Re: Question about (and problem with) pflash data access
On 13/02/2020 10:50, Philippe Mathieu-Daudé wrote: > Cc'ing Paolo and Alexey. > > On 2/13/20 12:09 AM, Guenter Roeck wrote: >> On Wed, Feb 12, 2020 at 10:39:30PM +0100, Philippe Mathieu-Daudé wrote: >>> Cc'ing Jean-Christophe and Peter. >>> >>> On 2/12/20 7:46 PM, Guenter Roeck wrote: Hi, I have been playing with pflash recently. For the most part it works, but I do have an odd problem when trying to instantiate pflash on sx1. My data file looks as follows. 000 0001 020 * 0002000 0002 0002020 * 0004000 0003 0004020 ... In the sx1 machine, this becomes: 000 6001 020 * 0002000 6002 0002020 * 0004000 6003 0004020 * ... pflash is instantiated with "-drive file=flash.32M.test,format=raw,if=pflash". I don't have much success with pflash tracing - data accesses don't show up there. I did find a number of problems with the sx1 emulation, but I have no clue what is going on with pflash. As far as I can see pflash works fine on other machines. Can someone give me a hint what to look out for ? >>> >>> This is specific to the SX1, introduced in commit 997641a84ff: >>> >>> 64 static uint64_t static_read(void *opaque, hwaddr offset, >>> 65 unsigned size) >>> 66 { >>> 67 uint32_t *val = (uint32_t *) opaque; >>> 68 uint32_t mask = (4 / size) - 1; >>> 69 >>> 70 return *val >> ((offset & mask) << 3); >>> 71 } >>> >>> Only guessing, this looks like some hw parity, and I imagine you need to >>> write the parity bits in your flash.32M file before starting QEMU, >>> then it >>> would appear "normal" within the guest. >>> >> I thought this might be related, but that is not the case. I added log >> messages, and even ran the code in gdb. static_read() and static_write() >> are not executed. >> >> Also, >> >> memory_region_init_io([0], NULL, _ops, , >> "sx1.cs0", OMAP_CS0_SIZE - flash_size); >> ^^ >> memory_region_add_subregion(address_space, >> OMAP_CS0_BASE + flash_size, [0]); >> ^^ >> >> suggests that the code is only executed for memory accesses _after_ >> the actual flash. The memory tree is: >> >> memory-region: system >> - (prio 0, i/o): system >> -01ff (prio 0, romd): omap_sx1.flash0-1 >> -01ff (prio 0, rom): omap_sx1.flash0-0 > > Eh two memory regions with same size and same priority... Is this legal? I'd say yes if used with memory_region_set_enabled() to make sure only one is enabled. Having both enabled is weird and we should print a warning. Thanks, > > (qemu) info mtree -f -d > FlatView #0 > AS "memory", root: system > AS "cpu-memory-0", root: system > Root memory region: system > -01ff (prio 0, romd): omap_sx1.flash0-1 > 0200-03ff (prio 0, i/o): sx1.cs0 > 0400-07ff (prio 0, i/o): sx1.cs1 > 0800-0bff (prio 0, i/o): sx1.cs3 > 1000-11ff (prio 0, ram): omap1.dram > 2000-2002 (prio 0, ram): omap1.sram > ... > Dispatch > Physical sections > #0 @.. (noname) [unassigned] > #1 @..01ff omap_sx1.flash0-1 [not dirty] > #2 @0200..03ff sx1.cs0 [ROM] > #3 @0400..07ff sx1.cs1 [watch] > #4 @0800..0bff sx1.cs3 > #5 @1000..11ff omap1.dram > #6 @2000..2002 omap1.sram > ... > Nodes (9 bits per level, 6 levels) ptr=[3] skip=4 > [0] > 0 skip=3 ptr=[3] > 1..511 skip=1 ptr=NIL > [1] > 0 skip=2 ptr=[3] > 1..511 skip=1 ptr=NIL > [2] > 0 skip=1 ptr=[3] > 1..511 skip=1 ptr=NIL > [3] > 0 skip=1 ptr=[4] > 1 skip=1 ptr=[5] > 2 skip=2 ptr=[7] > 3..13 skip=1 ptr=NIL > 14 skip=2 ptr=[9] > 15 skip=2 ptr=[11] > 16..511 skip=1 ptr=NIL >
Re: [Qemu-block] [PATCH v2 0/3] Don't write headers if BDS is INACTIVE
On 31/10/17 00:10, Jeff Cody wrote: > Changes from v1->v2: > > * Drop previous parallels patches, just check BDRV_O_INACTIVE now > (Kevin) > > git-backport-diff -r qemu/master.. -u github/master > Key: > [] : patches are identical > [] : number of functional differences between upstream/downstream patch > [down] : patch is downstream-only > The flags [FC] indicate (F)unctional and (C)ontextual differences, > respectively > > 001/3:[] [--] 'block/vhdx.c: Don't blindly update the header' > 002/3:[down] 'block/parallals: Do not update header or truncate image when > INMIGRATE' > 003/3:[] [--] 'qemu-iotests: update unsupported image formats in 194' Tested-by: Alexey Kardashevskiy <a...@ozlabs.ru> > > v1: > > VHDX and Parallels both blindly write headers to the image file > if the images are opened R/W. This causes an assert if the QEMU run > state is INMIGRATE. > > Jeff Cody (3): > block/vhdx.c: Don't blindly update the header > block/parallals: Do not update header or truncate image when INMIGRATE > qemu-iotests: update unsupported image formats in 194 > > block/parallels.c | 7 ++- > block/vhdx.c | 7 --- > tests/qemu-iotests/194 | 2 +- > 3 files changed, 3 insertions(+), 13 deletions(-) > -- Alexey
