| Message ID | 1438334990-11765-1-git-send-email-ldufour@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
From: Laurent Dufour > Sent: 31 July 2015 10:30 > This patch fixes several endianness issues detected when running the HVSI > driver in little endian mode. > > These issues are raised in little endian mode because the data exchanged in > memory between the kernel and the hypervisor has to be in big endian > format. ... > diff --git a/drivers/tty/hvc/hvsi.c b/drivers/tty/hvc/hvsi.c > index 41901997c0d6..a75146f600cb 100644 > --- a/drivers/tty/hvc/hvsi.c > +++ b/drivers/tty/hvc/hvsi.c > @@ -240,9 +240,9 @@ static void hvsi_recv_control(struct hvsi_struct *hp, uint8_t *packet, > { > struct hvsi_control *header = (struct hvsi_control *)packet; > > - switch (header->verb) { > + switch (be16_to_cpu(header->verb)) { > case VSV_MODEM_CTL_UPDATE: > - if ((header->word & HVSI_TSCD) == 0) { > + if ((be32_to_cpu(header->word) & HVSI_TSCD) == 0) { It is generally best to byteswap constants. David
On Fri, 2015-07-31 at 11:29 +0200, Laurent Dufour wrote: > This patch fixes several endianness issues detected when running the HVSI > driver in little endian mode. > > These issues are raised in little endian mode because the data exchanged in > memory between the kernel and the hypervisor has to be in big endian > format. Can you include the sparse output before and after? cheers
On 03/08/2015 13:00, David Laight wrote: > From: Laurent Dufour >> Sent: 31 July 2015 10:30 >> This patch fixes several endianness issues detected when running the HVSI >> driver in little endian mode. >> >> These issues are raised in little endian mode because the data exchanged in >> memory between the kernel and the hypervisor has to be in big endian >> format. > ... >> diff --git a/drivers/tty/hvc/hvsi.c b/drivers/tty/hvc/hvsi.c >> index 41901997c0d6..a75146f600cb 100644 >> --- a/drivers/tty/hvc/hvsi.c >> +++ b/drivers/tty/hvc/hvsi.c >> @@ -240,9 +240,9 @@ static void hvsi_recv_control(struct hvsi_struct *hp, uint8_t *packet, >> { >> struct hvsi_control *header = (struct hvsi_control *)packet; >> >> - switch (header->verb) { >> + switch (be16_to_cpu(header->verb)) { >> case VSV_MODEM_CTL_UPDATE: >> - if ((header->word & HVSI_TSCD) == 0) { >> + if ((be32_to_cpu(header->word) & HVSI_TSCD) == 0) { > > It is generally best to byteswap constants. > > David Thanks David for your review. Regarding the byte swapping of the constants, I'm wondering if this the best way here. For instance, Benjamin wrote a similar patch to fix another endianness issue (99fc1d91b8fc) and he doesn't convert the constant neither. It think that byte swapping the constant value will impact more code, and may not ease code reading. Cheers, Laurent.
