Jonathan Neuschäfer has uploaded this change for review. ( https://review.coreboot.org/23772
Change subject: arch/riscv: Don't set up virtual memory ......................................................................
arch/riscv: Don't set up virtual memory
Due to changes in the RISC-V Privileged Architecture specification, Linux can now be started in physical memory and it will setup its own page tables.
Thus we can delete most of virtual_memory.c.
Change-Id: I4e69d15f8ee540d2f98c342bc4ec0c00fb48def0 Signed-off-by: Jonathan Neuschäfer j.neuschaefer@gmx.net --- M src/Kconfig M src/arch/riscv/boot.c M src/arch/riscv/include/vm.h M src/arch/riscv/virtual_memory.c 4 files changed, 0 insertions(+), 266 deletions(-)
git pull ssh://review.coreboot.org:29418/coreboot refs/changes/72/23772/1
diff --git a/src/Kconfig b/src/Kconfig index 459280d..02abed2 100644 --- a/src/Kconfig +++ b/src/Kconfig @@ -975,14 +975,6 @@ Control debugging of the boot state machine. When selected displays the state boundaries in ramstage.
-config DEBUG_PRINT_PAGE_TABLES - bool "Print the page tables after construction" - default n - depends on ARCH_RISCV - help - After the page tables have been built, print them on the debug - console. - config DEBUG_ADA_CODE bool "Compile debug code in Ada sources" default n diff --git a/src/arch/riscv/boot.c b/src/arch/riscv/boot.c index ac095bb..b73f3ca 100644 --- a/src/arch/riscv/boot.c +++ b/src/arch/riscv/boot.c @@ -27,7 +27,6 @@
if (ENV_RAMSTAGE && prog_type(prog) == PROG_PAYLOAD) { printk(BIOS_SPEW, "Config string: '%s'\n", config); - initVirtualMemory(); printk(BIOS_SPEW, "OK, let's go\n"); riscvpayload(config, doit); } diff --git a/src/arch/riscv/include/vm.h b/src/arch/riscv/include/vm.h index 1a8f7ad..fd0a2c9 100644 --- a/src/arch/riscv/include/vm.h +++ b/src/arch/riscv/include/vm.h @@ -32,36 +32,11 @@ #include <stdint.h> #include <arch/encoding.h>
-#define SUPERPAGE_SIZE ((uintptr_t)(RISCV_PGSIZE << RISCV_PGLEVEL_BITS)) -#define VM_CHOICE VM_SV39 -#define VA_BITS 39 -#define MEGAPAGE_SIZE (SUPERPAGE_SIZE << RISCV_PGLEVEL_BITS) - #define EXTRACT_FIELD(val, which) (((val) & (which)) / ((which) & ~((which)-1))) #define INSERT_FIELD(val, which, fieldval) (((val) & ~(which)) | ((fieldval) * ((which) & ~((which)-1))))
-#define supervisor_paddr_valid(start, length) \ - ((uintptr_t)(start) >= current.first_user_vaddr + current.bias \ - && (uintptr_t)(start) + (length) < mem_size \ - && (uintptr_t)(start) + (length) >= (uintptr_t)(start)) - -typedef uintptr_t pte_t; -extern pte_t* root_page_table; - -void initVirtualMemory(void); - -size_t pte_ppn(pte_t pte); -pte_t ptd_create(uintptr_t ppn); -pte_t pte_create(uintptr_t ppn, int prot, int user); - -void print_page_table(void); - -void init_vm(uintptr_t virtMemStart, uintptr_t physMemStart, - pte_t *pageTableStart); void mstatus_init(void); // need to setup mstatus so we know we have virtual memory
-void flush_tlb(void); -
#define DEFINE_MPRV_READ(name, type, insn) \ static inline type name(type *p); \ diff --git a/src/arch/riscv/virtual_memory.c b/src/arch/riscv/virtual_memory.c index 27b4b82..702d6b7 100644 --- a/src/arch/riscv/virtual_memory.c +++ b/src/arch/riscv/virtual_memory.c @@ -14,13 +14,9 @@ * GNU General Public License for more details. */
-#include <arch/barrier.h> #include <arch/encoding.h> -#include <atomic.h> -#include <console/console.h> #include <stdint.h> #include <vm.h> -#include <symbols.h>
/* Delegate controls which traps are delegated to the payload. If you * wish to temporarily disable some or all delegation you can, in a @@ -38,234 +34,6 @@ | (1 << CAUSE_USER_ECALL) ;
