char *path,const char *prop,const u32 dflt) { -const fdt32_t *val;int off; off =fdt_path_offset (fdt, path);if (off <0 )return dflt; - val =fdt_getprop (fdt, off, prop,NULL ); -if (val) -return fdt32_to_cpu (*val); -else -return dflt; +return fdt_getprop_u32_default_node (fdt, off,0 , prop, dflt); }/**
include/fdt_support.h +2 -0
@@ -12,6 +12,8 @@#include <libfdt.h> +u32 fdt_getprop_u32_default_node (const void *fdt,int off,int cell, +const char *prop,const u32 dflt);u32 fdt_getprop_u32_default (const void *fdt,const char *path,const char *prop,const u32 dflt);int fdt_chosen (void *fdt,int force);
Signature · 1 line
1.7.10.4
PATCH v3 2/6 fdt_support: Add helper function to read "ranges" property
In reply to Alexander GrafThis patch adds a helper function that can be used to interpret most "ranges" properties in the device tree.
It reads the n'th range out of a "ranges" array and returns the node's virtual address of the range, the physical address that range starts at and the size of the range.
--- common/fdt_support.c | 94 +++++++++++++++++++++++++++++++++++++++++++++++++ include/fdt_support.h | 2 ++ 2 files changed, 96 insertions(+)
common/fdt_support.c +94 -0
@@ -1435,3 +1435,97 @@ u64 fdt_get_base_address(void *fdt, int node)return prop ?fdt_translate_address (fdt, node, prop + naddr) :0 ; } + +/* +* Read a property of size <prop_len>. Currently only supports 1 or 2 cells. +*/ +static int fdt_read_prop (const fdt32_t *prop,int prop_len,int cell_off, +uint64_t *val,int cells) +{ +const fdt32_t *prop32 = &prop[cell_off]; +const fdt64_t *prop64 = (const fdt64_t *)&prop[cell_off]; + +if ((cell_off + cells) > prop_len) +return -FDT_ERR_NOSPACE ; + +switch (cells) { +case 1 : + *val =fdt32_to_cpu (*prop32); +break ; +case 2 : + *val =fdt64_to_cpu (*prop64); +break ; +default : +return -FDT_ERR_NOSPACE ; + } + +return 0 ; +} + +/** +* fdt_read_range - Read a node's n'th range property +* +* @fdt: ptr to device tree +* @node: offset of node +* @n: range index +* @child_addr: pointer to storage for the "child address" field +* @addr: pointer to storage for the CPU view translated physical start +* @len: pointer to storage for the range length +* +* Convenience function that reads and interprets a specific range out of +* a number of the "ranges" property array. +*/ +int fdt_read_range (void *fdt,int node,int n,uint64_t *child_addr, +uint64_t *addr,uint64_t *len) +{ +int pnode =fdt_parent_offset (fdt, node); +const fdt32_t *ranges; +int pacells; +int acells; +int scells; +int ranges_len; +int cell =0 ; +int r =0 ; + +/* +* The "ranges" property is an array of +* { <child address> <parent address> <size in child address space> } +* +* All 3 elements can span a diffent number of cells. Fetch their size. +*/ + pacells =fdt_getprop_u32_default_node (fdt, pnode,0 ,"#address-cells" ,1 ); + acells =fdt_getprop_u32_default_node (fdt, node,0 ,"#address-cells" ,1 ); + scells =fdt_getprop_u32_default_node (fdt, node,0 ,"#size-cells" ,1 ); + +/* Now try to get the ranges property */ + ranges =fdt_getprop (fdt, node,"ranges" , &ranges_len); +if (!ranges) +return -FDT_ERR_NOTFOUND ; + ranges_len /=sizeof (uint32_t ); + +/* Jump to the n'th entry */ + cell = n * (pacells + acells + scells); + +/* Read <child address> */ +if (child_addr) { + r =fdt_read_prop (ranges, ranges_len, cell, child_addr, + acells); +if (r) +return r; + } + cell += acells; + +/* Read <parent address> */ +if (addr) + *addr =fdt_translate_address (fdt, node, ranges + cell); + cell += pacells; + +/* Read <size in child address space> */ +if (len) { + r =fdt_read_prop (ranges, ranges_len, cell, len, scells); +if (r) +return r; + } + +return 0 ; +}
include/fdt_support.h +2 -0
@@ -87,6 +87,8 @@ int fdt_add_edid(void *blob, const char *compat, unsigned char *buf);int fdt_verify_alias_address (void *fdt,int anode,const char *alias,u64 addr);u64 fdt_get_base_address (void *fdt,int node); +int fdt_read_range (void *fdt,int node,int n,uint64_t *child_addr, +uint64_t *addr,uint64_t *len);enum fdt_status {FDT_STATUS_OKAY ,
Signature · 1 line
1.7.10.4
PATCH v3 3/6 PPC: 85xx: Remove IVOR reset
In reply to Alexander GrafThere is no need to set IVORs to anything but their default values, so let's leave them where they are.
