oneechanhax/phunix
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

For /dev/mem, map in memory to be copied to memory's own address space

Browse Source 
one page at a time, and use safecopies to copy it to the requesting
process.

This lets /dev/mem access the entire physical address space, as the minix
page tables only allow access by default to physical RAM, which breaks
e.g. the VESA X driver in some cases.
This commit is contained in:
Ben Gras 2007-03-09 16:03:19 +00:00
parent f4b7a16f7b
commit 9843d7a625
1 changed files with 53 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

95
drivers/memory/memory.c
View File

@ -38,7 +38,6 @@ PRIVATE struct device m_geom[NR_DEVS]; /* base and size of each device */
PRIVATE int m_seg[NR_DEVS]; /* segment index of each device */
PRIVATE int m_device; /* current device */
PRIVATE struct kinfo kinfo; /* kernel information */
PRIVATE struct machine machine; /* machine information */
extern int errno; /* error number for PM calls */
@ -69,6 +68,12 @@ PRIVATE struct driver m_dtab = {
NULL
};
/* One page of temporary mapping area - enough to be able to page-align
* one page.
*/
static char pagedata_buf[2*PAGE_SIZE];
vir_bytes pagedata_aligned;
/* Buffer for the /dev/zero null byte feed. */
#define ZERO_BUF_SIZE 1024
PRIVATE char dev_zero[ZERO_BUF_SIZE];
@ -133,14 +138,11 @@ int safe; /* safe copies */
int seg;
unsigned count, left, chunk;
vir_bytes user_vir, vir_offset = 0;
phys_bytes user_phys;
struct device *dv;
unsigned long dv_size;
int s, r;
off_t position;
static int n = 0;
if(!safe) {
printf("m_transfer: unsafe?\n");
return EPERM;
@ -187,32 +189,56 @@ int safe; /* safe copies */
}
break;
/* Physical copying. Only used to access entire memory. */
/* Physical copying. Only used to access entire memory.
* Transfer one 'page window' at a time.
*/
case MEM_DEV:
if (position >= dv_size) {
printf("memory: read 0x%lx beyond physical memory of 0x%lx\n",
position, dv_size);
{
u32_t pagestart, page_off;
static u32_t pagestart_mapped;
static int any_mapped = 0;
int r;
u32_t subcount;
if (position >= dv_size)
return(OK); /* check for EOF */
}
if (position + count > dv_size) {
printf("memory: truncating count from %d to ", count);
if (position + count > dv_size)
count = dv_size - position;
printf("%d (size %d)\n", count, dv_size);
}
mem_phys = cv64ul(dv->dv_base) + position;
if((r=sys_umap(proc_nr, GRANT_SEG, user_vir,
count + vir_offset, &user_phys)) != OK) {
panic("MEM","sys_umap failed in m_transfer",r);
}
page_off = mem_phys % PAGE_SIZE;
pagestart = mem_phys - page_off;
/* All memory to the map call has to be page-aligned.
* Don't have to map same page over and over.
*/
if(!any_mapped || pagestart_mapped != pagestart) {
if((r=sys_vm_map(SELF, 1, pagedata_aligned,
PAGE_SIZE, pagestart)) != OK) {
printf("memory: sys_vm_map failed: %d\n", r);
return r;
}
any_mapped = 1;
pagestart_mapped = pagestart;
}
/* how much to be done within this page. */
subcount = PAGE_SIZE-page_off;
if(subcount > count)
subcount = count;
if (opcode == DEV_GATHER_S) { /* copy data */
sys_physcopy(NONE, PHYS_SEG, mem_phys,
NONE, PHYS_SEG, user_phys + vir_offset, count);
s=sys_safecopyto(proc_nr, user_vir,
vir_offset, pagedata_aligned+page_off, subcount, D);
} else {
sys_physcopy(NONE, PHYS_SEG, user_phys + vir_offset,
NONE, PHYS_SEG, mem_phys, count);
s=sys_safecopyfrom(proc_nr, user_vir,
vir_offset, pagedata_aligned+page_off, subcount, D);
}
if(s != OK)
return s;
count = subcount;
break;
}
/* Null byte stream generator. */
case ZERO_DEV:
@ -224,7 +250,7 @@ int safe; /* safe copies */
s=sys_safecopyto(proc_nr, user_vir,
vir_offset+suboffset, (vir_bytes) dev_zero, chunk, D);
if(s != OK)
report("MEM","sys_vircopy failed", s);
report("MEM","sys_safecopyto failed", s);
left -= chunk;
suboffset += chunk;
}
@ -290,30 +316,12 @@ PRIVATE void m_init()
/* Initialize this task. All minor devices are initialized one by one. */
u32_t ramdev_size;
u32_t ramdev_base;
message m;
int i, s;
phys_bytes mem_top = 0;
/* Physical memory, to check validity of /dev/mem access. */
#define MAX_MEM_RANGES 10
struct memory mem_chunks[MAX_MEM_RANGES];
if (OK != (s=sys_getkinfo(&kinfo))) {
panic("MEM","Couldn't get kernel information.",s);
}
/* Obtain physical memory chunks for /dev/mem memory. */
if(env_memory_parse(mem_chunks, MAX_MEM_RANGES) != OK)
printf("memory driver: no memory layout, /dev/mem won't work\n");
else {
for(i = 0; i < MAX_MEM_RANGES; i++) {
phys_bytes top;
top = mem_chunks[i].base + mem_chunks[i].size;
if(top > mem_top)
mem_top = top;
}
}
/* Install remote segment for /dev/kmem memory. */
m_geom[KMEM_DEV].dv_base = cvul64(kinfo.kmem_base);
m_geom[KMEM_DEV].dv_size = cvul64(kinfo.kmem_size);
@ -355,8 +363,12 @@ PRIVATE void m_init()
dev_zero[i] = '\0';
}
/* Page-align page pointer. */
pagedata_aligned = (u32_t) pagedata_buf + PAGE_SIZE;
pagedata_aligned -= pagedata_aligned % PAGE_SIZE;
/* Set up memory range for /dev/mem. */
m_geom[MEM_DEV].dv_size = cvul64(mem_top);
m_geom[MEM_DEV].dv_size = cvul64(0xffffffff);
}
/*===========================================================================*
@ -384,7 +396,6 @@ int safe;
u32_t ramdev_size;
phys_bytes ramdev_base;
message m;
int s;
/* A ramdisk can be created only once, and only on RAM disk device. */