On Wed, 2015-08-19 at 23:37 +0200, Laurent Dufour wrote: > On 03/08/2015 13:00, David Laight wrote: From: Laurent Dufour > > > Sent: 31 July 2015 10:30 > > > This patch fixes several endianness issues detected when running > > > the HVSI > > > driver in little endian mode. > > > > > > These issues are raised in little endian mode because the data > > > exchanged in > > > memory between the kernel and the hypervisor has to be in big > > > endian > > > format. > > ... > > > diff --git a/drivers/tty/hvc/hvsi.c b/drivers/tty/hvc/hvsi.c > > > index 41901997c0d6..a75146f600cb 100644 > > > --- a/drivers/tty/hvc/hvsi.c > > > +++ b/drivers/tty/hvc/hvsi.c > > > @@ -240,9 +240,9 @@ static void hvsi_recv_control(struct > > > hvsi_struct *hp, uint8_t *packet, > > > { > > > struct hvsi_control *header = (struct hvsi_control > > > *)packet; > > > > > > - switch (header->verb) { > > > + switch (be16_to_cpu(header->verb)) { > > > case VSV_MODEM_CTL_UPDATE: > > > - if ((header->word & HVSI_TSCD) == 0) { > > > + if ((be32_to_cpu(header->word) & > > > HVSI_TSCD) == 0) { > > > > It is generally best to byteswap constants. > > > > David > > Thanks David for your review. > Regarding the byte swapping of the constants, I'm wondering if this > the > best way here. > For instance, Benjamin wrote a similar patch to fix another > endianness > issue (99fc1d91b8fc) and he doesn't convert the constant neither. > It think that byte swapping the constant value will impact more code, > and may not ease code reading. Right, I disagree with byteswapping the constants at their definition point, however maybe he meant using cpu_to_be16(CONSTANT) ? In any case, it's pretty moot as we have the lhbrx instruction which will do the load and byteswap and for HVSI, even if it was a tad slower than a normal load, it would not make a noticeable difference. Ben.
On 04/08/2015 02:51, Michael Ellerman wrote: > On Fri, 2015-07-31 at 11:29 +0200, Laurent Dufour wrote: >> This patch fixes several endianness issues detected when running the HVSI >> driver in little endian mode. >> >> These issues are raised in little endian mode because the data exchanged in >> memory between the kernel and the hypervisor has to be in big endian >> format. > > Can you include the sparse output before and after? Hi Michael, Here is the output message displayed on the console when the bug occurred: [ 0.000517] irq: (null) didn't like hwirq-0x1000a00 to VIRQ16 mapping (rc=-22) [ 0.000578] hvsi_console_init: couldn't create irq mapping for 0x1000a00 With the patch is applied, the hvsi driver is initializing correctly and no message is displayed, except the one saying the number of device the hvsi driver has configured. For instance: [ 1.535783] HVSI: registered 1 devices Cheers, Laurent.
On Wed, 2015-08-19 at 23:53 +0200, Laurent Dufour wrote: > On 04/08/2015 02:51, Michael Ellerman wrote: > > On Fri, 2015-07-31 at 11:29 +0200, Laurent Dufour wrote: > >> This patch fixes several endianness issues detected when running the HVSI > >> driver in little endian mode. > >> > >> These issues are raised in little endian mode because the data exchanged in > >> memory between the kernel and the hypervisor has to be in big endian > >> format. > > > > Can you include the sparse output before and after? > > Hi Michael, > > Here is the output message displayed on the console when the bug occurred: > > [ 0.000517] irq: (null) didn't like hwirq-0x1000a00 to VIRQ16 mapping (rc=-22) > [ 0.000578] hvsi_console_init: couldn't create irq mapping for 0x1000a00 > > With the patch is applied, the hvsi driver is initializing