--- arch/powerpc/cpu/mpc85xx/cpu_init.c | 4 -- arch/powerpc/cpu/mpc85xx/fixed_ivor.S | 63 -------------------------- arch/powerpc/cpu/mpc85xx/release.S | 3 -- arch/powerpc/cpu/mpc85xx/start.S | 6 --- nand_spl/board/freescale/mpc8536ds/Makefile | 4 -- nand_spl/board/freescale/mpc8569mds/Makefile | 4 -- nand_spl/board/freescale/mpc8572ds/Makefile | 4 -- nand_spl/board/freescale/p1023rds/Makefile | 4 -- nand_spl/board/freescale/p1_p2_rdb/Makefile | 4 -- 9 files changed, 96 deletions(-) delete mode 100644 arch/powerpc/cpu/mpc85xx/fixed_ivor.S
arch/powerpc/cpu/mpc85xx/cpu_init.c +0 -4
@@ -705,8 +705,6 @@ skip_l2:return 0 ; } -extern void setup_ivors (void ); -void arch_preboot_os (void ) {u32 msr;
@@ -719,8 +717,6 @@ void arch_preboot_os(void) msr =mfmsr (); msr &= ~(MSR_ME |MSR_CE );mtmsr (msr); - -setup_ivors (); }#if defined(CONFIG_CMD_SATA ) && defined(CONFIG_FSL_SATA )
arch/powerpc/cpu/mpc85xx/fixed_ivor.S deleted +0 -63
63 lines not on this page · raw message
arch/powerpc/cpu/mpc85xx/release.S +0 -3
@@ -407,9 +407,6 @@ __second_half_boot_page: bne 3b isync - /* setup IVORs to match fixed offsets */ -#include "fixed_ivor.S" - /* get the upper bits of the addr */ lwz r11,ENTRY_ADDR_UPPER(r10)
arch/powerpc/cpu/mpc85xx/start.S +0 -6
@@ -1967,10 +1967,4 @@ flush_dcache: isync blr - -.globl setup_ivors -setup_ivors: - -#include "fixed_ivor.S" - blr #endif /* !MINIMAL_SPL */
nand_spl/board/freescale/mpc8536ds/Makefile +0 -4
@@ -80,10 +80,6 @@ $(obj)resetvec.S: @rm -f $@ ln -s $(SRCTREE)/$(CPUDIR)/resetvec.S $@ -$(obj)fixed_ivor.S: - @rm -f $@ - ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/fixed_ivor.S $@ - $(obj)start.S: $(obj)fixed_ivor.S @rm -f $@ ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/start.S $@
nand_spl/board/freescale/mpc8569mds/Makefile +0 -4
@@ -80,10 +80,6 @@ $(obj)resetvec.S: @rm -f $@ ln -s $(SRCTREE)/$(CPUDIR)/resetvec.S $@ -$(obj)fixed_ivor.S: - @rm -f $@ - ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/fixed_ivor.S $@ - $(obj)start.S: $(obj)fixed_ivor.S @rm -f $@ ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/start.S $@
nand_spl/board/freescale/mpc8572ds/Makefile +0 -4
@@ -80,10 +80,6 @@ $(obj)resetvec.S: @rm -f $@ ln -s $(SRCTREE)/$(CPUDIR)/resetvec.S $@ -$(obj)fixed_ivor.S: - @rm -f $@ - ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/fixed_ivor.S $@ - $(obj)start.S: $(obj)fixed_ivor.S @rm -f $@ ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/start.S $@
nand_spl/board/freescale/p1023rds/Makefile +0 -4
@@ -75,10 +75,6 @@ $(obj)resetvec.S: @rm -f $@ ln -s $(SRCTREE)/$(CPUDIR)/resetvec.S $@ -$(obj)fixed_ivor.S: - @rm -f $@ - ln -sf $(SRCTREE)/$(CPUDIR)/fixed_ivor.S $@ - $(obj)start.S: $(obj)fixed_ivor.S @rm -f $@ ln -sf $(SRCTREE)/$(CPUDIR)/start.S $@
nand_spl/board/freescale/p1_p2_rdb/Makefile +0 -4
@@ -80,10 +80,6 @@ $(obj)resetvec.S: @rm -f $@ ln -s $(SRCTREE)/$(CPUDIR)/resetvec.S $@ -$(obj)fixed_ivor.S: - @rm -f $@ - ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/fixed_ivor.S $@ - $(obj)start.S: $(obj)fixed_ivor.S @rm -f $@ ln -sf $(SRCTREE)/arch/powerpc/cpu/mpc85xx/start.S $@
Signature · 1 line
1.7.10.4
PATCH v3 4/6 PPC: 85xx: Generalize DDR TLB mapping function
In reply to Alexander GrafThe DDR mapping function really is just a generic virtual -> physical mapping function. Generalize it so it can support any virtual starting offset and IO maps just the same.
--- arch/powerpc/cpu/mpc85xx/tlb.c | 48 ++++++++++++++++++++++++++-------------- arch/powerpc/include/asm/mmu.h | 3 +++ 2 files changed, 34 insertions(+), 17 deletions(-)
arch/powerpc/cpu/mpc85xx/tlb.c +31 -17
@@ -236,20 +236,26 @@ void init_addr_map(void) }#endif -unsigned int -setup_ddr_tlbs_phys (phys_addr_t p_addr,unsigned int memsize_in_meg) +void +tlb_map_range (ulong v_addr,phys_addr_t p_addr,uint64_t size,bool is_ram) {int i;unsigned int tlb_size; -unsigned int wimge =MAS2_M ; -unsigned int ram_tlb_address = (unsigned int )CONFIG_SYS_DDR_SDRAM_BASE ; +unsigned int wimge; +unsigned int perm;unsigned int max_cam, tsize_mask; -u64 size, memsize = (u64 )memsize_in_meg <<20 ; +if (is_ram) { + perm =MAS3_SX |MAS3_SW |MAS3_SR ; + wimge =MAS2_M ;#ifdef CONFIG_SYS_PPC_DDR_WIMGE - wimge =CONFIG_SYS_PPC_DDR_WIMGE ; + wimge =CONFIG_SYS_PPC_DDR_WIMGE ;#endif - size =min (memsize,CONFIG_MAX_MEM_MAPPED ); + }else { + perm =MAS3_SW |MAS3_SR ; + wimge =MAS2_I |MAS2_G ; + } +if ((mfspr (SPRN_MMUCFG ) &MMUCFG_MAVN ) ==MMUCFG_MAVN_V1 ) {/* Convert (4^max) kB to (2^max) bytes */ max_cam = ((mfspr (SPRN_TLB1CFG ) >>16 ) &0xf ) *2 +10 ;
@@ -261,11 +267,11 @@ setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) }for (i =0 ; size && i <8 ; i++) { -int ram_tlb_index =find_free_tlbcam (); +int tlb_index =find_free_tlbcam ();u32 camsize =__ilog2_u64 (size) & tsize_mask; -u32 align =__ilog2 (ram_tlb_address) & tsize_mask; +u32 align =__ilog2 (v_addr) & tsize_mask; -if (ram_tlb_index ==-1 ) +if (tlb_index ==-1 )break ;if (align ==-2 ) align = max_cam;
@@ -277,18 +283,26 @@ setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) tlb_size = camsize -10 ; -set_tlb (1 , ram_tlb_address, p_addr, -MAS3_SX |MAS3_SW |MAS3_SR , wimge, -0 , ram_tlb_index, tlb_size,1 ); +set_tlb (1 , v_addr, p_addr, perm, wimge, +0 , tlb_index, tlb_size,1 ); size -=1ULL << camsize; - memsize -=1ULL << camsize; - ram_tlb_address +=1UL << camsize; + v_addr +=1UL << camsize; p_addr +=1UL << camsize; } -if (memsize) -print_size (memsize," left unmapped\n" ); +if (size) +print_size (size," left unmapped\n" ); +} + +unsigned int +setup_ddr_tlbs_phys (phys_addr_t p_addr,unsigned int memsize_in_meg) +{ +unsigned int ram_tlb_address = (unsigned int )CONFIG_SYS_DDR_SDRAM_BASE ; +u64 memsize = (u64 )memsize_in_meg <<20 ; + + memsize =min (memsize,CONFIG_MAX_MEM_MAPPED ); +tlb_map_range (ram_tlb_address, p_addr, memsize, true);return memsize_in_meg; }
arch/powerpc/include/asm/mmu.h +3 -0
@@ -509,6 +509,9 @@ extern void print_tlbcam(void);extern unsigned int setup_ddr_tlbs (unsigned int memsize_in_meg);extern void clear_ddr_tlbs (unsigned int memsize_in_meg); +extern void tlb_map_range (ulong v_addr,phys_addr_t p_addr,uint64_t size, +bool is_ram); +extern void write_tlb (u32 _mas0,u32 _mas1,u32 _mas2,u32 _mas3,u32 _mas7);#define SET_TLB_ENTRY (_tlb, _epn, _rpn, _perms, _wimge, _ts, _esel, _sz, _iprot) \
Signature · 1 line
1.7.10.4
PATCH v3 4/6 PPC: 85xx: Generalize DDR TLB mapping function
arch/powerpc/cpu/mpc85xx/tlb.c setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) · line 296
Quoted 5 lines · Alexander Graf
arch/powerpc/cpu/mpc85xx/tlb.c +3
-2-if (memsize) -print_size (memsize," left unmapped\n" ); +if (size) +print_size (size," left unmapped\n" ); +}
The print_size should move to the caller, with some way to pass back the amout left unmapped. Non-RAM callers would treat a non-zero unmapped value as an error.
