> Sent: Thursday, October 23, 2014 5:47 AM > To: kvmarm@xxxxxxxxxxxxxxxxxxxxx > Cc: Wang, Yalin; peter.maydell@xxxxxxxxxx > Subject: arm64: virtio broken in upstream kernel > > Initially I thought this was a qemu bug, so I was tracking it here. > > https://bugs.launchpad.net/qemu/+bug/1383857 > > In brief, virtio devices don't show up in the guest when using kernel > 3.18.0+rc1. I'm also using 64k pages, but I don't know if that is related. > > I did a bisect and it pointed at: > > 421520ba98290a73b35b7644e877a48f18e06004 is the first bad commit > commit 421520ba98290a73b35b7644e877a48f18e06004 > Author: Yalin Wang <Yalin.Wang@xxxxxxxxxxxxxx> > Date: Fri Sep 26 03:07:09 2014 +0100 > > ARM: 8167/1: extend the reserved memory for initrd to be page aligned > > This patch extends the start and end address of initrd to be page > aligned, > so that we can free all memory including the un-page aligned head or > tail > page of initrd, if the start or end address of initrd are not page > aligned, the page can't be freed by free_initrd_mem() function. > > Signed-off-by: Yalin Wang <yalin.wang@xxxxxxxxxxxxxx> > Acked-by: Catalin Marinas <catalin.marinas@xxxxxxx> > Signed-off-by: Russell King <rmk+kernel@xxxxxxxxxxxxxxxx> > > :040000 040000 23bd54d302533c173a4ae592969dd2868794e9ed > f1833b44ee7a389902f6f9d2fb55f4b89ba0de16 M arch > > I also reverted this patch on top of the 3.18.0+rc1 kernel, and I see: > > supermin: internal insmod virtio.ko > supermin: internal insmod virtio_ring.ko > supermin: internal insmod virtio_blk.ko > supermin: internal insmod virtio-rng.ko > supermin: internal insmod virtio_console.ko > supermin: internal insmod virtio_net.ko > supermin: internal insmod scsi_transport_spi.ko > supermin: internal insmod virtio_scsi.ko > supermin: internal insmod virtio_balloon.ko > supermin: internal insmod virtio_mmio.ko > [ 2.765223] BUG: failure at > include/linux/virtio_config.h:125/virtio_device_ready()! > [ 2.765932] Kernel panic - not syncing: BUG! > [ 2.766792] CPU: 0 PID: 18 Comm: kworker/0:1 Not tainted 3.18.0-rc1+ > #46 > [ 2.767532] Workqueue: events control_work_handler [virtio_console] > [ 2.768194] Call trace: > [ 2.768779] [<fffffe0000096cc8>] dump_backtrace+0x0/0x164 > [ 2.769252] [<fffffe0000096e48>] show_stack+0x1c/0x28 > [ 2.769618] [<fffffe0000726ff4>] dump_stack+0x74/0x94 > [ 2.769947] [<fffffe00007264a4>] panic+0xe8/0x228 > [ 2.770434] [<fffffdfffc14252c>] add_port+0x378/0x37c [virtio_console] > [ 2.770983] [<fffffdfffc142fb4>] control_work_handler+0x374/0x390 > [virtio_console] > [ 2.771552] [<fffffe00000cf8cc>] process_one_work+0x148/0x3a4 > [ 2.772067] [<fffffe00000d0358>] worker_thread+0x13c/0x488 > [ 2.772758] [<fffffe00000d53fc>] kthread+0xe0/0xf8 > [ 2.773801] Rebooting in 1 seconds..Reboot failed -- System halted > > So I guess this patch really is to blame. > > Not sure I understand exactly why aligning the initrd affects virtio > however ... > > Rich. > > -- > Richard Jones, Virtualization Group, Red Hat > http://people.redhat.com/~rjones Read my programming and virtualization > blog: http://rwmj.wordpress.com virt-builder quickly builds VMs from > scratch http://libguestfs.org/virt-builder.1.html Could you try this patch? If my patch is suspicious . --- this patch extend the start and end address of initrd to be page aligned, so that we can free all memory including the un-page aligned head or tail page of initrd, if the start or end address of initrd are not page aligned, the page can't be freed by free_initrd_mem() function. Signed-off-by: Yalin Wang <yalin.wang@xxxxxxxxxxxxxx> --- arch/arm/mm/init.c | 19 +++++++++++++++++-- arch/arm64/mm/init.c | 37 +++++++++++++++++++++++++++++++++---- 2 files changed, 50 insertions(+), 6 deletions(-) diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c index 659c75d..8490b70 100644 --- a/arch/arm/mm/init.c +++ b/arch/arm/mm/init.c @@ -277,6 +277,8 @@ phys_addr_t __init arm_memblock_steal(phys_addr_t