correctly and > no message is displayed, except the one saying the number of device the > hvsi driver has configured. For instance: > [ 1.535783] HVSI: registered 1 devices OK that's also good. I was talking about the output from sparse endian checking, as the driver seems to already be endian annotated. ie, before: $ make C=2 CF=-D__CHECK_ENDIAN__ ... drivers/tty/hvc/hvsi.c:245:36: warning: restricted __be32 degrades to integer drivers/tty/hvc/hvsi.c:243:23: warning: restricted __be16 degrades to integer drivers/tty/hvc/hvsi.c:243:23: warning: restricted __be16 degrades to integer drivers/tty/hvc/hvsi.c:277:36: warning: restricted __be32 degrades to integer drivers/tty/hvc/hvsi.c:279:36: warning: restricted __be32 degrades to integer drivers/tty/hvc/hvsi.c:298:26: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:298:26: expected restricted __be16 [assigned] [usertype] seqno drivers/tty/hvc/hvsi.c:298:26: got int drivers/tty/hvc/hvsi.c:299:21: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:299:21: expected restricted __be16 [assigned] [usertype] verb drivers/tty/hvc/hvsi.c:299:21: got int drivers/tty/hvc/hvsi.c:301:28: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:301:28: expected restricted __be16 [assigned] [usertype] query_seqno drivers/tty/hvc/hvsi.c:301:28: got int drivers/tty/hvc/hvsi.c:322:60: warning: incorrect type in argument 2 (different base types) drivers/tty/hvc/hvsi.c:322:60: expected unsigned short [unsigned] [usertype] query_seqno drivers/tty/hvc/hvsi.c:322:60: got restricted __be16 [usertype] seqno drivers/tty/hvc/hvsi.c:558:26: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:558:26: expected restricted __be16 [assigned] [usertype] seqno drivers/tty/hvc/hvsi.c:558:26: got int drivers/tty/hvc/hvsi.c:559:21: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:559:21: expected restricted __be16 [assigned] [usertype] verb drivers/tty/hvc/hvsi.c:559:21: got unsigned short [unsigned] [usertype] verb drivers/tty/hvc/hvsi.c:600:26: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:600:26: expected restricted __be16 [assigned] [usertype] seqno drivers/tty/hvc/hvsi.c:600:26: got int drivers/tty/hvc/hvsi.c:602:21: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:602:21: expected restricted __be16 [assigned] [usertype] verb drivers/tty/hvc/hvsi.c:602:21: got int drivers/tty/hvc/hvsi.c:603:21: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:603:21: expected restricted __be32 [assigned] [usertype] mask drivers/tty/hvc/hvsi.c:603:21: got int drivers/tty/hvc/hvsi.c:606:29: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:606:29: expected restricted __be32 [assigned] [usertype] word drivers/tty/hvc/hvsi.c:606:29: got int drivers/tty/hvc/hvsi.c:683:26: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:683:26: expected restricted __be16 [assigned] [usertype] seqno drivers/tty/hvc/hvsi.c:683:26: got int drivers/tty/hvc/hvsi.c:700:26: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:700:26: expected restricted __be16 [assigned] [usertype] seqno drivers/tty/hvc/hvsi.c:700:26: got int drivers/tty/hvc/hvsi.c:702:21: warning: incorrect type in assignment (different base types) drivers/tty/hvc/hvsi.c:702:21: expected restricted __be16 [assigned] [usertype] verb drivers/tty/hvc/hvsi.c:702:21: got int ... And with your patch applied there are no warnings from hvsi.c! So it seems you fixed all the issues, or at least all the issues we can detect with sparse. So I'll merge this as-is with an updated changelog. cheers