arch/powerpc/cpu/mpc85xx/tlb.c setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) · line 307
Quoted 10 lines
arch/powerpc/cpu/mpc85xx/tlb.c +8
-0+unsigned int +setup_ddr_tlbs_phys (phys_addr_t p_addr,unsigned int memsize_in_meg) +{ +unsigned int ram_tlb_address = (unsigned int )CONFIG_SYS_DDR_SDRAM_BASE ; +u64 memsize = (u64 )memsize_in_meg <<20 ; + + memsize =min (memsize,CONFIG_MAX_MEM_MAPPED ); +tlb_map_range (ram_tlb_address, p_addr, memsize, true);return memsize_in_meg; }
Here you seem to be hiding the message for RAM.
York, are you OK with just removing the message altogether, and having tlb_map_range return a normal error code if it can't map everything (with DDR size reduced in advance as above)?
arch/powerpc/include/asm/mmu.h extern void print_tlbcam(void); · line 513
Quoted 10 lines
arch/powerpc/include/asm/mmu.h +2
-0@@ -509,6 +509,9 @@ extern void print_tlbcam(void);extern unsigned int setup_ddr_tlbs (unsigned int memsize_in_meg);extern void clear_ddr_tlbs (unsigned int memsize_in_meg); +extern void tlb_map_range (ulong v_addr,phys_addr_t p_addr,uint64_t size, +bool is_ram);
bool arguments tend to be hard to read at call sites -- flags (or enum) with something like MAP_RAM/MAP_IO would be nicer (I'm not insisting, though).
-Scott
PATCH v3 4/6 PPC: 85xx: Generalize DDR TLB mapping function
Quoted 11 lines · Scott Wood
arch/powerpc/cpu/mpc85xx/tlb.c setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) · line 296
Quoted 5 lines · Alexander Graf
arch/powerpc/cpu/mpc85xx/tlb.c +3
-2-if (memsize) -print_size (memsize," left unmapped\n" ); +if (size) +print_size (size," left unmapped\n" ); +}The
print_sizeshould move to the caller, with some way to pass back the amout left unmapped. Non-RAM callers would treat a non-zero unmapped value as an error.
Quoted 12 lines
Quoted 10 lines
arch/powerpc/cpu/mpc85xx/tlb.c +8
-0+unsigned int +setup_ddr_tlbs_phys (phys_addr_t p_addr,unsigned int memsize_in_meg) +{ +unsigned int ram_tlb_address = (unsigned int )CONFIG_SYS_DDR_SDRAM_BASE ; +u64 memsize = (u64 )memsize_in_meg <<20 ; + + memsize =min (memsize,CONFIG_MAX_MEM_MAPPED ); +tlb_map_range (ram_tlb_address, p_addr, memsize, true);return memsize_in_meg; }Here you seem to be hiding the message for RAM.
It could still fail if we're running out of TLB entries, no?
Quoted 3 lines
York, are you OK with just removing the message altogether, and having
tlb_map_rangereturn a normal error code if it can't map everything (with DDR size reduced in advance as above)?
How about we just change the return value of tlb_map_range to uint64_t and return size? That way we can 1:1 move the print code out of the function into the RAM map code and IO callers can just call assert(r != 0).
Quoted 9 lines
Quoted 9 lines
arch/powerpc/include/asm/mmu.h +0
-0@@ -509,6 +509,9 @@ extern void print_tlbcam(void);extern unsigned int setup_ddr_tlbs (unsigned int memsize_in_meg);extern void clear_ddr_tlbs (unsigned int memsize_in_meg);
Quoted 6 lines
arch/powerpc/include/asm/mmu.h extern void print_tlbcam(void); · line 513
Quoted 2 lines
arch/powerpc/include/asm/mmu.h +2
-0+extern void tlb_map_range (ulong v_addr,phys_addr_t p_addr,uint64_t size, +bool is_ram);bool arguments tend to be hard to read at call sites -- flags (or enum) with something like MAP_RAM/MAP_IO would be nicer (I'm not insisting, though).
Good point.
Alex
PATCH v3 4/6 PPC: 85xx: Generalize DDR TLB mapping function
Quoted 26 lines · Alexander Graf
Quoted 11 lines · Scott Wood
arch/powerpc/cpu/mpc85xx/tlb.c setup_ddr_tlbs_phys(phys_addr_t p_addr, unsigned int memsize_in_meg) · line 296
Quoted 5 lines · Alexander Graf
arch/powerpc/cpu/mpc85xx/tlb.c +3
-2-if (memsize) -print_size (memsize," left unmapped\n" ); +if (size) +print_size (size," left unmapped\n" ); +}The
print_sizeshould move to the caller, with some way to pass back the amout left unmapped. Non-RAM callers would treat a non-zero unmapped value as an error.Quoted 12 lines
Quoted 10 lines
arch/powerpc/cpu/mpc85xx/tlb.c +8
-0+unsigned int +setup_ddr_tlbs_phys (phys_addr_t p_addr,unsigned int memsize_in_meg) +{ +unsigned int ram_tlb_address = (unsigned int )CONFIG_SYS_DDR_SDRAM_BASE ; +u64 memsize = (u64 )memsize_in_meg <<20 ; + + memsize =min (memsize,CONFIG_MAX_MEM_MAPPED ); +tlb_map_range (ram_tlb_address, p_addr, memsize, true);return memsize_in_meg; }Here you seem to be hiding the message for RAM.
Quoted 1 line
It could still fail if we're running out of TLB entries, no?
That's not the usual reason for that message to be printed.
Quoted 8 lines
Quoted 3 lines
York, are you OK with just removing the message altogether, and having
tlb_map_rangereturn a normal error code if it can't map everything (with DDR size reduced in advance as above)?How about we just change the return value of
tlb_map_rangetouint64_tand return size? That way we can 1:1 move the print code out of the function into the RAM map code and IO callers can just call assert(r != 0).
That's fine. I was just wondering if the message had value at all, given that it's expected if you have more than 2 GiB of RAM.
-Scott
PATCH v3 5/6 PPC 85xx: Add ELF entry point
In reply to Alexander GrafWe want to be able to directly execute the ELF binary without going through the u-boot.bin one.
To know where we have to start executing this ELF binary we have to tell the linker where our entry point is.
