From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:36544) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YaHXb-0005jC-V2 for qemu-devel@nongnu.org; Tue, 24 Mar 2015 01:38:20 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1YaHXW-0005bx-Le for qemu-devel@nongnu.org; Tue, 24 Mar 2015 01:38:15 -0400 Received: from ozlabs.org ([103.22.144.67]:50825) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YaHXW-0005bS-2X for qemu-devel@nongnu.org; Tue, 24 Mar 2015 01:38:10 -0400 Date: Tue, 24 Mar 2015 16:38:54 +1100 From: David Gibson Message-ID: <20150324053854.GE25043@voom.fritz.box> References: <1424883128-9841-1-git-send-email-dgilbert@redhat.com> <1424883128-9841-41-git-send-email-dgilbert@redhat.com> MIME-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="5FDBdOf8+srCecG/" Content-Disposition: inline In-Reply-To: <1424883128-9841-41-git-send-email-dgilbert@redhat.com> Subject: Re: [Qemu-devel] [PATCH v5 40/45] Postcopy; Handle userfault requests List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: "Dr. David Alan Gilbert (git)" Cc: aarcange@redhat.com, yamahata@private.email.ne.jp, quintela@redhat.com, qemu-devel@nongnu.org, amit.shah@redhat.com, pbonzini@redhat.com, yanghy@cn.fujitsu.com --5FDBdOf8+srCecG/ Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Wed, Feb 25, 2015 at 04:52:03PM +0000, Dr. David Alan Gilbert (git) wrot= e: > From: "Dr. David Alan Gilbert" >=20 > userfaultfd is a Linux syscall that gives an fd that receives a stream > of notifications of accesses to pages registered with it and allows > the program to acknowledge those stalls and tell the accessing > thread to carry on. >=20 > Signed-off-by: Dr. David Alan Gilbert > --- > include/migration/migration.h | 4 + > migration/postcopy-ram.c | 217 ++++++++++++++++++++++++++++++++++++= ++++-- > trace-events | 12 +++ > 3 files changed, 223 insertions(+), 10 deletions(-) >=20 > diff --git a/include/migration/migration.h b/include/migration/migration.h > index 139bb1b..cec064f 100644 > --- a/include/migration/migration.h > +++ b/include/migration/migration.h > @@ -86,11 +86,15 @@ struct MigrationIncomingState { > =20 > PostcopyState postcopy_state; > =20 > + bool have_fault_thread; > QemuThread fault_thread; > QemuSemaphore fault_thread_sem; > =20 > /* For the kernel to send us notifications */ > int userfault_fd; > + /* To tell the fault_thread to quit */ > + int userfault_quit_fd; > + > QEMUFile *return_path; > QemuMutex rp_mutex; /* We send replies from multiple threads= */ > PostcopyPMI postcopy_pmi; > diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c > index 86fa5a0..abc039e 100644 > --- a/migration/postcopy-ram.c > +++ b/migration/postcopy-ram.c > @@ -47,6 +47,8 @@ struct PostcopyDiscardState { > */ > #if defined(__linux__) > =20 > +#include > +#include > #include > #include > #include > @@ -264,7 +266,7 @@ void postcopy_pmi_dump(MigrationIncomingState *mis) > void postcopy_hook_early_receive(MigrationIncomingState *mis, > size_t bitmap_index) > { > - if (mis->postcopy_state =3D=3D POSTCOPY_INCOMING_ADVISE) { > + if (postcopy_state_get(mis) =3D=3D POSTCOPY_INCOMING_ADVISE) { It kind of looks like that's a fix which should be folded into an earlier patch. > /* > * If we're in precopy-advise mode we need to track received pag= es even > * though we don't need to place pages atomically yet. > @@ -489,15 +491,40 @@ int postcopy_ram_incoming_init(MigrationIncomingSta= te *mis, size_t ram_pages) > */ > int