From: Laurent Vivier <laurent@vivier.eu>
To: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>,
qemu-devel@nongnu.org, pbonzini@redhat.com, fam@euphon.net
Subject: Re: [PATCH v2 33/42] esp: defer command completion interrupt on incoming data transfers
Date: Wed, 3 Mar 2021 20:52:34 +0100 [thread overview]
Message-ID: <ba080d8e-1e21-e5e9-8fff-568db5f73b3f@vivier.eu> (raw)
In-Reply-To: <6cc4e1c8-5a3b-2d2f-5a64-5d2bf3cb3a69@ilande.co.uk>
Le 18/02/2021 à 18:25, Mark Cave-Ayland a écrit :
> On 09/02/2021 19:30, Mark Cave-Ayland wrote:
>
>> The MacOS toolbox ROM issues a command to the ESP controller as part of its
>> "FAST" SCSI routines and then proceeds to read the incoming data soon after
>> receiving the command completion interrupt.
>>
>> Unfortunately due to SCSI block transfers being asynchronous the incoming data
>> may not yet be present causing an underflow error. Resolve this by waiting for
>> the SCSI subsystem transfer_data callback before raising the command completion
>> interrupt.
>>
>> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
>> ---
>> hw/scsi/esp.c | 54 +++++++++++++++++++++++++++++++++++++++----
>> include/hw/scsi/esp.h | 1 +
>> 2 files changed, 51 insertions(+), 4 deletions(-)
>>
>> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
>> index 728d4ddf99..ce6a7a1ed0 100644
>> --- a/hw/scsi/esp.c
>> +++ b/hw/scsi/esp.c
>> @@ -183,6 +183,14 @@ static int esp_select(ESPState *s)
>> esp_raise_irq(s);
>> return -1;
>> }
>> +
>> + /*
>> + * Note that we deliberately don't raise the IRQ here: this will be done
>> + * either in do_busid_cmd() for DATA OUT transfers or by the deferred
>> + * IRQ mechanism in esp_transfer_data() for DATA IN transfers
>> + */
>> + s->rregs[ESP_RINTR] |= INTR_FC;
>> + s->rregs[ESP_RSEQ] = SEQ_CD;
>> return 0;
>> }
>> @@ -237,18 +245,24 @@ static void do_busid_cmd(ESPState *s, uint8_t *buf, uint8_t busid)
>> s->ti_size = datalen;
>> if (datalen != 0) {
>> s->rregs[ESP_RSTAT] = STAT_TC;
>> + s->rregs[ESP_RSEQ] = SEQ_CD;
>> esp_set_tc(s, 0);
>> if (datalen > 0) {
>> + /*
>> + * Switch to DATA IN phase but wait until initial data xfer is
>> + * complete before raising the command completion interrupt
>> + */
>> + s->data_in_ready = false;
>> s->rregs[ESP_RSTAT] |= STAT_DI;
>> } else {
>> s->rregs[ESP_RSTAT] |= STAT_DO;
>> + s->rregs[ESP_RINTR] |= INTR_BS | INTR_FC;
>> + esp_raise_irq(s);
>> + esp_lower_drq(s);
>> }
>> scsi_req_continue(s->current_req);
>> + return;
>> }
>> - s->rregs[ESP_RINTR] |= INTR_BS | INTR_FC;
>> - s->rregs[ESP_RSEQ] = SEQ_CD;
>> - esp_raise_irq(s);
>> - esp_lower_drq(s);
>> }
>> static void do_cmd(ESPState *s)
>> @@ -603,12 +617,35 @@ void esp_command_complete(SCSIRequest *req, uint32_t status,
>> void esp_transfer_data(SCSIRequest *req, uint32_t len)
>> {
>> ESPState *s = req->hba_private;
>> + int to_device = ((s->rregs[ESP_RSTAT] & 7) == STAT_DO);
>> uint32_t dmalen = esp_get_tc(s);