--- arch/powerpc/cpu/mpc85xx/u-boot.lds | 1 + 1 file changed, 1 insertion(+)
arch/powerpc/cpu/mpc85xx/u-boot.lds +1 -0
@@ -13,6 +13,7 @@ #endif OUTPUT_ARCH(powerpc) +ENTRY(_start_e500) PHDRS {
Signature · 1 line
1.7.10.4
PATCH v3 6/6 PPC 85xx: Add qemu-ppce500 machine
In reply to Alexander GrafFor KVM we have a special PV machine type called "ppce500". This machine is inspired by the MPC8544DS board, but implements a lot less features than that one.
It also provides more PCI slots and is supposed to be enumerated by device tree only.
This patch adds support for the generic ppce500 machine and tries to rely solely on device tree for device enumeration.
---
v1 -> v2:
- Write device tree offset directly into global variable
- use r4 rather than r2 for that
- access fdt directly from in-memory copy
- remove unneeded header includes
- clean up pci enumeration
- coding style fixes
- populate and only use fdt_addr_r
- remove unused exported functions
- remove unused TLB0 entries
- make TLB1 I/O maps non-executable
- remove unused defines in board header
- make -kernel boot variables more clear
- remove TLB0 invalidation
- use tlb1.14 for temporary as=1 map
- use CONFIG_SYS_MPC85XX_NO_RESETVEC
- store fdt pointer in gd through cpu_init_early_f()
- replace fixup_tlb1() with dynamic TLB creation hook
v2 -> v3:
- merge dt patches into this one
- Document and rename CONFIG_SYS_USE_DYNAMIC_TLBS
- remove CONFIG_DDR_SPD
- Reduce PCI message to only base address
- Map AS=1 fdt behind CCSRBAR
- Use CONFIG_SYS_CCSR_DO_NOT_RELOCATE
- remove unnecessary ccsr map
- s/cell size/cell count/
- add temporary map define
- use fdt_get_base_address() to find region
- use new APIs
- remove myfdt helpers
--- README | 5 + arch/powerpc/cpu/mpc85xx/Makefile | 2 + arch/powerpc/cpu/mpc85xx/cpu.c | 3 +- arch/powerpc/cpu/mpc85xx/cpu_init_early.c | 10 +- arch/powerpc/cpu/mpc85xx/start.S | 13 ++ arch/powerpc/cpu/mpc85xx/tlb.c | 4 + arch/powerpc/include/asm/config_mpc85xx.h | 4 + board/freescale/qemu-ppce500/Makefile | 10 + board/freescale/qemu-ppce500/qemu-ppce500.c | 334 +++++++++++++++++++++++++++ board/freescale/qemu-ppce500/tlb.c | 16 ++ boards.cfg | 1 + include/configs/qemu-ppce500.h | 206 +++++++++++++++++ 12 files changed, 606 insertions(+), 2 deletions(-) create mode 100644 board/freescale/qemu-ppce500/Makefile create mode 100644 board/freescale/qemu-ppce500/qemu-ppce500.c create mode 100644 board/freescale/qemu-ppce500/tlb.c create mode 100644 include/configs/qemu-ppce500.h
README +5 -0
@@ -379,6 +379,11 @@ The following options need to be configured: symbol should be set to the TLB1 entry to be used for this purpose. + CONFIG_SYS_USE_DYNAMIC_TLBS + + Create dynamic TLB entries in early CPU bringup. If you enable + this, make sure to implement void init_tlbs_dynamic(void). + CONFIG_SYS_FSL_ERRATUM_A004510 Enables a workaround for erratum A004510. If set,
arch/powerpc/cpu/mpc85xx/Makefile +2 -0
@@ -102,7 +102,9 @@ obj-y += cpu.o obj-y += cpu_init.o obj-y += cpu_init_early.o obj-y += interrupts.o +ifneq ($(CONFIG_QEMU_E500),y) obj-y += speed.o +endif obj-y += tlb.o obj-y += traps.o
arch/powerpc/cpu/mpc85xx/cpu.c +2 -1
@@ -338,7 +338,8 @@ void mpc85xx_reginfo(void) !defined(CONFIG_SYS_INIT_L2_ADDR)phys_size_t initdram (int board_type) { -#if defined(CONFIG_SPD_EEPROM ) || defined(CONFIG_DDR_SPD ) +#if defined(CONFIG_SPD_EEPROM ) || defined(CONFIG_DDR_SPD ) || \ + defined(CONFIG_QEMU_E500 )return fsl_ddr_sdram_size ();#else return (phys_size_t )CONFIG_SYS_SDRAM_SIZE *1024 *1024 ;
arch/powerpc/cpu/mpc85xx/cpu_init_early.c +9 -1
@@ -79,7 +79,7 @@ void setup_ifc(void)#endif /* We run cpu_init_early_f in AS = 1 */ -void cpu_init_early_f (void ) +void cpu_init_early_f (void *fdt) {u32 mas0, mas1, mas2, mas3, mas7;int i;
@@ -102,6 +102,14 @@ void cpu_init_early_f(void)for (i =0 ; i <sizeof (gd_t); i++) ((char *)gd)[i] =0 ; +#ifdef CONFIG_QEMU_E500 +/* +* CONFIG_SYS_CCSRBAR_PHYS below will use gd->fdt_blob, so we need to +* populate it before it accesses it. +*/ + gd->fdt_blob = fdt; +#endif + mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (13 ); mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); mas2 =FSL_BOOKE_MAS2 (CONFIG_SYS_CCSRBAR ,MAS2_I |MAS2_G );
arch/powerpc/cpu/mpc85xx/start.S +13 -0
@@ -80,6 +80,11 @@ _start_e500: li r1,MSR_DE mtmsr r1 +#ifdef CONFIG_QEMU_E500 + /* Save our ePAPR device tree pointer before we clobber it */ + mr r24, r3 +#endif + #ifdef CONFIG_SYS_FSL_ERRATUM_A004510 mfspr r3,SPRN_SVR rlwinm r3,r3,0,0xff
@@ -514,6 +519,7 @@ nexti: mflr r1 /* R1 = our PC */ * As a general rule, TLB0 is used for short-term TLBs, and TLB1 is used for * long-term TLBs, so we use TLB0 here. */ +#if !defined(CONFIG_DYNAMIC_CCSRBAR) #if (CONFIG_SYS_CCSRBAR_DEFAULT != CONFIG_SYS_CCSRBAR_PHYS) #if !defined(CONFIG_SYS_CCSRBAR_PHYS_HIGH) || !defined(CONFIG_SYS_CCSRBAR_PHYS_LOW)
@@ -698,6 +704,7 @@ delete_temp_tlbs: delete_tlb0_entry 1, CONFIG_SYS_CCSRBAR + 0x1000, MAS2_I|MAS2_G, r3 #endif /* #if (CONFIG_SYS_CCSRBAR_DEFAULT != CONFIG_SYS_CCSRBAR_PHYS) */ +#endif /* #if !defined(CONFIG_DYNAMIC_CCSRBAR) */ #if defined(CONFIG_SYS_FSL_QORIQ_CHASSIS2) && defined(CONFIG_E6500) create_ccsr_l2_tlb:
@@ -1144,6 +1151,12 @@ _start_cont: mr r1,r3 /* Transfer to SP(r1) */ GET_GOT + +#ifdef CONFIG_QEMU_E500 + /* Pass the ePAPR device tree pointer to cpu_init_early_f */ + mr r3, r24 +#endif + bl cpu_init_early_f /* switch back to AS = 0 */
arch/powerpc/cpu/mpc85xx/tlb.c +4 -0
@@ -36,6 +36,10 @@ void init_tlbs(void) tlb_table[i].mas7); } +#ifdef CONFIG_SYS_USE_DYNAMIC_TLBS +init_tlbs_dynamic (); +#endif +return ; }
arch/powerpc/include/asm/config_mpc85xx.h +4 -0
@@ -776,6 +776,10 @@ defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)#define CONFIG_SYS_CCSRBAR_DEFAULT 0xff700000#define CONFIG_SYS_FSL_ERRATUM_A005125 +#elif defined(CONFIG_QEMU_E500 ) +#define CONFIG_MAX_CPUS 1 +#define CONFIG_SYS_CCSRBAR_DEFAULT 0xe0000000 +#else #error Processor type not defined for this platform#endif
board/freescale/qemu-ppce500/Makefile new file +10 -0
@@ -0,0 +1,10 @@ +# +# Copyright 2007 Freescale Semiconductor, Inc. +# (C) Copyright 2001-2006 +# Wolfgang Denk, DENX Software Engineering, wd at denx.de. +# +# SPDX-License-Identifier: GPL-2.0+ +# + +obj-y += qemu-ppce500.o +obj-y += tlb.o
board/freescale/qemu-ppce500/qemu-ppce500.c new file +334 -0
334 lines not on this page · raw message
board/freescale/qemu-ppce500/tlb.c new file +16 -0
@@ -0,0 +1,16 @@ +/* +* Copyright 2008 Freescale Semiconductor, Inc. +* +* (C) Copyright 2000 +* Wolfgang Denk, DENX Software Engineering, wd at denx.de. +* +* SPDX-License-Identifier: GPL-2.0+ +*/ + +#include <common.h> +#include <asm/mmu.h> + +struct fsl_e_tlb_entry tlb_table[] = { +}; + +int num_tlb_entries =ARRAY_SIZE (tlb_table);
boards.cfg +1 -0
@@ -986,6 +986,7 @@ Active powerpc mpc85xx - freescale t2080qds Active powerpc mpc85xx - freescale t2080qds T2080QDS_SPIFLASH T2080QDS:PPC_T2080,RAMBOOT_PBL,SPIFLASH,SYS_TEXT_BASE=0xFFF80000 Active powerpc mpc85xx - freescale t2080qds T2080QDS_NAND T2080QDS:PPC_T2080,RAMBOOT_PBL,NAND,SYS_TEXT_BASE=0xFFF80000 Active powerpc mpc85xx - freescale t2080qds T2080QDS_SRIO_PCIE_BOOT T2080QDS:PPC_T2080,SRIO_PCIE_BOOT_SLAVE,SYS_TEXT_BASE=0xFFF80000 +Active powerpc mpc85xx - freescale qemu-ppce500 qemu-ppce500 - Alexander Graf <[email protected]> Active powerpc mpc85xx - gdsys p1022 controlcenterd_36BIT_SDCARD controlcenterd:36BIT,SDCARD Dirk Eibach <[email protected]> Active powerpc mpc85xx - gdsys p1022 controlcenterd_36BIT_SDCARD_DEVELOP controlcenterd:36BIT,SDCARD,DEVELOP Dirk Eibach <[email protected]> Active powerpc mpc85xx - gdsys p1022 controlcenterd_TRAILBLAZER controlcenterd:TRAILBLAZER,SPIFLASH Dirk Eibach <[email protected]>
include/configs/qemu-ppce500.h new file +206 -0
206 lines not on this page · raw message
Signature · 1 line
1.7.10.4
PATCH v3 6/6 PPC 85xx: Add qemu-ppce500 machine
arch/powerpc/cpu/mpc85xx/start.S _start_e500: · line 86
Quoted 12 lines · Alexander Graf
arch/powerpc/cpu/mpc85xx/start.S +4
-0@@ -80,6 +80,11 @@ _start_e500: li r1,MSR_DE mtmsr r1 +#ifdef CONFIG_QEMU_E500 + /* Save our ePAPR device tree pointer before we clobber it */ + mr r24, r3 +#endif
FWIW, it should be harmless to do this unconditionally (though in that case I'd insert "(if we have one)" in the comment.
arch/powerpc/cpu/mpc85xx/start.S nexti: mflr r1 /* R1 = our PC */ · line 523
Quoted 9 lines
arch/powerpc/cpu/mpc85xx/start.S +1
-0#ifdef CONFIG_SYS_FSL_ERRATUM_A004510 mfspr r3,SPRN_SVR rlwinm r3,r3,0,0xff@@ -514,6 +519,7 @@ nexti: mflr r1 /* R1 = our PC */ * As a general rule, TLB0 is used for short-term TLBs, and TLB1 is used for * long-term TLBs, so we use TLB0 here. */ +#if !defined(CONFIG_DYNAMIC_CCSRBAR) #if (CONFIG_SYS_CCSRBAR_DEFAULT != CONFIG_SYS_CCSRBAR_PHYS)
This shouldn't be necessary, if you haveCONFIG_SYS_CCSR_DO_NOT_RELOCATE.
arch/powerpc/cpu/mpc85xx/tlb.c void init_tlbs(void) · line 41
Quoted 11 lines
arch/powerpc/cpu/mpc85xx/tlb.c +3
-0@@ -36,6 +36,10 @@ void init_tlbs(void) tlb_table[i].mas7); } +#ifdef CONFIG_SYS_USE_DYNAMIC_TLBS +init_tlbs_dynamic (); +#endif
You could avoid the ifdef by moving a stub implementation into the header -- or possibly better, avoid the config symbol entirely by using a weak symbol.
Better still, make init_tlbs() weak. Then you don't need a config symbol, a new function name, or a dummy tlb_table[] -- you just redefine init_tlbs() in board code.
Quoted 22 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +16
-0@@ -0,0 +1,334 @@ +/* +* Copyright 2007,2009-2014 Freescale Semiconductor, Inc. +* +* SPDX-License-Identifier: GPL-2.0+ +*/ + +#include <common.h> +#include <command.h> +#include <pci.h> +#include <asm/processor.h> +#include <asm/mmu.h> +#include <asm/fsl_pci.h> +#include <asm/io.h> +#include <libfdt.h> +#include <fdt_support.h> +#include <netdev.h>
board/freescale/qemu-ppce500/qemu-ppce500.c line 26
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+#include <fdtdec.h> +#include <errno.h> +#include <malloc.h> + +DECLARE_GLOBAL_DATA_PTR ; + +static void *get_fdt (void ) +{ +return (void *)gd->fdt_blob; +}
Does this return virtual or physical?
