On 31.07.24 23:37, Samuel Thibault wrote:
Juergen Gross, le mer. 31 juil. 2024 15:00:26 +0200, a ecrit:+static int set_readonly_func(unsigned long va, unsigned int lvl, bool is_leaf, + pgentry_t *pte, void *par) +{ + struct set_readonly_par *ro = par;+ mmu_updates[ro->count].ptr = virt_to_mach(pte);+ mmu_updates[ro->count].val = *pte & ~_PAGE_RW; + ro->count++; + + if ( (ro->count == L1_PAGETABLE_ENTRIES || + va + 2 * PAGE_SIZE > ro->etext) && + HYPERVISOR_mmu_update(mmu_updates, ro->count, NULL, DOMID_SELF) < 0 ) + { + printk("ERROR: set_readonly(): PTE could not be updated\n"); + do_exit(); + }Don't we also want to set ro->count to 0?
Oh, indeed. Thanks for catching this.
And assert that it is 0 after calling walk_pt in set_readonly, to make sure the va + 2 * PAGE_SIZE > ro->etext test did work properly (personally I would have rather made set_readonly call a last HYPERVISOR_mmu_update in case ro->count is not 0, which looks more robust that a quite magic-looking va + 2 * PAGE_SIZE > ro->etext test)
I think you are right. I'll do that. Juergen