Re: [Qemu-block] [Qemu-devel] [PATCH] qcow2: allocate cluster_cache/cluster_data on demand
On 31/08/17 03:20, Stefan Hajnoczi wrote: > On Tue, Aug 22, 2017 at 02:56:00PM +1000, Alexey Kardashevskiy wrote: >> On 19/08/17 12:46, Alexey Kardashevskiy wrote: >>> On 19/08/17 01:18, Eric Blake wrote: >>>> On 08/18/2017 08:31 AM, Stefan Hajnoczi wrote: >>>>> Most qcow2 files are uncompressed so it is wasteful to allocate (32 + 1) >>>>> * cluster_size + 512 bytes upfront. Allocate s->cluster_cache and >>>>> s->cluster_data when the first read operation is performance on a >>>>> compressed cluster. >>>>> >>>>> The buffers are freed in .bdrv_close(). .bdrv_open() no longer has any >>>>> code paths that can allocate these buffers, so remove the free functions >>>>> in the error code path. >>>>> >>>>> Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru> >>>>> Cc: Kevin Wolf <kw...@redhat.com> >>>>> Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> >>>>> --- >>>>> Alexey: Does this improve your memory profiling results? >>>> >>>> Is this a regression from earlier versions? >>> >>> Hm, I have not thought about this. >>> >>> So. I did bisect and this started happening from >>> 9a4c0e220d8a4f82b5665d0ee95ef94d8e1509d5 >>> "hw/virtio-pci: fix virtio behaviour" >>> >>> Before that, the very same command line would take less than 1GB of >>> resident memory. That thing basically enforces virtio-1.0 for QEMU <=2.6 >>> which means that upstream with "-machine pseries-2.6" works fine (less than >>> 1GB), "-machine pseries-2.7" does not (close to 7GB, sometime even 9GB). >>> >>> Then I tried bisecting again, with >>> "scsi=off,disable-modern=off,disable-legacy=on" on my 150 virtio-block >>> devices, started from >>> e266d421490e0 "virtio-pci: add flags to enable/disable legacy/modern" (it >>> added the disable-modern switch) which uses 2GB of memory. >>> >>> I ended up with ada434cd0b44 "virtio-pci: implement cfg capability". >>> >>> Then I removed proxy->modern_as on v2.10.0-rc3 (see below) and got 1.5GB of >>> used memory (yay!) >>> >>> I do not really know how to reinterpret all of this, do you? >> >> >> Anyone, ping? Should I move the conversation to the original thread? Any >> hacks to try with libc? > > I suggest a new top-level thread with Michael Tsirkin CCed. I am continuing in the original "Memory use with >100 virtio devices" thread and the problem is more generic than virtio, it is just easier to reproduce it with virtio, that's all. -- Alexey
Re: [Qemu-block] [Qemu-devel] [PATCH] qcow2: allocate cluster_cache/cluster_data on demand
On 19/08/17 12:46, Alexey Kardashevskiy wrote: > On 19/08/17 01:18, Eric Blake wrote: >> On 08/18/2017 08:31 AM, Stefan Hajnoczi wrote: >>> Most qcow2 files are uncompressed so it is wasteful to allocate (32 + 1) >>> * cluster_size + 512 bytes upfront. Allocate s->cluster_cache and >>> s->cluster_data when the first read operation is performance on a >>> compressed cluster. >>> >>> The buffers are freed in .bdrv_close(). .bdrv_open() no longer has any >>> code paths that can allocate these buffers, so remove the free functions >>> in the error code path. >>> >>> Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru> >>> Cc: Kevin Wolf <kw...