postcopy_ram_incoming_cleanup(MigrationIncomingState *mis) > { > - /* TODO: Join the fault thread once we're sure it will exit */ > - if (qemu_ram_foreach_block(cleanup_area, mis)) { > - return -1; > + trace_postcopy_ram_incoming_cleanup_entry(); > + > + if (mis->have_fault_thread) { > + uint64_t tmp64; > + > + if (qemu_ram_foreach_block(cleanup_area, mis)) { > + return -1; > + } > + /* > + * Tell the fault_thread to exit, it's an eventfd that should > + * currently be at 0, we're going to inc it to 1 > + */ > + tmp64 =3D 1; > + if (write(mis->userfault_quit_fd, &tmp64, 8) =3D=3D 8) { > + trace_postcopy_ram_incoming_cleanup_join(); > + qemu_thread_join(&mis->fault_thread); > + } else { > + /* Not much we can do here, but may as well report it */ > + perror("incing userfault_quit_fd"); > + } > + trace_postcopy_ram_incoming_cleanup_closeuf(); > + close(mis->userfault_fd); > + close(mis->userfault_quit_fd); > + mis->have_fault_thread =3D false; > } > =20 > + postcopy_state_set(mis, POSTCOPY_INCOMING_END); > + migrate_send_rp_shut(mis, qemu_file_get_error(mis->file) !=3D 0); > + > if (mis->postcopy_tmp_page) { > munmap(mis->postcopy_tmp_page, getpagesize()); > mis->postcopy_tmp_page =3D NULL; > } > + trace_postcopy_ram_incoming_cleanup_exit(); > return 0; > } > =20 > @@ -531,36 +558,206 @@ static int ram_block_enable_notify(const char *blo= ck_name, void *host_addr, > } > =20 > /* > + * Tell the kernel that we've now got some memory it previously asked fo= r. > + */ > +static int ack_userfault(MigrationIncomingState *mis, void *start, size_= t len) > +{ > + struct uffdio_range range_struct; > + > + range_struct.start =3D (uint64_t)(uintptr_t)start; > + range_struct.len =3D (uint64_t)len; > + > + errno =3D 0; > + if (ioctl(mis->userfault_fd, UFFDIO_WAKE, &range_struct)) { > + int e =3D errno; > + > + if (e =3D=3D ENOENT) { > + /* Kernel said it wasn't waiting - one case where this can > + * happen is where two threads triggered the userfault > + * and we receive the page and ack it just after we received > + * the 2nd request and that ends up deciding it should ack it > + * We could optimise it out, but it's rare. > + */ > + /*fprintf(stderr, "ack_userfault: %p/%zx ENOENT\n", start, l= en); */ > + return 0; > + } > + error_report("postcopy_ram: Failed to notify kernel for %p/%zx (= %d)", > + start, len, e); > + return -e; > + } > + > + return 0; > +} > + > +/* > * Handle faults detected by the USERFAULT markings > */ > static void *postcopy_ram_fault_thread(void *opaque) > { > MigrationIncomingState *mis =3D (MigrationIncomingState *)opaque; > - > - fprintf(stderr, "postcopy_ram_fault_thread\n"); > - /* TODO: In later patch */ > + uint64_t hostaddr; /* The kernel always gives us 64 bit, not a point= er */ > + int ret; > + size_t hostpagesize =3D getpagesize(); > + RAMBlock *rb =3D NULL; > + RAMBlock *last_rb =3D NULL; /* last RAMBlock we sent part of */ > + uint8_t *local_tmp_page; > + > + trace_postcopy_ram_fault_thread_entry(); > qemu_sem_post(&mis->fault_thread_sem); > - while (1) { > - /* TODO: In later patch */ > + > + local_tmp_page =3D mmap(NULL, getpagesize(), > + PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON= YMOUS, > + -1, 0); > + if (!local_tmp_page) { > + perror("mapping local tmp page"); > + return NULL; > } > + if (madvise(local_tmp_page, getpagesize(), MADV_DONTFORK)) { > + munmap(local_tmp_page, getpagesize()); > + perror("postcpy local page DONTFORK"); > + return NULL; > + } > + > + while (true) { > + PostcopyPMIState