| Message ID | 20171207114545.23845-1-marc.zyngier@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On 07/12/17 11:45, Marc Zyngier wrote: > When we unmap the HYP memory, we try to be clever and unmap one > PGD at a time. If we start with a non-PGD aligned address and try > to unmap a whole PGD, things go horribly wrong in unmap_hyp_range > (addr and end can never match, and it all goes really badly as we > keep incrementing pgd and parse random memory as page tables...). > > The obvious fix is to let unmap_hyp_range do what it does best, > which is to iterate over a range. > > Cc: stable@vger.kernel.org > Reported-by: Andre Przywara <andre.przywara@arm.com> > Signed-off-by: Marc Zyngier <marc.zyngier@arm.com> Thanks for the patch (and the hours of analysis that preceded it)! So yes, that fixes the crash with 4.15-rc1 on my Midway (with the original DT). As expected, KVM gets shuts down in the process, so no one is at home under chardev 10/232 anymore - with this patch only, that is. Tested-by: Andre Przywara <andre.przywara@arm.com> Cheers, Andre. > --- > virt/kvm/arm/mmu.c | 10 ++++------ > 1 file changed, 4 insertions(+), 6 deletions(-) > > diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c > index b36945d49986..b4b69c2d1012 100644 > --- a/virt/kvm/arm/mmu.c > +++ b/virt/kvm/arm/mmu.c > @@ -509,8 +509,6 @@ static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size) > */ > void free_hyp_pgds(void) > { > - unsigned long addr; > - > mutex_lock(&kvm_hyp_pgd_mutex); > > if (boot_hyp_pgd) { > @@ -521,10 +519,10 @@ void free_hyp_pgds(void) > > if (hyp_pgd) { > unmap_hyp_range(hyp_pgd, hyp_idmap_start, PAGE_SIZE); > - for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) > - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); > - for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) > - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); > + unmap_hyp_range(hyp_pgd, kern_hyp_va(PAGE_OFFSET), > + (uintptr_t)high_memory - PAGE_OFFSET); > + unmap_hyp_range(hyp_pgd, kern_hyp_va(VMALLOC_START), > + VMALLOC_END - VMALLOC_START); > > free_pages((unsigned long)hyp_pgd, hyp_pgd_order); > hyp_pgd = NULL; >
On Thu, Dec 07, 2017 at 11:45:45AM +0000, Marc Zyngier wrote: > When we unmap the HYP memory, we try to be clever and unmap one > PGD at a time. If we start with a non-PGD aligned address and try > to unmap a whole PGD, things go horribly wrong in unmap_hyp_range > (addr and end can never match, and it all goes really badly as we > keep incrementing pgd and parse random memory as page tables...). > > The obvious fix is to let unmap_hyp_range do what it does best, > which is to iterate over a range. Would you mind terribly if I add the following to the commit message? The size of the linear mapping, which begins at PAGE_OFFSET, can be easily calculated by subtracting PAGE_OFFSET form high_memory, because high_memory is defined as the linear map address of the last byte of DRAM, plus one. The size of the vmalloc region is given trivially by VMALLOC_END - VMALLOC_START. Otherwise: Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org> > > Cc: stable@vger.kernel.org > Reported-by: Andre Przywara <andre.przywara@arm.com> > Signed-off-by: Marc Zyngier <marc.zyngier@arm.com> > --- > virt/kvm/arm/mmu.c | 10 ++++------ > 1 file changed, 4 insertions(+), 6 deletions(-) > > diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c > index b36945d49986..b4b69c2d1012 100644 > --- a/virt/kvm/arm/mmu.c > +++ b/virt/kvm/arm/mmu.c > @@ -509,8 +509,6 @@ static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size) > */ > void free_hyp_pgds(void) > { > - unsigned long addr; > - > mutex_lock(&kvm_hyp_pgd_mutex); > > if (boot_hyp_pgd) { > @@ -521,10 +519,10 @@ void free_hyp_pgds(void) > > if (hyp_pgd) { > unmap_hyp_range(hyp_pgd, hyp_idmap_start, PAGE_SIZE); > - for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) > - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); > - for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) > - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); > + unmap_hyp_range(hyp_pgd, kern_hyp_va(PAGE_OFFSET), > + (uintptr_t)high_memory - PAGE_OFFSET); > + unmap_hyp_range(hyp_pgd, kern_hyp_va(VMALLOC_START), > + VMALLOC_END - VMALLOC_START); > > free_pages((unsigned long)hyp_pgd, hyp_pgd_order); > hyp_pgd = NULL; > -- > 2.14.2 >
On 11/12/17 09:05, Christoffer Dall wrote: > On Thu, Dec 07, 2017 at 11:45:45AM +0000, Marc Zyngier wrote: >> When we unmap the HYP memory, we try to be clever and unmap one >> PGD at a time. If we start with a non-PGD aligned address and try >> to unmap a whole PGD, things go horribly wrong in unmap_hyp_range >> (addr and end can never match, and it all goes really badly as we >> keep incrementing pgd and parse random memory as page tables...). >> >> The obvious fix is to let unmap_hyp_range do what it does best, >> which is to iterate over a range. > > Would you mind terribly if I add the following to the commit message? > > The size of the linear mapping, which begins at PAGE_OFFSET, can be > easily calculated by subtracting PAGE_OFFSET form high_memory, because > high_memory is defined as the linear map address of the last byte of > DRAM, plus one. > > The size of the vmalloc region is given trivially by VMALLOC_END - > VMALLOC_START. Please do! > Otherwise: > > Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org> Thanks, M.
diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c index b36945d49986..b4b69c2d1012 100644 --- a/virt/kvm/arm/mmu.c +++ b/virt/kvm/arm/mmu.c @@ -509,8 +509,6 @@ static void unmap_hyp_range(pgd_t *pgdp, phys_addr_t start, u64 size) */ void free_hyp_pgds(void) { - unsigned long addr; - mutex_lock(&kvm_hyp_pgd_mutex); if (boot_hyp_pgd) { @@ -521,10 +519,10 @@ void free_hyp_pgds(void) if (hyp_pgd) { unmap_hyp_range(hyp_pgd, hyp_idmap_start, PAGE_SIZE); - for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); - for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) - unmap_hyp_range(hyp_pgd, kern_hyp_va(addr), PGDIR_SIZE); + unmap_hyp_range(hyp_pgd, kern_hyp_va(PAGE_OFFSET), + (uintptr_t)high_memory - PAGE_OFFSET); + unmap_hyp_range(hyp_pgd, kern_hyp_va(VMALLOC_START), + VMALLOC_END - VMALLOC_START); free_pages((unsigned long)hyp_pgd, hyp_pgd_order); hyp_pgd = NULL;
When we unmap the HYP memory, we try to be clever and unmap one PGD at a time. If we start with a non-PGD aligned address and try to unmap a whole PGD, things go horribly wrong in unmap_hyp_range (addr and end can never match, and it all goes really badly as we keep incrementing pgd and parse random memory as page tables...). The obvious fix is to let unmap_hyp_range do what it does best, which is to iterate over a range. Cc: stable@vger.kernel.org Reported-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com> --- virt/kvm/arm/mmu.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)