>> assert(!s->do_cmd);
>> trace_esp_transfer_data(dmalen, s->ti_size);
>> s->async_len = len;
>> s->async_buf = scsi_req_get_buf(req);
>> +
>> + if (!to_device && !s->data_in_ready) {
>> + /*
>> + * Initial incoming data xfer is complete so raise command
>> + * completion interrupt
>> + */
>> + s->data_in_ready = true;
>> + s->rregs[ESP_RSTAT] |= STAT_TC;
>> + s->rregs[ESP_RINTR] |= INTR_BS;
>> + esp_raise_irq(s);
>> +
>> + /*
>> + * If data is ready to transfer and the TI command has already
>> + * been executed, start DMA immediately. Otherwise DMA will start
>> + * when host sends the TI command
>> + */
>> + if (s->ti_size && (s->rregs[ESP_CMD] == (CMD_TI | CMD_DMA))) {
>> + esp_do_dma(s);
>> + }
>> + return;
>> + }
>> +
>> if (dmalen) {
>> esp_do_dma(s);
>> } else if (s->ti_size <= 0) {
>> @@ -870,6 +907,14 @@ static bool esp_is_before_version_5(void *opaque, int version_id)
>> return version_id < 5;
>> }
>> +static bool esp_is_version_5(void *opaque, int version_id)
>> +{
>> + ESPState *s = ESP(opaque);
>> +
>> + version_id = MIN(version_id, s->mig_version_id);
>> + return version_id == 5;
>> +}
>> +
>> static int esp_pre_save(void *opaque)
>> {
>> ESPState *s = ESP(opaque);
>> @@ -914,6 +959,7 @@ const VMStateDescription vmstate_esp = {
>> VMSTATE_UINT32(cmdlen, ESPState),
>> VMSTATE_UINT32(do_cmd, ESPState),
>> VMSTATE_UINT32_TEST(mig_dma_left, ESPState, esp_is_before_version_5),
>> + VMSTATE_BOOL_TEST(data_in_ready, ESPState, esp_is_version_5),
>> VMSTATE_END_OF_LIST()
>> },
>> };
>> diff --git a/include/hw/scsi/esp.h b/include/hw/scsi/esp.h
>> index 6618f4e091..3b69aedebe 100644
>> --- a/include/hw/scsi/esp.h
>> +++ b/include/hw/scsi/esp.h
>> @@ -41,6 +41,7 @@ struct ESPState {
>> uint32_t cmdlen;
>> uint32_t do_cmd;
>> + bool data_in_ready;
>> int dma_enabled;
>> uint32_t async_len;
>
> Whilst doing some testing earlier, I discovered that the same change is required in
> do_dma_pdma_cb(): it seems during boot the ROM attempts several 128k DMA requests in a row, and with
> heavy debugging enabled it's enough to trigger the same underflow problem.
>
> Fortunately the fix is easy, and I'll squash this into the v3 series:
>
> diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
> index a175191718..248c1ce27a 100644
> --- a/hw/scsi/esp.c
> +++ b/hw/scsi/esp.c
> @@ -453,17 +453,13 @@ static void do_dma_pdma_cb(ESPState *s)
> } else {
> if (s->async_len == 0) {
> if (s->current_req) {
> + /*
> + * Defer until the scsi layer has completed.
> + */
> scsi_req_continue(s->current_req);
> + s->data_in_ready = false;
> }
> -
> - /*
> - * If there is still data to be read from the device then
> - * complete the DMA operation immediately. Otherwise defer
> - * until the scsi layer has completed.
> - */
> - if (esp_get_tc(s) != 0 || s->ti_size == 0) {
> - return;
> - }
> + return;
> }
>
> if (esp_get_tc(s) != 0) {
>
>
> ATB,
>
> Mark.