On Thu, 2015-08-20 at 11:40 +1000, Michael Ellerman wrote: > On Wed, 2015-08-19 at 23:53 +0200, Laurent Dufour wrote: > > On 04/08/2015 02:51, Michael Ellerman wrote: > > > On Fri, 2015-07-31 at 11:29 +0200, Laurent Dufour wrote: > > >> This patch fixes several endianness issues detected when running the HVSI > > >> driver in little endian mode. > > >> > > >> These issues are raised in little endian mode because the data exchanged in > > >> memory between the kernel and the hypervisor has to be in big endian > > >> format. ... > > So it seems you fixed all the issues, or at least all the issues we can detect > with sparse. So I'll merge this as-is with an updated changelog. Unless Greg wants to merge it? But I'll assume he's not bothered unless he says otherwise as it's a powerpc only driver. cheers
On Fri, 2015-31-07 at 09:29:50 UTC, Laurent Dufour wrote: > This patch fixes several endianness issues detected when running the HVSI > driver in little endian mode. > > These issues are raised in little endian mode because the data exchanged in > memory between the kernel and the hypervisor has to be in big endian > format. > > Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com> > CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org> > CC: Jiri Slaby <jslaby@suse.cz> > CC: linuxppc-dev@lists.ozlabs.org > CC: linux-kernel@vger.kernel.org Applied to powerpc next, thanks. https://git.kernel.org/powerpc/c/480798044eb268a31f6b cheers
diff --git a/drivers/tty/hvc/hvsi.c b/drivers/tty/hvc/hvsi.c index 41901997c0d6..a75146f600cb 100644 --- a/drivers/tty/hvc/hvsi.c +++ b/drivers/tty/hvc/hvsi.c @@ -240,9 +240,9 @@ static void hvsi_recv_control(struct hvsi_struct *hp, uint8_t *packet, { struct hvsi_control *header = (struct hvsi_control *)packet; - switch (header->verb) { + switch (be16_to_cpu(header->verb)) { case VSV_MODEM_CTL_UPDATE: - if ((header->word & HVSI_TSCD) == 0) { + if ((be32_to_cpu(header->word) & HVSI_TSCD) == 0) { /* CD went away; no more connection */ pr_debug("hvsi%i: CD dropped\n", hp->index); hp->mctrl &= TIOCM_CD; @@ -267,6 +267,7 @@ static void hvsi_recv_control(struct hvsi_struct *hp, uint8_t *packet, static void hvsi_recv_response(struct hvsi_struct *hp, uint8_t *packet) { struct hvsi_query_response *resp = (struct hvsi_query_response *)packet; + uint32_t mctrl_word; switch (hp->state) { case HVSI_WAIT_FOR_VER_RESPONSE: @@ -274,9 +275,10 @@ static void hvsi_recv_response(struct hvsi_struct *hp, uint8_t *packet) break; case HVSI_WAIT_FOR_MCTRL_RESPONSE: hp->mctrl = 0; - if (resp->u.mctrl_word & HVSI_TSDTR) + mctrl_word = be32_to_cpu(resp->u.mctrl_word); + if (mctrl_word & HVSI_TSDTR) hp->mctrl |= TIOCM_DTR; - if (resp->u.mctrl_word & HVSI_TSCD) + if (mctrl_word & HVSI_TSCD) hp->mctrl |= TIOCM_CD; __set_state(hp, HVSI_OPEN); break; @@ -295,10 +297,10 @@ static int hvsi_version_respond(struct hvsi_struct *hp, uint16_t query_seqno) packet.hdr.type = VS_QUERY_RESPONSE_PACKET_HEADER; packet.hdr.len = sizeof(struct hvsi_query_response); - packet.hdr.seqno = atomic_inc_return(&hp->seqno); - packet.verb = VSV_SEND_VERSION_NUMBER; + packet.hdr.seqno = cpu_to_be16(atomic_inc_return(&hp->seqno)); + packet.verb = cpu_to_be16(VSV_SEND_VERSION_NUMBER); packet.u.version = HVSI_VERSION; - packet.query_seqno = query_seqno+1; + packet.query_seqno = cpu_to_be16(query_seqno+1); pr_debug("%s: sending %i bytes\n", __func__, packet.hdr.len); dbg_dump_hex((uint8_t*)&packet, packet.hdr.len); @@ -319,7 +321,7 @@ static void hvsi_recv_query(struct hvsi_struct *hp, uint8_t *packet) switch (hp->state) { case HVSI_WAIT_FOR_VER_QUERY: - hvsi_version_respond(hp, query->hdr.seqno); + hvsi_version_respond(hp, be16_to_cpu(query->hdr.seqno)); __set_state(hp, HVSI_OPEN); break; default: @@ -555,8 +557,8 @@ static int hvsi_query(struct hvsi_struct *hp, uint16_t verb) packet.hdr.type = VS_QUERY_PACKET_HEADER; packet.hdr.len = sizeof(struct hvsi_query); - packet.hdr.seqno = atomic_inc_return(&hp->seqno); - packet.verb = verb; + packet.hdr.seqno = cpu_to_be16(atomic_inc_return(&hp->seqno)); + packet.verb = cpu_to_be16(verb); pr_debug("%s: sending %i bytes\n", __func__, packet.hdr.len); dbg_dump_hex((uint8_t*)&packet, packet.hdr.len); @@ -596,14 +598,14 @@ static int hvsi_set_mctrl(struct hvsi_struct *hp, uint16_t mctrl) struct hvsi_control packet __ALIGNED__; int wrote; - packet.hdr.type = VS_CONTROL_PACKET_HEADER, - packet.hdr.seqno = atomic_inc_return(&hp->seqno); + packet.hdr.type = VS_CONTROL_PACKET_HEADER; + packet.hdr.seqno = cpu_to_be16(atomic_inc_return(&hp->seqno)); packet.hdr.len = sizeof(struct hvsi_control); - packet.verb = VSV_SET_MODEM_CTL; - packet.mask = HVSI_TSDTR; + packet.verb = cpu_to_be16(VSV_SET_MODEM_CTL); + packet.mask = cpu_to_be32(HVSI_TSDTR); if (mctrl & TIOCM_DTR) - packet.word = HVSI_TSDTR; + packet.word = cpu_to_be32(HVSI_TSDTR); pr_debug("%s: sending %i bytes\n", __func__, packet.hdr.len); dbg_dump_hex((uint8_t*)&packet, packet.hdr.len); @@ -680,7 +682,7 @@ static int hvsi_put_chars(struct hvsi_struct *hp, const char *buf, int count) BUG_ON(count > HVSI_MAX_OUTGOING_DATA); packet.hdr.type = VS_DATA_PACKET_HEADER; - packet.hdr.seqno = atomic_inc_return(&hp->seqno); + packet.hdr.seqno = cpu_to_be16(atomic_inc_return(&hp->seqno)); packet.hdr.len = count + sizeof(struct hvsi_header); memcpy(&packet.data, buf, count); @@ -697,9 +699,9 @@ static void hvsi_close_protocol(struct hvsi_struct *hp) struct hvsi_control packet __ALIGNED__; packet.hdr.type = VS_CONTROL_PACKET_HEADER; - packet.hdr.seqno = atomic_inc_return(&hp->seqno); + packet.hdr.seqno = cpu_to_be16(atomic_inc_return(&hp->seqno)); packet.hdr.len = 6; - packet.verb = VSV_CLOSE_PROTOCOL; + packet.verb = cpu_to_be16(VSV_CLOSE_PROTOCOL); pr_debug("%s: sending %i bytes\n", __func__, packet.hdr.len); dbg_dump_hex((uint8_t*)&packet, packet.hdr.len); @@ -1180,7 +1182,7 @@ static int __init hvsi_console_init(void) /* search device tree for vty nodes */ for_each_compatible_node(vty, "serial", "hvterm-protocol") { struct hvsi_struct *hp; - const uint32_t *vtermno, *irq; + const __be32 *vtermno, *irq; vtermno = of_get_property(vty, "reg", NULL); irq = of_get_property(vty, "interrupts", NULL); @@ -1202,11 +1204,11 @@ static int __init hvsi_console_init(void) hp->index = hvsi_count; hp->inbuf_end = hp->inbuf; hp->state = HVSI_CLOSED; - hp->vtermno = *vtermno; - hp->virq = irq_create_mapping(NULL, irq[0]); + hp->vtermno = be32_to_cpup(vtermno); + hp->virq = irq_create_mapping(NULL, be32_to_cpup(irq)); if (hp->virq == 0) { printk(KERN_ERR "%s: couldn't create irq mapping for 0x%x\n", - __func__, irq[0]); + __func__, be32_to_cpup(irq)); tty_port_destroy(&hp->port); continue; }
This patch fixes several endianness issues detected when running the HVSI driver in little endian mode. These issues are raised in little endian mode because the data exchanged in memory between the kernel and the hypervisor has to be in big endian format. Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com> CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org> CC: Jiri Slaby <jslaby@suse.cz> CC: linuxppc-dev@lists.ozlabs.org CC: linux-kernel@vger.kernel.org --- drivers/tty/hvc/hvsi.c | 46 ++++++++++++++++++++++++---------------------- 1 file changed, 24 insertions(+), 22 deletions(-)