-pte_t* root_page_table; - -/* Indent the following text by 2*level spaces */ -static void indent(int level) -{ - int i; - for (i = 0; i < level; i++) - printk(BIOS_DEBUG, " "); -} - -/* - * Convert a page table index at a given page table level to a virtual address - * offset - */ -static uintptr_t index_to_virt_addr(int index, int level) -{ - /* - * Index is at most RISCV_PGLEVEL_BITS bits wide (not considering the - * leading zeroes. If level==0, the below expression thus shifts index - * into the highest bits of a 64-bit number, and then shifts it down - * with sign extension. - * - * If level>0, then the expression should work as expected, without any - * magic. - */ - return ((intptr_t)index) - << (64 - RISCV_PGLEVEL_BITS - level * RISCV_PGLEVEL_BITS) - >> (64 - VA_BITS); -} - -/* Dump the page table structures to the console -- helper function */ -static void print_page_table_at(pte_t *pt, intptr_t virt_addr, int level) -{ - int i; - - indent(level); - printk(BIOS_DEBUG, "Level %d page table at 0x%p\n", level, pt); - - for (i = 0; i < RISCV_PGSIZE / sizeof(pte_t); i++) { - char urwx[8]; - uintptr_t pointer; - intptr_t next_virt_addr; - - if (!(pt[i] & PTE_V)) - continue; - - urwx[0] = (pt[i] & PTE_U)? 'u' : '-'; - urwx[1] = (pt[i] & PTE_R)? 'r' : '-'; - urwx[2] = (pt[i] & PTE_W)? 'w' : '-'; - urwx[3] = (pt[i] & PTE_X)? 'x' : '-'; - urwx[4] = '\0'; - - next_virt_addr = virt_addr + index_to_virt_addr(i, level); - - pointer = ((uintptr_t)pt[i] >> 10) << RISCV_PGSHIFT; - - indent(level + 1); - printk(BIOS_DEBUG, "Valid PTE at index %d (0x%016zx -> 0x%zx), ", - i, (size_t) next_virt_addr, (size_t) pointer); - if (PTE_TABLE(pt[i])) - printk(BIOS_DEBUG, "page table\n"); - else - printk(BIOS_DEBUG, "protections %s\n", urwx); - - if (PTE_TABLE(pt[i])) { - print_page_table_at((pte_t *)pointer, next_virt_addr, level + 1); - } - } -} - -/* Print the page table structures to the console */ -void print_page_table(void) { - print_page_table_at((void *)(read_csr(sptbr) << RISCV_PGSHIFT), 0, 0); -} - -void flush_tlb(void) -{ - asm volatile("sfence.vm"); -} - -size_t pte_ppn(pte_t pte) -{ - return pte >> PTE_PPN_SHIFT; -} - -pte_t ptd_create(uintptr_t ppn) -{ - return (ppn << PTE_PPN_SHIFT) | PTE_V; -} - -pte_t pte_create(uintptr_t ppn, int prot, int user) -{ - pte_t pte = (ppn << PTE_PPN_SHIFT) | PTE_R | PTE_V; - if (prot & PTE_W) - pte |= PTE_W; - if (prot & PTE_X) - pte |= PTE_X; - if (user) - pte |= PTE_U; - return pte; -} - -// The current RISCV *physical* address space is this: -// * 0 - 2 GiB: miscellaneous IO devices -// * 2 GiB - 4 GiB DRAM -// We have determined, also, that if code references a physical address -// not backed by a device, we'll take a fault. In other words, we don't -// need to finely map the memory-mapped devices as we would on an x86. -// We can use GiB mappings for the IO space and we will take a trap -// if we reference hardware that does not exist. -// -// The intent of the RISCV designers is that pages be set up in M mode -// for lower privilege software. They have also told me that they -// expect, unlike other platforms, that next level software use these -// page tables. Some kernels (Linux) prefer the old fashioned model, -// where kernel starts with an identity (ID) map and sets up page tables as -// it sees fit. Other kernels (harvey) are fine with using whatever -// firmware sets up. We need to accommodate both. So, we set up the -// identity map for Linux, but also set up the map for kernels that -// are more willing to conform to the RISCV model. The map is as -// follows: -// -// ID map: map IO space and all of DRAM 1:1 using 1 GiB PTEs -// I.e. we use 1 GiB PTEs for 4 GiB. -// Linux/BSD uses this mapping just enough to replace it. -// -// Top 2G map: map the 2 Gib - 4 GiB of physical address space to -// 0xffffffff_80000000. This will be needed until the GNU toolchain can compile -// code to