@redhat.com> >>> Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> >>> --- >>> Alexey: Does this improve your memory profiling results? >> >> Is this a regression from earlier versions? > > Hm, I have not thought about this. > > So. I did bisect and this started happening from > 9a4c0e220d8a4f82b5665d0ee95ef94d8e1509d5 > "hw/virtio-pci: fix virtio behaviour" > > Before that, the very same command line would take less than 1GB of > resident memory. That thing basically enforces virtio-1.0 for QEMU <=2.6 > which means that upstream with "-machine pseries-2.6" works fine (less than > 1GB), "-machine pseries-2.7" does not (close to 7GB, sometime even 9GB). > > Then I tried bisecting again, with > "scsi=off,disable-modern=off,disable-legacy=on" on my 150 virtio-block > devices, started from > e266d421490e0 "virtio-pci: add flags to enable/disable legacy/modern" (it > added the disable-modern switch) which uses 2GB of memory. > > I ended up with ada434cd0b44 "virtio-pci: implement cfg capability". > > Then I removed proxy->modern_as on v2.10.0-rc3 (see below) and got 1.5GB of > used memory (yay!) > > I do not really know how to reinterpret all of this, do you? Anyone, ping? Should I move the conversation to the original thread? Any hacks to try with libc? > > > Note: 1GB..9GB numbers from below are the peak values from valgrind's > massif. This is pretty much resident memory used by QEMU process. In my > testing I did not enable KVM and I did not start the guest (i.e. used -S). > 150 virtio-block devices, 2GB RAM for the guest. > > Also, while bisecting, I only paid attention if it is 1..2GB or 6..9GB - > all tests did fit these 2 ranges, for any given sha1 the amount of memory > would be stable but among "good" commits it could change between 1GB and 2GB. > > > > diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c > index 5d14bd6..7ad447a 100644 > --- a/hw/virtio/virtio-pci.c > +++ b/hw/virtio/virtio-pci.c > @@ -1783,6 +1783,7 @@ static void virtio_pci_realize(PCIDevice *pci_dev, > Error **errp) > /* PCI BAR regions must be powers of 2 */ > pow2ceil(proxy->notify.offset + proxy->notify.size)); > > +#if 0 > memory_region_init_alias(>modern_cfg, > OBJECT(proxy), > "virtio-pci-cfg", > @@ -1791,7 +1792,7 @@ static void virtio_pci_realize(PCIDevice *pci_dev, > Error **errp) > memory_region_size(>modern_bar)); > > address_space_init(>modern_as, >modern_cfg, > "virtio-pci-cfg-as"); > - > +#endif > if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) { > proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; > } > @@ -1860,10 +1861,10 @@ static void virtio_pci_realize(PCIDevice *pci_dev, > Error **errp) > > static void virtio_pci_exit(PCIDevice *pci_dev) > { > -VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); > +//VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); > > msix_uninit_exclusive_bar(pci_dev); > -address_space_destroy(>modern_as); > +//address_space_destroy(>modern_as); > } > > static void virtio_pci_reset(DeviceState *qdev) > > > > > >> Likely, it is NOT -rc4 >> material, and thus can wait for 2.11; but it should be fine for -stable >> as part of 2.10.1 down the road. >> >>> +++ b/block/qcow2-cluster.c >>> @@ -1516,6 +1516,23 @@ int qcow2_decompress_cluster(BlockDriverState *bs, >>> uint64_t cluster_offset) >>> nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) >>> + 1; >>> sector_offset = coffset & 511; >>> csize = nb_csectors * 512 - sector_offset; >>>
Re: [Qemu-block] [Qemu-devel] [PATCH] qcow2: allocate cluster_cache/cluster_data on demand
On 19/08/17 12:46, Alexey Kardashevskiy wrote: > On 19/08/17 01:18, Eric Blake wrote: >> On 08/18/2017 08:31 AM, Stefan Hajnoczi wrote: >>> Most qcow2 files are uncompressed so it is wasteful to allocate (32 + 1) >>> * cluster_size + 512 bytes upfront. Allocate s->cluster_cache and >>> s->cluster_data when the first read operation is performance on a >>> compressed cluster. >>> >>> The buffers are freed in .bdrv_close(). .bdrv_open() no longer has any >>> code paths that can allocate these buffers, so remove the free functions >>> in the error code path. >>> >>> Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru> >>> Cc: Kevin Wolf <kw...