Quoted 7 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +7
-0+ +uint64_t get_phys_ccsrbar_addr_early (void ) +{ +u32 mas0, mas1, mas2, mas3, mas7; +ulong fdt = (ulong )get_fdt (); +uint64_t r; +
board/freescale/qemu-ppce500/qemu-ppce500.c line 43
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+/* +* To be able to read the FDT we need to create a temporary TLB +* map for it. +*/ + + mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (10 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (fdt,0 ); + mas3 =FSL_BOOKE_MAS3 (fdt,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (fdt);
Don't use the physical address as a virtual address. Use a known-good virtual address. Using unknown physical addresses as virtual addresses is generally a bad idea (it's different for stuff whose physical address is hardcoded in U-Boot), but it's particularly bad here because you'll create an overlapping TLB entry if the fdt happens to live within your initial memory mapping.
OK, I see you use CONFIG_SYS_TMPVIRT for this elsewhere, but I guess forgot to use it here (why is fdt mapping code duplicated?).
board/freescale/qemu-ppce500/qemu-ppce500.c line 109
Quoted 9 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +9
-0+void pci_init_board (void ) +{ +struct pci_controller *pci_hoses; +void *fdt =get_fdt (); +int pci_node; +int pci_num =0 ; +int pci_count =0 ; +const char *compat ="fsl,mpc8540-pci" ; +ulong map_addr;
May want to look for arbitrary PCI host controllers (device_type would be simplest), in case the QEMU machine ever gets an upgrade to PCIe.
board/freescale/qemu-ppce500/qemu-ppce500.c line 114
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ +puts ("\n" ); + +/* Start MMIO and PIO range maps above RAM */ + map_addr =CONFIG_MAX_MEM_MAPPED ;
It'd be better to hardcode virtual addresses for this (as other boards do), and limit the size you map to the smaller of the hardcoded size or the device tree size.
Quoted 13 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +13
-0+/* Count and allocate PCI buses */ + pci_node =fdt_node_offset_by_compatible (fdt,-1 , compat); +while (pci_node != -FDT_ERR_NOTFOUND ) { + pci_node =fdt_node_offset_by_compatible (fdt, pci_node, compat); + pci_count++; + } + +if (pci_count) { + pci_hoses =malloc (sizeof (struct pci_controller ) * pci_count); + }else { +printf ("PCI: disabled\n\n" ); +return ; + }
board/freescale/qemu-ppce500/qemu-ppce500.c line 138
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+ +/* Spawn PCI buses based on device tree */ + pci_node =fdt_node_offset_by_compatible (fdt,-1 , compat); +while (pci_node != -FDT_ERR_NOTFOUND ) { +struct fsl_pci_info pci_info = { }; +const fdt32_t *reg; +int r; + + reg =fdt_getprop (fdt, pci_node,"reg" ,NULL ); + pci_info.regs =fdt_translate_address ((void *)fdt, pci_node, reg);
Unnecessary cast.
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+void init_tlbs_dynamic (void ) +{ +unsigned long fdt_phys = (unsigned long )get_fdt (); +unsigned long fdt_phys_tlb = fdt_phys & ~0xffffful ; +unsigned long fdt_virt_tlb =CONFIG_SYS_TMPVIRT ;
board/freescale/qemu-ppce500/qemu-ppce500.c line 256
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+unsigned long fdt_virt = fdt_virt_tlb + (fdt_phys &0xffffful ); +u32 mas0, mas1, mas2, mas3, mas7; +phys_size_t ram_size; + +/* +* Create a temporary AS=1 map for the fdt +* +* We use ESEL=0 here to overwrite the previous AS=0 map for ourselves +* which was only 4k big. This way we don't have to clear any other maps. +*/
I don't think it's generally safe to assume that this entry is ESEL 0 -- though I'm wondering if the current TLB code is assuming that (or at least, assuming that whatever entry is used gets overwritten by the TLB table).
board/freescale/qemu-ppce500/qemu-ppce500.c line 261
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (0 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (fdt_virt_tlb,0 ); + mas3 =FSL_BOOKE_MAS3 (fdt_phys_tlb,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (fdt_phys_tlb);
What if the fdt straddles a 1M boundary?
board/freescale/qemu-ppce500/qemu-ppce500.c line 267
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+write_tlb (mas0, mas1, mas2, mas3, mas7); + gd->fdt_blob = (void *)fdt_virt; + +/* Fetch RAM size from the fdt */ + ram_size =fixed_sdram ();
Why not just call get_linear_ram_size() directly?
board/freescale/qemu-ppce500/qemu-ppce500.c line 280
Quoted 12 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +12
-0+/* And remove our fdt map again */ + gd->fdt_blob = (void *)fdt_phys; +disable_tlb (0 ); + +/* Create a dynamic AS=0 CCSRBAR mapping */ + mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (13 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (CONFIG_SYS_CCSRBAR ,MAS2_I |MAS2_G ); + mas3 =FSL_BOOKE_MAS3 (CONFIG_SYS_CCSRBAR_PHYS ,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (CONFIG_SYS_CCSRBAR_PHYS ); + +write_tlb (mas0, mas1, mas2, mas3, mas7);
Why is this not done via tlb_map_range (after calling init_used_tlb_cams, of course)?
Quoted 20 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +20
-0+ +/* Create a RAM map that spans all accessible RAM */ +init_used_tlb_cams (); +setup_ddr_tlbs (ram_size >>20 ); +} + +void init_laws (void ) +{ +/* We don't emulate LAWs yet */ +} + +static uint32_t get_cpu_freq (void ) +{ +void *fdt =get_fdt (); +int cpus_node =fdt_path_offset (fdt,"/cpus" ); +int cpu_node =fdt_first_subnode (fdt, cpus_node); +const char *prop ="clock-frequency" ; +return fdt_getprop_u32_default_node (fdt, cpu_node,0 , prop,0 ); +} +
board/freescale/qemu-ppce500/qemu-ppce500.c line 310
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+void get_sys_info (sys_info_t *sys_info) +{ +int freq =get_cpu_freq (); + +memset (sys_info,0 ,sizeof (sys_info_t)); + sys_info->freq_systembus = freq; + sys_info->freq_ddrbus = freq; + sys_info->freq_processor[0 ] = freq; +} +
Again, if you're doing this you really should override get_tbclk() instead of letting it use the much faster cpufreq/8.
board/freescale/qemu-ppce500/qemu-ppce500.c line 320
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+int get_clocks (void ) +{ +sys_info_t sys_info; + +get_sys_info (&sys_info); + + gd->cpu_clk = sys_info.freq_processor[0 ]; + gd->bus_clk = sys_info.freq_systembus; + gd->mem_clk = sys_info.freq_ddrbus; + gd->arch.lbc_clk = sys_info.freq_ddrbus;
I wonder if with higher frequencies we'll trigger overflows in some drivers. :-)
board/freescale/qemu-ppce500/qemu-ppce500.c line 161
Quoted 4 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +4
-0+/* Physical address should be a function call */ +#ifndef __ASSEMBLY__ +extern unsigned long long get_phys_ccsrbar_addr_early (void ); +#endif
Where does this get called?
