On Mon Mar 31, 2025 at 11:25 PM AEST, Corey Minyard wrote:
> On Mon, Mar 31, 2025 at 10:57:24PM +1000, Nicholas Piggin wrote:
>> Linux issues this command when booting a powernv machine.
>
> This is good, just a couple of nits.
>
>>
>> Signed-off-by: Nicholas Piggin <[email protected]>
>> ---
>> include/hw/ipmi/ipmi.h | 14 +++++++++++
>> hw/ipmi/ipmi_bmc_sim.c | 56 ++++++++++++++++++++++++++++++++++++++++--
>> hw/ipmi/ipmi_bt.c | 2 ++
>> hw/ipmi/ipmi_kcs.c | 1 +
>> 4 files changed, 71 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/hw/ipmi/ipmi.h b/include/hw/ipmi/ipmi.h
>> index 77a7213ed93..5f01a50cd86 100644
>> --- a/include/hw/ipmi/ipmi.h
>> +++ b/include/hw/ipmi/ipmi.h
>> @@ -41,6 +41,15 @@ enum ipmi_op {
>> IPMI_SEND_NMI
>> };
>>
>> +/* Channel properties */
>> +#define IPMI_CHANNEL_IPMB 0x00
>> +#define IPMI_CHANNEL_SYSTEM 0x0f
>> +#define IPMI_CH_MEDIUM_IPMB 0x01
>> +#define IPMI_CH_MEDIUM_SYSTEM 0x0c
>> +#define IPMI_CH_PROTOCOL_IPMB 0x01
>> +#define IPMI_CH_PROTOCOL_KCS 0x05
>> +#define IPMI_CH_PROTOCOL_BT_15 0x08
>
> I know it's picky, but could you spell out CHANNEL here?
Sure.
>> +
>> #define IPMI_CC_INVALID_CMD 0xc1
>> #define IPMI_CC_COMMAND_INVALID_FOR_LUN 0xc2
>> #define IPMI_CC_TIMEOUT 0xc3
>> @@ -170,6 +179,11 @@ struct IPMIInterfaceClass {
>> * Return the firmware info for a device.
>> */
>> void (*get_fwinfo)(struct IPMIInterface *s, IPMIFwInfo *info);
>> +
>> + /*
>> + * IPMI channel protocol type number.
>> + */
>> + uint8_t protocol;
>> };
>>
>> /*
>> diff --git a/hw/ipmi/ipmi_bmc_sim.c b/hw/ipmi/ipmi_bmc_sim.c
>> index 8c3313aa65f..9198f854bd9 100644
>> --- a/hw/ipmi/ipmi_bmc_sim.c
>> +++ b/hw/ipmi/ipmi_bmc_sim.c
>> @@ -70,6 +70,7 @@
>> #define IPMI_CMD_GET_MSG 0x33
>> #define IPMI_CMD_SEND_MSG 0x34
>> #define IPMI_CMD_READ_EVT_MSG_BUF 0x35
>> +#define IPMI_CMD_GET_CHANNEL_INFO 0x42
>>
>> #define IPMI_NETFN_STORAGE 0x0a
>>
>> @@ -1033,8 +1034,8 @@ static void send_msg(IPMIBmcSim *ibs,
>> uint8_t *buf;
>> uint8_t netfn, rqLun, rsLun, rqSeq;
>>
>> - if (cmd[2] != 0) {
>> - /* We only handle channel 0 with no options */
>> + if (cmd[2] != IPMI_CHANNEL_IPMB) {
>> + /* We only handle channel 0h (IPMB) with no options */
>> rsp_buffer_set_error(rsp, IPMI_CC_INVALID_DATA_FIELD);
>> return;
>> }
>> @@ -1232,6 +1233,56 @@ static void get_watchdog_timer(IPMIBmcSim *ibs,
>> }
>> }
>>
>> +static void get_channel_info(IPMIBmcSim *ibs,
>> + uint8_t *cmd, unsigned int cmd_len,
>> + RspBuffer *rsp)
>> +{
>> + IPMIInterface *s = ibs->parent.intf;
>> + IPMIInterfaceClass *k = IPMI_INTERFACE_GET_CLASS(s);
>> + uint8_t ch = cmd[1] & 0x0f;
>> +
>> + /* Only define channel 0h (IPMB) and Fh (system interface) */
>> +
>> + if (ch == 0x0e) { /* "This channel" */
>> + ch = IPMI_CHANNEL_SYSTEM;
>> + }
>> + rsp_buffer_push(rsp, ch);
>> +
>> + if (ch != IPMI_CHANNEL_IPMB && ch != IPMI_CHANNEL_SYSTEM) {
>> + /* Not supported */
>
> I think that an all zero response is a valid response. I think you
> should return a IPMI_CC_INVALID_DATA_FIELD instead, right?
I can't remember, I dug the patch out from a while ago. I can't actually
see anywhere it is made clear in the spec, do you? I agree an invalid
error sounds better. I'll try to see how ipmi tools handles it.
>> + int i;
>> + for (i = 0; i < 8; i++) {
>> + rsp_buffer_push(rsp, 0x00);
>> + }
>> + return;
>> + }
>> +
>> + if (ch == IPMI_CHANNEL_IPMB) {
>> + rsp_buffer_push(rsp, IPMI_CH_MEDIUM_IPMB);
>> + rsp_buffer_push(rsp, IPMI_CH_PROTOCOL_IPMB);
>> + } else { /* IPMI_CHANNEL_SYSTEM */
>> + rsp_buffer_push(rsp, IPMI_CH_MEDIUM_SYSTEM);
>> + rsp_buffer_push(rsp, k->protocol);
>> + }
>> +
>> + rsp_buffer_push(rsp, 0x00); /* Session-less */
>> +
>> + /* IPMI Vendor ID */
>> + rsp_buffer_push(rsp, 0xf2);
>> + rsp_buffer_push(rsp, 0x1b);
>> + rsp_buffer_push(rsp, 0x00);
>
> Where does this come from?
IPMI spec Get Channel Info Command, search "IPMI Enterprise Number"
>From my reading, all channel info responses contain this.
>> +
>> + if (ch == IPMI_CHANNEL_SYSTEM) {
>> + /* IRQ assigned by ACPI/PnP (XXX?) */
>> + rsp_buffer_push(rsp, 0x60);
>> + rsp_buffer_push(rsp, 0x60);
>
> The interrupt should be available. For the isa versions there is a
> get_fwinfo function pointer that you can fetch this with. For PCI it's
> more complicated, unfortunately.
These are for the two interrupts. QEMU seems to tie both to the
same line, I guess that's okay?
That interface looks good, but what I was concerned about is whether
that implies the irq is hard coded or whether the platform can assign
it, does it mean unassigned? I don't know a lot about irq routing or
what IPMI clients would use it for.
Anyhow I'll respin with changes.
Thanks,
Nick