run at 0xffffffc000000000, i.e. the start of Sv39. -// -// Only Harvey/Plan 9 uses this Mapping, and temporarily. -// -// standard RISCV map long term: Map IO space, and all of DRAM, to the *lowest* -// possible negative address for this implementation, -// e.g. 0xffffffc000000000 for Sv39 CPUs. For now we can use GiB PTEs. -// -// RISCV map for now: map IO space, and all of DRAM, starting at -// 0xffff_ffc0_0000_0000, i.e. just as for Sv39. -// -// It is our intent on Harvey (and eventually Akaros) that we use -// this map, once the toolchain can correctly support it. -// We have tested this arrangement and it lets us boot harvey to user mode. -void init_vm(uintptr_t virtMemStart, uintptr_t physMemStart, pte_t *pt) -{ - // 0xFFF... - 0xFFFFFFFF81000000 - RISCV_PGSIZE - intptr_t memorySize = 0x7F000000; - - // middle page table - pte_t* middle_pt = (void*)pt; - size_t num_middle_pts = 2; // 3 level page table, 39 bit virtual address space for now - - // root page table - pte_t* root_pt = (void*)middle_pt + num_middle_pts * RISCV_PGSIZE; - memset(middle_pt, 0, (num_middle_pts + 1) * RISCV_PGSIZE); // 0's out middle_pt and root_pt - for (size_t i = 0; i < num_middle_pts; i++) - root_pt[(1<<RISCV_PGLEVEL_BITS)-num_middle_pts+i] = ptd_create(((uintptr_t)middle_pt >> RISCV_PGSHIFT) + i); - - // fill the middle page table - for (uintptr_t vaddr = virtMemStart, paddr = physMemStart; - paddr < physMemStart + memorySize; - vaddr += SUPERPAGE_SIZE, paddr += SUPERPAGE_SIZE) { - int l2_shift = RISCV_PGLEVEL_BITS + RISCV_PGSHIFT; - size_t l2_idx = (virtMemStart >> l2_shift) & ((1 << RISCV_PGLEVEL_BITS)-1); - l2_idx += ((vaddr - virtMemStart) >> l2_shift); - middle_pt[l2_idx] = pte_create(paddr >> RISCV_PGSHIFT, - PTE_U|PTE_R|PTE_W|PTE_X, 0); - } - - // IO space. Identity mapped. - root_pt[0x000] = pte_create(0x00000000 >> RISCV_PGSHIFT, - PTE_R | PTE_W, 0); - root_pt[0x001] = pte_create(0x40000000 >> RISCV_PGSHIFT, - PTE_R | PTE_W, 0); - root_pt[0x002] = pte_create(0x80000000 >> RISCV_PGSHIFT, - PTE_R | PTE_W | PTE_X, 0); - root_pt[0x003] = pte_create(0xc0000000 >> RISCV_PGSHIFT, - PTE_R | PTE_W | PTE_X, 0); - - // Negative address space map at 0xffffffc000000000 - root_pt[0x100] = root_pt[0]; - root_pt[0x101] = root_pt[1]; - root_pt[0x102] = root_pt[2]; - root_pt[0x103] = root_pt[3]; - - mb(); - root_page_table = root_pt; - uintptr_t ptbr = ((uintptr_t) root_pt) >> RISCV_PGSHIFT; - write_csr(sptbr, ptbr); -} - -void initVirtualMemory(void) { - uintptr_t ms; - - ms = read_csr(mstatus); - ms = INSERT_FIELD(ms, MSTATUS_VM, VM_CHOICE); - write_csr(mstatus, ms); - ms = read_csr(mstatus); - - if (EXTRACT_FIELD(ms, MSTATUS_VM) != VM_CHOICE) { - printk(BIOS_DEBUG, "We don't have virtual memory...\n"); - return; - } else { - printk(BIOS_DEBUG, "-----------------------------\n"); - printk(BIOS_DEBUG, "Virtual memory status enabled\n"); - printk(BIOS_DEBUG, "-----------------------------\n"); - } - - // TODO: Figure out how to grab this from cbfs - // N.B. We used to map physical from 0x81000000, - // but since kernels need to be able to see the page tables - // created by firmware, we're going to map from start of RAM. - // All this is subject to change as we learn more. Much - // about RISCV is still in flux. - printk(BIOS_DEBUG, "Initializing virtual memory...\n"); - uintptr_t physicalStart = 0x80000000; - uintptr_t virtualStart = 0xffffffff80000000; - init_vm(virtualStart, physicalStart, (pte_t *)_pagetables); - mb(); - flush_tlb(); - -#if IS_ENABLED(CONFIG_DEBUG_PRINT_PAGE_TABLES) - printk(BIOS_DEBUG, "Finished initializing virtual memory, starting walk...\n"); - print_page_table(); -#else - printk(BIOS_DEBUG, "Finished initializing virtual memory\n"); -#endif -} - void mstatus_init(void) { uintptr_t ms = 0;