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
next prev parent reply other threads:[~2021-03-03 19:53 UTC|newest]
Thread overview: 135+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-09 19:29 [PATCH v2 00/42] esp: consolidate PDMA transfer buffers and other fixes Mark Cave-Ayland
2021-02-09 19:29 ` [PATCH v2 01/42] esp: checkpatch fixes Mark Cave-Ayland
2021-03-01 19:43 ` Laurent Vivier
2021-03-03 8:33 ` Mark Cave-Ayland
2021-02-09 19:29 ` [PATCH v2 02/42] esp: rename existing ESP QOM type to SYSBUS_ESP Mark Cave-Ayland
2021-02-10 22:29 ` Philippe Mathieu-Daudé
2021-03-01 19:52 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 03/42] esp: QOMify the internal ESP device state Mark Cave-Ayland
2021-02-12 18:51 ` Philippe Mathieu-Daudé
2021-02-15 22:29 ` Mark Cave-Ayland
2021-03-01 20:11 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 04/42] esp: add vmstate_esp version to embedded ESPState Mark Cave-Ayland
2021-02-16 7:35 ` Philippe Mathieu-Daudé
2021-03-01 20:21 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 05/42] esp: add trace event when receiving a TI command Mark Cave-Ayland
2021-03-01 20:24 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 06/42] esp: fix esp_reg_read() trace event Mark Cave-Ayland
2021-03-01 20:29 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 07/42] esp: add PDMA trace events Mark Cave-Ayland
2021-03-01 20:32 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 08/42] esp: determine transfer direction directly from SCSI phase Mark Cave-Ayland
2021-02-16 7:36 ` Philippe Mathieu-Daudé
2021-03-01 21:18 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 09/42] esp: introduce esp_get_tc() and esp_set_tc() Mark Cave-Ayland
2021-03-01 21:24 ` Laurent Vivier
2021-03-03 8:35 ` Mark Cave-Ayland
2021-02-09 19:29 ` [PATCH v2 10/42] esp: introduce esp_get_stc() Mark Cave-Ayland
2021-02-10 22:33 ` Philippe Mathieu-Daudé
2021-02-11 7:53 ` Mark Cave-Ayland
2021-02-11 10:07 ` Philippe Mathieu-Daudé
2021-02-15 22:27 ` Mark Cave-Ayland
2021-03-01 21:28 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 11/42] esp: apply transfer length adjustment when STC is zero at TC load time Mark Cave-Ayland
2021-02-16 7:33 ` Philippe Mathieu-Daudé
2021-02-16 21:52 ` Mark Cave-Ayland
2021-03-01 21:35 ` Laurent Vivier
2021-03-03 8:44 ` Mark Cave-Ayland
2021-02-09 19:29 ` [PATCH v2 12/42] esp: remove dma_counter from ESPState Mark Cave-Ayland
2021-02-10 22:37 ` Philippe Mathieu-Daudé
2021-03-01 21:44 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 13/42] esp: remove dma_left " Mark Cave-Ayland
2021-02-23 21:22 ` Philippe Mathieu-Daudé
2021-03-01 21:50 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 14/42] esp: remove minlen restriction in handle_ti Mark Cave-Ayland
2021-02-23 18:24 ` Philippe Mathieu-Daudé
2021-03-01 22:04 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 15/42] esp: introduce esp_pdma_read() and esp_pdma_write() functions Mark Cave-Ayland
2021-02-10 22:51 ` Philippe Mathieu-Daudé
2021-02-11 8:01 ` Mark Cave-Ayland
2021-02-11 10:09 ` Philippe Mathieu-Daudé
2021-03-01 22:06 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 16/42] esp: use pdma_origin directly in esp_pdma_read()/esp_pdma_write() Mark Cave-Ayland
2021-02-23 18:25 ` Philippe Mathieu-Daudé
2021-03-01 22:07 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 17/42] esp: move pdma_len and TC logic into esp_pdma_read()/esp_pdma_write() Mark Cave-Ayland
2021-02-23 21:23 ` Philippe Mathieu-Daudé
2021-03-01 22:09 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 18/42] esp: accumulate SCSI commands for PDMA transfers in cmdbuf instead of pdma_buf Mark Cave-Ayland
2021-02-23 21:25 ` Philippe Mathieu-Daudé
2021-03-02 17:02 ` Laurent Vivier
2021-03-02 17:34 ` Mark Cave-Ayland
2021-03-02 17:59 ` Laurent Vivier
2021-03-02 19:29 ` Mark Cave-Ayland
2021-03-02 21:21 ` Laurent Vivier
2021-03-02 21:22 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 19/42] esp: remove buf parameter from do_cmd() Mark Cave-Ayland
2021-02-23 18:27 ` Philippe Mathieu-Daudé
2021-03-02 17:03 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 20/42] esp: remove the buf and buflen parameters from get_cmd() Mark Cave-Ayland
2021-02-16 7:31 ` Philippe Mathieu-Daudé
