On 25.01.19 02:50, JJZhu wrote:
在 2019年1月24日星期四 UTC+8下午5:17:33，J. Kiszka写道：
On 24.01.19 07:47, JJZhu wrote:
Hi all,
I am using ARM64 Ultrazed board (ZYNQ Ultrascale+ serials). I want to allocate
memeory in the initialization functions.
I use a simplified malloc and free function from other project.I modified it a
little on the start address. I use the stack_top as the beginning of first
available memory address.
Then I tested the malloc functions, it can return correct value according to
the size we need.
I can allocate memory from 0x1c000(stack_top) to 0x200000. But when it reach
0x200000, The system crashed without any print information.
Is there any limitation about stack size? 0x200000?
We only map the first 2M during arch_mmu_enable in the inmate. If you want more
(and changed the inmate config to provide more to the guest), you need to map
those ranges in your inmate setup yourself (map_range).
Another question is that if I start another bare-metal inmate, does it share
the same stack as the first inmate or it has its own stack also available from
0x1c000 on it own memory?
If you create a different config which assigns different physical memory to the
virtual range of the inmate, there will be no sharing, of course.
Jan
--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux
Hi Jan,
I have checked the arch_mmu_enable(). It only maps two 0x10000 size of memory.
void arch_mmu_enable(void)
{
unsigned long mair, sctlr;
map_range((void*)CONFIG_INMATE_BASE, 0x10000, MAP_CACHED);
map_range((void*)COMM_REGION_BASE, PAGE_SIZE, MAP_CACHED);
/*
* ARMv7: Use attributes 0 and 1 in MAIR0
* ARMv8: Use attributes 0 and 1 in MAIR
*
* Attributes 0: inner/outer: normal memory, outer write-back
* non-transient
* Attributes 1: device memory
*/
mair = MAIR_ATTR(1, MAIR_ATTR_DEVICE) | MAIR_ATTR(0, MAIR_ATTR_WBRWA);
arm_write_sysreg(MAIR, mair);
arm_write_sysreg(TRANSL_CONT_REG, TRANSL_CONT_REG_SETTINGS);
arm_write_sysreg(TTBR0, page_directory);
/* This barrier ensures that TTBR0 is set before enabling the MMU. */
instruction_barrier();
arm_read_sysreg(SCTLR, sctlr);
sctlr |= SCTLR_MMU_CACHES;
arm_write_sysreg(SCTLR, sctlr);
/* This barrier ensures that the MMU is actually on */
instruction_barrier();
/* MMU is enabled from now on */
}
Where is the 2M memory? Can I just change one paremeter somewhere to make it
double?
The current implementation of map_range on ARM performs that mapping at 2M
granularity (HUGE_PAGE_SIZE). So that 64K request is rounded up to 2M.
For the config file:
I have modified the config file which should be enough.
/* RAM */ {
.phys_start = 0x43000000,
.virt_start = 0,
.size = 0x10000000,
.flags = JAILHOUSE_MEM_READ | JAILHOUSE_MEM_WRITE |
JAILHOUSE_MEM_EXECUTE | JAILHOUSE_MEM_LOADABLE,
},
Could you give some more detailed suggestion on where to modify?
Add some map_range call to your main function. That could actually map that
whole range (map_range(0, 0x10000000, MAP_CACHED)).
Jan
--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux
--
You received this message because you are subscribed to the Google Groups
"Jailhouse" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