@redhat.com> >>> Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> >>> --- >>> Alexey: Does this improve your memory profiling results? >> >> Is this a regression from earlier versions? > > Hm, I have not thought about this. > > So. I did bisect and this started happening from > 9a4c0e220d8a4f82b5665d0ee95ef94d8e1509d5 > "hw/virtio-pci: fix virtio behaviour" > > Before that, the very same command line would take less than 1GB of > resident memory. That thing basically enforces virtio-1.0 for QEMU <=2.6 > which means that upstream with "-machine pseries-2.6" works fine (less than > 1GB), "-machine pseries-2.7" does not (close to 7GB, sometime even 9GB). > > Then I tried bisecting again, with > "scsi=off,disable-modern=off,disable-legacy=on" on my 150 virtio-block > devices, started from > e266d421490e0 "virtio-pci: add flags to enable/disable legacy/modern" (it > added the disable-modern switch) which uses 2GB of memory. > > I ended up with ada434cd0b44 "virtio-pci: implement cfg capability". > > Then I removed proxy->modern_as on v2.10.0-rc3 (see below) and got 1.5GB of > used memory (yay!) > > I do not really know how to reinterpret all of this, do you? > > > Note: 1GB..9GB numbers from below are the peak values from valgrind's s/from below/from above/ , sorry, bad cut-n-paste :) -- Alexey signature.asc Description: OpenPGP digital signature
Re: [Qemu-block] [PATCH] qcow2: allocate cluster_cache/cluster_data on demand
On 18/08/17 23:31, Stefan Hajnoczi wrote: > Most qcow2 files are uncompressed so it is wasteful to allocate (32 + 1) > * cluster_size + 512 bytes upfront. Allocate s->cluster_cache and > s->cluster_data when the first read operation is performance on a > compressed cluster. > > The buffers are freed in .bdrv_close(). .bdrv_open() no longer has any > code paths that can allocate these buffers, so remove the free functions > in the error code path. > > Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru> > Cc: Kevin Wolf <kw...@redhat.com> > Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> > --- > Alexey: Does this improve your memory profiling results? Yes, it does: was: 12.81% (1,023,193,088B) now: 05.36% (393,893,888B) Tested-by: Alexey Kardashevskiy <a...@ozlabs.ru> > block/qcow2-cluster.c | 17 + > block/qcow2.c | 12 > 2 files changed, 17 insertions(+), 12 deletions(-) > > diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c > index f06c08f64c..c47600a44e 100644 > --- a/block/qcow2-cluster.c > +++ b/block/qcow2-cluster.c > @@ -1516,6 +1516,23 @@ int qcow2_decompress_cluster(BlockDriverState *bs, > uint64_t cluster_offset) > nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + > 1; > sector_offset = coffset & 511; > csize = nb_csectors * 512 - sector_offset; > + > +/* Allocate buffers on first decompress operation, most images are > + * uncompressed and the memory overhead can be avoided. The buffers > + * are freed in .bdrv_close(). > + */ > +if (!s->cluster_data) { > +/* one more sector for decompressed data alignment */ > +s->cluster_data = qemu_try_blockalign(bs->file->bs, > +QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512); > +if (!s->cluster_data) { > +return -EIO; > +} > +} > +if (!s->cluster_cache) { > +s->cluster_cache = g_malloc(s->cluster_size); > +} > + > BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED); > ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, > nb_csectors); > diff --git a/block/qcow2.c b/block/qcow2.c > index 40ba26c111..0ac201910a 100644 > --- a/block/qcow2.c > +++ b/block/qcow2.c > @@ -1360,16 +1360,6 @@ static int qcow2_do_open(BlockDriverState *bs, QDict > *options, int flags, > goto fail; > } > > -s->cluster_cache = g_malloc(s->cluster_size); > -/* one more sector for decompressed data alignment */ > -s->cluster_data = qemu_try_blockalign(bs->file->bs, > QCOW_MAX_CRYPT_CLUSTERS > -* s->cluster_size + 512); > -if (s->cluster_data == NULL) { > -error_setg(errp, "Could not allocate temporary cluster buffer"); > -ret = -ENOMEM; > -goto fail; > -} > - > s->cluster_cache_offset = -1; > s->flags = flags; > > @@ -1507,8 +1497,6 @@ static int qcow2_do_open(BlockDriverState *bs, QDict > *options, int flags, > if (s->refcount_block_cache) { > qcow2_cache_destroy(bs, s->refcount_block_cache); > } > -g_free(s->cluster_cache); > -qemu_vfree(s->cluster_data); > qcrypto_block_free(s->crypto); > qapi_free_QCryptoBlockOpenOptions(s->crypto_opts); > return ret; > -- Alexey
Re: [Qemu-block] [Qemu-devel] [PATCH] qcow2: allocate cluster_cache/cluster_data on demand