2021-03-02 17:03 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 21/42] esp: remove redundant pdma_start from ESPState Mark Cave-Ayland
2021-03-02 21:22 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 22/42] esp: move PDMA length adjustments into esp_pdma_read()/esp_pdma_write() Mark Cave-Ayland
2021-03-02 21:44 ` Laurent Vivier
2021-02-09 19:29 ` [PATCH v2 23/42] esp: use ti_wptr/ti_rptr to manage the current FIFO position for PDMA Mark Cave-Ayland
2021-02-23 21:29 ` Philippe Mathieu-Daudé
2021-03-02 21:47 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 24/42] esp: use in-built TC to determine PDMA transfer length Mark Cave-Ayland
2021-02-23 18:32 ` Philippe Mathieu-Daudé
2021-03-02 21:48 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 25/42] esp: remove CMD pdma_origin Mark Cave-Ayland
2021-02-23 18:34 ` Philippe Mathieu-Daudé
2021-03-02 21:49 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 26/42] esp: rename get_cmd_cb() to esp_select() Mark Cave-Ayland
2021-03-02 21:51 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 27/42] esp: fix PDMA target selection Mark Cave-Ayland
2021-03-02 21:57 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 28/42] esp: use FIFO for PDMA transfers between initiator and device Mark Cave-Ayland
2021-03-02 22:02 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 29/42] esp: remove pdma_origin from ESPState Mark Cave-Ayland
2021-03-02 22:03 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 30/42] esp: add 4 byte PDMA read and write transfers Mark Cave-Ayland
2021-02-12 18:56 ` Philippe Mathieu-Daudé
2021-02-15 22:35 ` Mark Cave-Ayland
2021-02-16 7:30 ` Philippe Mathieu-Daudé
2021-02-16 21:36 ` Mark Cave-Ayland
2021-02-23 8:24 ` Mark Cave-Ayland
2021-02-23 18:55 ` Philippe Mathieu-Daudé
2021-03-02 22:05 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 31/42] esp: implement FIFO flush command Mark Cave-Ayland
2021-03-03 19:32 ` Laurent Vivier
2021-03-04 18:26 ` Mark Cave-Ayland
2021-02-09 19:30 ` [PATCH v2 32/42] esp: latch individual bits in ESP_RINTR register Mark Cave-Ayland
2021-03-03 19:48 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 33/42] esp: defer command completion interrupt on incoming data transfers Mark Cave-Ayland
2021-02-18 17:25 ` Mark Cave-Ayland
2021-03-03 19:52 ` Laurent Vivier [this message]
2021-03-04 18:33 ` Mark Cave-Ayland
2021-02-09 19:30 ` [PATCH v2 34/42] esp: remove old deferred command completion mechanism Mark Cave-Ayland
2021-03-03 19:55 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 35/42] esp: raise interrupt after every non-DMA byte transferred to the FIFO Mark Cave-Ayland
2021-03-03 19:59 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 36/42] esp: add maxlen parameter to get_cmd() Mark Cave-Ayland
2021-02-23 18:43 ` Philippe Mathieu-Daudé
2021-03-03 20:04 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 37/42] esp: transition to message out phase after SATN and stop command Mark Cave-Ayland
2021-03-03 20:06 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 38/42] esp: convert ti_buf from array to Fifo8 Mark Cave-Ayland
2021-02-23 18:49 ` Philippe Mathieu-Daudé
2021-02-25 9:15 ` Mark Cave-Ayland
2021-03-03 20:11 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 39/42] esp: convert cmdbuf " Mark Cave-Ayland
2021-03-03 20:16 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 40/42] esp: add trivial implementation of the ESP_RFLAGS register Mark Cave-Ayland
2021-02-23 18:51 ` Philippe Mathieu-Daudé
2021-03-03 20:19 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 41/42] esp: implement non-DMA transfers in PDMA mode Mark Cave-Ayland
2021-03-03 20:22 ` Laurent Vivier
2021-02-09 19:30 ` [PATCH v2 42/42] esp: add support for unaligned accesses Mark Cave-Ayland
2021-03-03 20:22 ` Laurent Vivier
2021-02-23 21:32 ` [PATCH v2 00/42] esp: consolidate PDMA transfer buffers and other fixes Philippe Mathieu-Daudé
2021-02-25 9:54 ` Mark Cave-Ayland
2021-02-25 10:50 ` Philippe Mathieu-Daudé
2021-02-25 19:17 ` Mark Cave-Ayland
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ba080d8e-1e21-e5e9-8fff-568db5f73b3f@vivier.eu \
--to=laurent@vivier.eu \
--cc=fam@euphon.net \
--cc=mark.cave-ayland@ilande.co.uk \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).