-Scott
PATCH v3 6/6 PPC 85xx: Add qemu-ppce500 machine
Quoted 9 lines · Scott Wood
Quoted 8 lines · Alexander Graf
arch/powerpc/cpu/mpc85xx/start.S +0
-0@@ -80,6 +80,11 @@ _start_e500: li r1,MSR_DE mtmsr r1
Quoted 7 lines
arch/powerpc/cpu/mpc85xx/start.S _start_e500: · line 86
Quoted 4 lines
arch/powerpc/cpu/mpc85xx/start.S +4
-0+#ifdef CONFIG_QEMU_E500 + /* Save our ePAPR device tree pointer before we clobber it */ + mr r24, r3 +#endifFWIW, it should be harmless to do this unconditionally (though in that case I'd insert "(if we have one)" in the comment.
Sure.
Quoted 13 lines
arch/powerpc/cpu/mpc85xx/start.S nexti: mflr r1 /* R1 = our PC */ · line 523
Quoted 10 lines
arch/powerpc/cpu/mpc85xx/start.S +1
-0#ifdef CONFIG_SYS_FSL_ERRATUM_A004510 mfspr r3,SPRN_SVR rlwinm r3,r3,0,0xff@@ -514,6 +519,7 @@ nexti: mflr r1 /* R1 = our PC */ * As a general rule, TLB0 is used for short-term TLBs, and TLB1 is used for * long-term TLBs, so we use TLB0 here. */ +#if !defined(CONFIG_DYNAMIC_CCSRBAR) #if (CONFIG_SYS_CCSRBAR_DEFAULT != CONFIG_SYS_CCSRBAR_PHYS)This shouldn't be necessary, if you have
CONFIG_SYS_CCSR_DO_NOT_RELOCATE.
You're right :).
Quoted 17 lines
arch/powerpc/cpu/mpc85xx/tlb.c void init_tlbs(void) · line 41
Quoted 12 lines
arch/powerpc/cpu/mpc85xx/tlb.c +3
-0@@ -36,6 +36,10 @@ void init_tlbs(void) tlb_table[i].mas7); } +#ifdef CONFIG_SYS_USE_DYNAMIC_TLBS +init_tlbs_dynamic (); +#endif You could avoid the ifdef by moving a stub implementation into the header -- or possibly better, avoid the config symbol entirely by using a weak symbol.
Quoted 3 lines
Better still, make
init_tlbs()weak. Then you don't need a config symbol, a new function name, or a dummy tlb_table[] -- you just redefineinit_tlbs()in board code.
Nice idea. That radically simplifies the code.
Quoted 27 lines
Quoted 27 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +20
-0@@ -0,0 +1,334 @@ +/* +* Copyright 2007,2009-2014 Freescale Semiconductor, Inc. +* +* SPDX-License-Identifier: GPL-2.0+ +*/ + +#include <common.h> +#include <command.h> +#include <pci.h> +#include <asm/processor.h> +#include <asm/mmu.h> +#include <asm/fsl_pci.h> +#include <asm/io.h> +#include <libfdt.h> +#include <fdt_support.h> +#include <netdev.h> +#include <fdtdec.h> +#include <errno.h> +#include <malloc.h> +
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 26
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +6
-0+DECLARE_GLOBAL_DATA_PTR ; + +static void *get_fdt (void ) +{ +return (void *)gd->fdt_blob; +}Does this return virtual or physical?
I've split this into a get_fdt_virt() and get_fdt_phys() function to make this more explicit.
Quoted 17 lines
Quoted 17 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +16
-0+ +uint64_t get_phys_ccsrbar_addr_early (void ) +{ +u32 mas0, mas1, mas2, mas3, mas7; +ulong fdt = (ulong )get_fdt (); +uint64_t r; + +/* +* To be able to read the FDT we need to create a temporary TLB +* map for it. +*/ + + mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (10 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (fdt,0 ); + mas3 =FSL_BOOKE_MAS3 (fdt,0 ,MAS3_SW |MAS3_SR );
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 43
Quoted 1 line
board/freescale/qemu-ppce500/qemu-ppce500.c new file +1
-0+ mas7 =FSL_BOOKE_MAS7 (fdt);Don't use the physical address as a virtual address. Use a known-good virtual address. Using unknown physical addresses as virtual addresses is generally a bad idea (it's different for stuff whose physical address is hardcoded in U-Boot), but it's particularly bad here because you'll create an overlapping TLB entry if the fdt happens to live within your initial memory mapping.
Ok.
Quoted 3 lines
OK, I see you use
CONFIG_SYS_TMPVIRTfor this elsewhere, but I guess forgot to use it here (why is fdt mapping code duplicated?).
I've folded the AS=1 mapping functions into a single function and used tlb_map_range for the AS=0 one.
Quoted 13 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 109
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +9
-0+void pci_init_board (void ) +{ +struct pci_controller *pci_hoses; +void *fdt =get_fdt (); +int pci_node; +int pci_num =0 ; +int pci_count =0 ; +const char *compat ="fsl,mpc8540-pci" ; +ulong map_addr;May want to look for arbitrary PCI host controllers (
device_typewould be simplest), in case the QEMU machine ever gets an upgrade to PCIe.
Ok.
Quoted 10 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 114
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ +puts ("\n" ); + +/* Start MMIO and PIO range maps above RAM */ + map_addr =CONFIG_MAX_MEM_MAPPED ;It'd be better to hardcode virtual addresses for this (as other boards do), and limit the size you map to the smaller of the hardcoded size or the device tree size.
I don't understand this comment. CONFIG_MAX_MEM_MAPPED is basically the first address available to IO maps, so with this it is properly hardcoded and ensured to always map IO to the same physical address regardless of memory passed in.
Quoted 18 lines
Quoted 18 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +17
-0+/* Count and allocate PCI buses */ + pci_node =fdt_node_offset_by_compatible (fdt,-1 , compat); +while (pci_node != -FDT_ERR_NOTFOUND ) { + pci_node =fdt_node_offset_by_compatible (fdt, pci_node, compat); + pci_count++; + } + +if (pci_count) { + pci_hoses =malloc (sizeof (struct pci_controller ) * pci_count); + }else { +printf ("PCI: disabled\n\n" ); +return ; + } + +/* Spawn PCI buses based on device tree */ + pci_node =fdt_node_offset_by_compatible (fdt,-1 , compat); +while (pci_node != -FDT_ERR_NOTFOUND ) {
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 138
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +6
-0+struct fsl_pci_info pci_info = { }; +const fdt32_t *reg; +int r; + + reg =fdt_getprop (fdt, pci_node,"reg" ,NULL ); + pci_info.regs =fdt_translate_address ((void *)fdt, pci_node, reg);Unnecessary cast.
Removed.