On 19/08/17 01:18, Eric Blake wrote: > On 08/18/2017 08:31 AM, Stefan Hajnoczi wrote: >> Most qcow2 files are uncompressed so it is wasteful to allocate (32 + 1) >> * cluster_size + 512 bytes upfront. Allocate s->cluster_cache and >> s->cluster_data when the first read operation is performance on a >> compressed cluster. >> >> The buffers are freed in .bdrv_close(). .bdrv_open() no longer has any >> code paths that can allocate these buffers, so remove the free functions >> in the error code path. >> >> Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru> >> Cc: Kevin Wolf <kw...@redhat.com> >> Signed-off-by: Stefan Hajnoczi <stefa...@redhat.com> >> --- >> Alexey: Does this improve your memory profiling results? > > Is this a regression from earlier versions? Hm, I have not thought about this. So. I did bisect and this started happening from 9a4c0e220d8a4f82b5665d0ee95ef94d8e1509d5 "hw/virtio-pci: fix virtio behaviour" Before that, the very same command line would take less than 1GB of resident memory. That thing basically enforces virtio-1.0 for QEMU <=2.6 which means that upstream with "-machine pseries-2.6" works fine (less than 1GB), "-machine pseries-2.7" does not (close to 7GB, sometime even 9GB). Then I tried bisecting again, with "scsi=off,disable-modern=off,disable-legacy=on" on my 150 virtio-block devices, started from e266d421490e0 "virtio-pci: add flags to enable/disable legacy/modern" (it added the disable-modern switch) which uses 2GB of memory. I ended up with ada434cd0b44 "virtio-pci: implement cfg capability". Then I removed proxy->modern_as on v2.10.0-rc3 (see below) and got 1.5GB of used memory (yay!) I do not really know how to reinterpret all of this, do you? Note: 1GB..9GB numbers from below are the peak values from valgrind's massif. This is pretty much resident memory used by QEMU process. In my testing I did not enable KVM and I did not start the guest (i.e. used -S). 150 virtio-block devices, 2GB RAM for the guest. Also, while bisecting, I only paid attention if it is 1..2GB or 6..9GB - all tests did fit these 2 ranges, for any given sha1 the amount of memory would be stable but among "good" commits it could change between 1GB and 2GB. diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c index 5d14bd6..7ad447a 100644 --- a/hw/virtio/virtio-pci.c +++ b/hw/virtio/virtio-pci.c @@ -1783,6 +1783,7 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) /* PCI BAR regions must be powers of 2 */ pow2ceil(proxy->notify.offset + proxy->notify.size)); +#if 0 memory_region_init_alias(>modern_cfg, OBJECT(proxy), "virtio-pci-cfg", @@ -1791,7 +1792,7 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) memory_region_size(>modern_bar)); address_space_init(>modern_as, >modern_cfg, "virtio-pci-cfg-as"); - +#endif if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) { proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF; } @@ -1860,10 +1861,10 @@ static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp) static void virtio_pci_exit(PCIDevice *pci_dev) { -VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); +//VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev); msix_uninit_exclusive_bar(pci_dev); -address_space_destroy(>modern_as); +//address_space_destroy(>modern_as); } static void virtio_pci_reset(DeviceState *qdev) > Likely, it is NOT -rc4 > material, and thus can wait for 2.11; but it should be fine for -stable > as part of 2.10.1 down the road. > >> +++ b/block/qcow2-cluster.c >> @@ -1516,6 +1516,23 @@ int qcow2_decompress_cluster(BlockDriverState *bs, >> uint64_t cluster_offset) >> nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) >> + 1; >> sector_offset = coffset & 511; >> csize = nb_csectors * 512 - sector_offset; >> + >> +/* Allocate buffers on first decompress operation, most images are >> + * uncompressed and the memory overhead can be avoided. The buffers >> + * are freed in .bdrv_close(). >> + */ >> +if (!s->cluster_data) { >> +/* one more sector for decompressed data alignment */ >> +s->cluster_data = qemu_try_blockalign(bs->file->bs, >> +QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size + 512); >> +if (!s->cluster_data) { >> +return -EIO; > > Is -ENOMEM any better than -EIO here? > -- Alexey signature.asc Description: OpenPGP digital signature