old_state, tmp_state; > + ram_addr_t rb_offset; > + ram_addr_t in_raspace; > + unsigned long bitmap_index; > + struct pollfd pfd[2]; > + > + /* > + * We're mainly waiting for the kernel to give us a faulting HVA, > + * however we can be told to quit via userfault_quit_fd which is > + * an eventfd > + */ > + pfd[0].fd =3D mis->userfault_fd; > + pfd[0].events =3D POLLIN; > + pfd[0].revents =3D 0; > + pfd[1].fd =3D mis->userfault_quit_fd; > + pfd[1].events =3D POLLIN; /* Waiting for eventfd to go positive = */ > + pfd[1].revents =3D 0; > + > + if (poll(pfd, 2, -1 /* Wait forever */) =3D=3D -1) { > + perror("userfault poll"); > + break; > + } > + > + if (pfd[1].revents) { > + trace_postcopy_ram_fault_thread_quit(); > + break; > + } > + > + ret =3D read(mis->userfault_fd, &hostaddr, sizeof(hostaddr)); > + if (ret !=3D sizeof(hostaddr)) { > + if (ret < 0) { > + perror("Failed to read full userfault hostaddr"); > + break; > + } else { > + error_report("%s: Read %d bytes from userfaultfd expecte= d %zd", > + __func__, ret, sizeof(hostaddr)); > + break; /* Lost alignment, don't know what we'd read next= */ > + } > + } > + > + rb =3D qemu_ram_block_from_host((void *)(uintptr_t)hostaddr, tru= e, > + &in_raspace, &rb_offset, &bitmap_i= ndex); > + if (!rb) { > + error_report("postcopy_ram_fault_thread: Fault outside guest= : %" > + PRIx64, hostaddr); > + break; > + } > =20 > + trace_postcopy_ram_fault_thread_request(hostaddr, bitmap_index, > + qemu_ram_get_idstr(rb), > + rb_offset); > + > + tmp_state =3D postcopy_pmi_get_state(mis, bitmap_index); > + do { > + old_state =3D tmp_state; > + > + switch (old_state) { > + case POSTCOPY_PMI_REQUESTED: > + /* Do nothing - it's already requested */ > + break; > + > + case POSTCOPY_PMI_RECEIVED: > + /* Already arrived - no state change, just kick the kern= el */ > + trace_postcopy_ram_fault_thread_notify_pre(hostaddr); > + if (ack_userfault(mis, > + (void *)((uintptr_t)hostaddr > + & ~(hostpagesize - 1)), > + hostpagesize)) { > + assert(0); > + } > + break; > + > + case POSTCOPY_PMI_MISSING: > + tmp_state =3D postcopy_pmi_change_state(mis, bitmap_inde= x, > + old_state, POSTCOPY_PMI_REQUE= STED); > + if (tmp_state =3D=3D POSTCOPY_PMI_MISSING) { > + /* > + * Send the request to the source - we want to reque= st one > + * of our host page sizes (which is >=3D TPS) > + */ > + if (rb !=3D last_rb) { > + last_rb =3D rb; > + migrate_send_rp_req_pages(mis, qemu_ram_get_idst= r(rb), > + rb_offset, hostpagesize= ); > + } else { > + /* Save some space */ > + migrate_send_rp_req_pages(mis, NULL, > + rb_offset, hostpagesize= ); > + } > + } /* else it just arrived from the source and the kernel= will > + be kicked during the receive */ > + break; > + } > + } while (tmp_state !=3D old_state); Again, I think using separate requested/received bits rather than treating them as a single state could avoid this clunky loop. > + } > + munmap(local_tmp_page, getpagesize()); > + trace_postcopy_ram_fault_thread_exit(); > return NULL; > } > =20 > int postcopy_ram_enable_notify(MigrationIncomingState *mis) > { > - /* Create the fault handler thread and wait for it to be ready */ > + /* Open the fd for the kernel to give us userfaults */ > + mis->userfault_fd =3D syscall(__NR_userfaultfd, O_CLOEXEC); I think it would be good to declare your own userfaultfd() wrappers around syscall(). That way it will be easier to clean up once libc knows about them. > + if (mis->userfault_fd =3D=3D -1) { > + perror("Failed