Quoted 13 lines
Quoted 13 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +12
-0+void init_tlbs_dynamic (void ) +{ +unsigned long fdt_phys = (unsigned long )get_fdt (); +unsigned long fdt_phys_tlb = fdt_phys & ~0xffffful ; +unsigned long fdt_virt_tlb =CONFIG_SYS_TMPVIRT ; +unsigned long fdt_virt = fdt_virt_tlb + (fdt_phys &0xffffful ); +u32 mas0, mas1, mas2, mas3, mas7; +phys_size_t ram_size; + +/* +* Create a temporary AS=1 map for the fdt +*
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 256
Quoted 3 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +3
-0+* We use ESEL=0 here to overwrite the previous AS=0 map for ourselves +* which was only 4k big. This way we don't have to clear any other maps. +*/ I don't think it's generally safe to assume that this entry is ESEL 0 -- though I'm wondering if the current TLB code is assuming that (or at least, assuming that whatever entry is used gets overwritten by the TLB table).
That's at least the way I understand the current code :). There's definitely no code anywhere that removes the 4k map:
0x0000000000f00000 0x0000000000f00000 4K 0 0 SRWXURWX ----- U----
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 261
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (0 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (fdt_virt_tlb,0 ); + mas3 =FSL_BOOKE_MAS3 (fdt_phys_tlb,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (fdt_phys_tlb);What if the fdt straddles a 1M boundary?
Then we fix the hypervisor ;). Even the 1MB is only an approximation. We don't know the size of the fdt. But I think we can expect the hypervisor to align it on 1MB. The masks here really are just to be nice to a hypervisor if it's broken (or knows exactly what it's doing).
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 267
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+write_tlb (mas0, mas1, mas2, mas3, mas7); + gd->fdt_blob = (void *)fdt_virt; + +/* Fetch RAM size from the fdt */ + ram_size =fixed_sdram ();Why not just call
get_linear_ram_size()directly?
*shrug*. Fixed :).
Quoted 16 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 280
Quoted 13 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +12
-0+/* And remove our fdt map again */ + gd->fdt_blob = (void *)fdt_phys; +disable_tlb (0 ); + +/* Create a dynamic AS=0 CCSRBAR mapping */ + mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (13 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (CONFIG_SYS_CCSRBAR ,MAS2_I |MAS2_G ); + mas3 =FSL_BOOKE_MAS3 (CONFIG_SYS_CCSRBAR_PHYS ,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (CONFIG_SYS_CCSRBAR_PHYS ); + +write_tlb (mas0, mas1, mas2, mas3, mas7);Why is this not done via
tlb_map_range(after callinginit_used_tlb_cams, of course)?
Only because this code predates the generic tlb_map_range function. I've fixed it.
Quoted 26 lines
Quoted 26 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +25
-0+ +/* Create a RAM map that spans all accessible RAM */ +init_used_tlb_cams (); +setup_ddr_tlbs (ram_size >>20 ); +} + +void init_laws (void ) +{ +/* We don't emulate LAWs yet */ +} + +static uint32_t get_cpu_freq (void ) +{ +void *fdt =get_fdt (); +int cpus_node =fdt_path_offset (fdt,"/cpus" ); +int cpu_node =fdt_first_subnode (fdt, cpus_node); +const char *prop ="clock-frequency" ; +return fdt_getprop_u32_default_node (fdt, cpu_node,0 , prop,0 ); +} + +void get_sys_info (sys_info_t *sys_info) +{ +int freq =get_cpu_freq (); + +memset (sys_info,0 ,sizeof (sys_info_t));
Quoted 8 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 310
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ sys_info->freq_systembus = freq; + sys_info->freq_ddrbus = freq; + sys_info->freq_processor[0 ] = freq; +} +Again, if you're doing this you really should override
get_tbclk()instead of letting it use the much faster cpufreq/8.
Ok, I've made get_tbclk weak and override it with a function that reads timebase-frequency from the first cpu node in dt.
Quoted 15 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 320
Quoted 11 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +10
-0+int get_clocks (void ) +{ +sys_info_t sys_info; + +get_sys_info (&sys_info); + + gd->cpu_clk = sys_info.freq_processor[0 ]; + gd->bus_clk = sys_info.freq_systembus; + gd->mem_clk = sys_info.freq_ddrbus; + gd->arch.lbc_clk = sys_info.freq_ddrbus;I wonder if with higher frequencies we'll trigger overflows in some drivers. :-)
Quoted 6 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 161
Quoted 4 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +4
-0+/* Physical address should be a function call */ +#ifndef __ASSEMBLY__ +extern unsigned long long get_phys_ccsrbar_addr_early (void ); +#endif Where does this get called?
CONFIG_SYS_CCSRBAR_PHYS is defined to this which gets used in cpu_init_early_f().
Alex
PATCH v3 6/6 PPC 85xx: Add qemu-ppce500 machine
Quoted 12 lines · Alexander Graf
Quoted 9 lines · Scott Wood
board/freescale/qemu-ppce500/qemu-ppce500.c line 114
Quoted 4 lines · Alexander Graf
board/freescale/qemu-ppce500/qemu-ppce500.c new file +4
-0+puts ("\n" ); + +/* Start MMIO and PIO range maps above RAM */ + map_addr =CONFIG_MAX_MEM_MAPPED ;It'd be better to hardcode virtual addresses for this (as other boards do), and limit the size you map to the smaller of the hardcoded size or the device tree size.
Quoted 4 lines
I don't understand this comment.
CONFIG_MAX_MEM_MAPPEDis basically the first address available to IO maps, so with this it is properly hardcoded and ensured to always map IO to the same physical address regardless of memory passed in.
I mean an explicit address in the board config file, rather than hiding it here. It helps to have the full address map in one place. Consider what would happen if some other part of the code tried the same trick with CONFIG_MAX_MEM_MAPPED. :-)
Quoted 8 lines
Quoted 7 lines
board/freescale/qemu-ppce500/qemu-ppce500.c line 261
Quoted 5 lines
board/freescale/qemu-ppce500/qemu-ppce500.c new file +5
-0+ mas0 =MAS0_TLBSEL (1 ) |MAS0_ESEL (0 ); + mas1 =MAS1_VALID |MAS1_TID (0 ) |MAS1_TS |MAS1_TSIZE (BOOKE_PAGESZ_1M ); + mas2 =FSL_BOOKE_MAS2 (fdt_virt_tlb,0 ); + mas3 =FSL_BOOKE_MAS3 (fdt_phys_tlb,0 ,MAS3_SW |MAS3_SR ); + mas7 =FSL_BOOKE_MAS7 (fdt_phys_tlb);What if the fdt straddles a 1M boundary?
Quoted 4 lines
Then we fix the hypervisor ;). Even the 1MB is only an approximation. We don't know the size of the fdt. But I think we can expect the hypervisor to align it on 1MB. The masks here really are just to be nice to a hypervisor if it's broken (or knows exactly what it's doing).
What's special about 1 MiB? If you want to rely on DTC_PAD_MASK not changing to simplify this code, since it's QEMU-specific, fine -- but I wouldn't consider it "broken" for an arbitrary hypervisor to do differently.
-Scott