size, phys_addr_t align) void __init arm_memblock_init(const struct machine_desc *mdesc) { /* Register the kernel text, kernel data and initrd with memblock. */ + phys_addr_t phys_initrd_start_orig __maybe_unused; + phys_addr_t phys_initrd_size_orig __maybe_unused; #ifdef CONFIG_XIP_KERNEL memblock_reserve(__pa(_sdata), _end - _sdata); #else @@ -289,6 +291,13 @@ void __init arm_memblock_init(const struct machine_desc *mdesc) phys_initrd_size = initrd_end - initrd_start; } initrd_start = initrd_end = 0; + phys_initrd_start_orig = phys_initrd_start; + phys_initrd_size_orig = phys_initrd_size; + /* make sure the start and end address are page aligned */ + phys_initrd_size = round_up(phys_initrd_start + phys_initrd_size, PAGE_SIZE); + phys_initrd_start = round_down(phys_initrd_start, PAGE_SIZE); + phys_initrd_size -= phys_initrd_start; + if (phys_initrd_size && !memblock_is_region_memory(phys_initrd_start, phys_initrd_size)) { pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region - disabling initrd\n", @@ -305,9 +314,10 @@ void __init arm_memblock_init(const struct machine_desc *mdesc) memblock_reserve(phys_initrd_start, phys_initrd_size); /* Now convert initrd to virtual addresses */ - initrd_start = __phys_to_virt(phys_initrd_start); - initrd_end = initrd_start + phys_initrd_size; + initrd_start = __phys_to_virt(phys_initrd_start_orig); + initrd_end = initrd_start + phys_initrd_size_orig; } + #endif arm_mm_memblock_reserve(); @@ -636,6 +646,11 @@ static int keep_initrd; void free_initrd_mem(unsigned long start, unsigned long end) { if (!keep_initrd) { + if (start == initrd_start) + start = round_down(start, PAGE_SIZE); + if (end == initrd_end) + end = round_up(end, PAGE_SIZE); + poison_init_mem((void *)start, PAGE_ALIGN(end) - start); free_reserved_area((void *)start, (void *)end, -1, "initrd"); } diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c index 5472c24..9dfd9a6 100644 --- a/arch/arm64/mm/init.c +++ b/arch/arm64/mm/init.c @@ -138,15 +138,38 @@ static void arm64_memory_present(void) void __init arm64_memblock_init(void) { phys_addr_t dma_phys_limit = 0; - + phys_addr_t phys_initrd_start; + phys_addr_t phys_initrd_size; /* * Register the kernel text, kernel data, initrd, and initial * pagetables with memblock. */ memblock_reserve(__pa(_text), _end - _text); #ifdef CONFIG_BLK_DEV_INITRD - if (initrd_start) - memblock_reserve(__virt_to_phys(initrd_start), initrd_end - initrd_start); + if (initrd_start) { + phys_initrd_start = __virt_to_phys(initrd_start); + phys_initrd_size = initrd_end - initrd_start; + /* make sure the start and end address are page aligned */ + phys_initrd_size = round_up(phys_initrd_start + phys_initrd_size, PAGE_SIZE); + phys_initrd_start = round_down(phys_initrd_start, PAGE_SIZE); + phys_initrd_size -= phys_initrd_start; + if (phys_initrd_size && + !memblock_is_region_memory(phys_initrd_start, phys_initrd_size)) { + pr_err("INITRD: %pa+%pa is not a memory region - disabling initrd\n", + &phys_initrd_start, &phys_initrd_size); + phys_initrd_start = phys_initrd_size = 0; + } + if (phys_initrd_size && + memblock_is_region_reserved(phys_initrd_start, phys_initrd_size)) { + pr_err("INITRD: %pa+%pa overlaps in-use memory region - disabling initrd\n", + &phys_initrd_start, &phys_initrd_size); + phys_initrd_start = phys_initrd_size = 0; + } + if (phys_initrd_size) + memblock_reserve(phys_initrd_start, phys_initrd_size); + else + initrd_start = initrd_end = 0; + } #endif if (!efi_enabled(EFI_MEMMAP)) @@ -334,8 +357,14 @@ static int keep_initrd; void free_initrd_mem(unsigned long start, unsigned long end) { - if (!keep_initrd) + if (!keep_initrd) { + if (start == initrd_start) + start = round_down(start, PAGE_SIZE); + if (end == initrd_end) + end = round_up(end, PAGE_SIZE); + free_reserved_area((void *)start, (void *)end, 0, "initrd"); + } } static int __init keepinitrd_setup(char *__unused) -- 2.1.0 _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