to open userfault fd"); > + return -1; > + } > + > + /* > + * Although the host check already tested the API, we need to > + * do the check again as an ABI handshake on the new fd. > + */ > + if (!ufd_version_check(mis->userfault_fd)) { > + return -1; > + } > + > + /* Now an eventfd we use to tell the fault-thread to quit */ > + mis->userfault_quit_fd =3D eventfd(0, EFD_CLOEXEC); > + if (mis->userfault_quit_fd =3D=3D -1) { > + perror("Opening userfault_quit_fd"); > + close(mis->userfault_fd); > + return -1; > + } > + > qemu_sem_init(&mis->fault_thread_sem, 0); > qemu_thread_create(&mis->fault_thread, "postcopy/fault", > postcopy_ram_fault_thread, mis, QEMU_THREAD_JOINA= BLE); > qemu_sem_wait(&mis->fault_thread_sem); > qemu_sem_destroy(&mis->fault_thread_sem); > + mis->have_fault_thread =3D true; > =20 > /* Mark so that we get notified of accesses to unwritten areas */ > if (qemu_ram_foreach_block(ram_block_enable_notify, mis)) { > return -1; > } > =20 > + trace_postcopy_ram_enable_notify(); > + > return 0; > } > =20 > diff --git a/trace-events b/trace-events > index 16a91d9..d955a28 100644 > --- a/trace-events > +++ b/trace-events > @@ -1498,6 +1498,18 @@ postcopy_discard_send_finish(const char *ramblock,= int nwords, int ncmds) "%s ma > postcopy_cleanup_area(const char *ramblock, void *host_addr, size_t offs= et, size_t length) "%s: %p offset=3D%zx length=3D%zx" > postcopy_init_area(const char *ramblock, void *host_addr, size_t offset,= size_t length) "%s: %p offset=3D%zx length=3D%zx" > postcopy_place_page(unsigned long offset, void *host_addr, bool all_zero= , int old_state) "offset=3D%lx host=3D%p all_zero=3D%d old_state=3D%d" > +postcopy_ram_enable_notify(void) "" > +postcopy_ram_fault_thread_entry(void) "" > +postcopy_ram_fault_thread_exit(void) "" > +postcopy_ram_fault_thread_quit(void) "" > +postcopy_ram_fault_thread_request(uint64_t hostaddr, unsigned long index= , const char *ramblock, size_t offset) "Request for HVA=3D%" PRIx64 " index= =3D%lx rb=3D%s offset=3D%zx" > +postcopy_ram_fault_thread_notify_pre(uint64_t hostaddr) "%" PRIx64 > +postcopy_ram_fault_thread_notify_zero(void *hostaddr) "%p" > +postcopy_ram_fault_thread_notify_zero_ack(void *hostaddr, unsigned long = bitmap_index) "%p %lx" > +postcopy_ram_incoming_cleanup_closeuf(void) "" > +postcopy_ram_incoming_cleanup_entry(void) "" > +postcopy_ram_incoming_cleanup_exit(void) "" > +postcopy_ram_incoming_cleanup_join(void) "" > =20 > # kvm-all.c > kvm_ioctl(int type, void *arg) "type 0x%x, arg %p" --=20 David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson --5FDBdOf8+srCecG/ Content-Type: application/pgp-signature -----BEGIN PGP SIGNATURE----- Version: GnuPG v1 iQIcBAEBAgAGBQJVEPhuAAoJEGw4ysog2bOSkQ4P/Ao/4QQWbRDhQWmlY8DUS1Vb LvyeBeTR8rOtKkpYCdBXqwEjIO+/Cb/2str6q9s9YyZVn99DtaTghOcngTzkl1Xu uCUNUc9vdziV0DUocPgGXAEZCJEqHmnFtPbhry7JlNAXPMD0WiK4UlGgMPnrNnww fsck5988m+fz5WDTxZDgV/r0S/crvxdQbf3Z8jecOeD5/0WuN7rT26VQKHchjgXY r7VGQxqqtAyJjPMCe0zReb4jTrS1trrt77EhA+8V4PQhDU4noVabeeVXCFPorXrm rKAvtepVe8OscLi19OQNdFQgDCEL41T+xHlM9FqYAncQ5bmC9ySVbcPShZk4Ev4s rW1IWV/AQkKS7LEZnUdzuCwrCEh3dwIYpDPylYLifSpF08YiRi3ENumjD/nRhXox mbBv00YQHM35fgXYeWpMNM9Gi0tqriVzpfS94tNGW9AFt5+V9HY5LPZuvZgXUIn5 ReQKqJ5kAM+CX0txE/zMb68Cq5hgxOp6k+EyiZ1a7e0okGZgtZgak2ipsqpmji3X V2XKDmeAqle1TV//U8e/v7a56dsEcw0kC+cjMYdToRIq0YU/4Q5tZnP8dXoK8xKy JWhIVY/fm72yk0JPDOkoAnzQFMktmOwd4QVylWL/Cdz4N9+JsNhi4ha02RXT1+MS W97OxlMG7WNR0fzVLqEN =pPzd -----END PGP SIGNATURE----- --5FDBdOf8+srCecG/--