| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAPDyKFoiroN0XX0CLHGxZYHCZmWe11z5avH=KnS7qpkuNdwxiw@mail.gmail.com>
Date: Tue, 3 Oct 2023 14:15:05 +0200
From: Ulf Hansson <ulf.hansson@...aro.org>
To: Victor Shih <victorshihgli@...il.com>
Cc: adrian.hunter@...el.com, linux-mmc@...r.kernel.org,
linux-kernel@...r.kernel.org, benchuanggli@...il.com,
HL.Liu@...esyslogic.com.tw, Greg.tu@...esyslogic.com.tw,
takahiro.akashi@...aro.org, dlunev@...omium.org,
Ben Chuang <ben.chuang@...esyslogic.com.tw>,
Victor Shih <victor.shih@...esyslogic.com.tw>
Subject: Re: [PATCH V12 18/23] mmc: sdhci-uhs2: add request() and others
On Fri, 15 Sept 2023 at 11:44, Victor Shih <victorshihgli@...il.com> wrote:
>
> From: Victor Shih <victor.shih@...esyslogic.com.tw>
>
> This is a sdhci version of mmc's request operation.
> It covers both UHS-I and UHS-II.
Okay, but again, please elaborate on why we need/want this.
>
> Signed-off-by: Ben Chuang <ben.chuang@...esyslogic.com.tw>
> Signed-off-by: AKASHI Takahiro <takahiro.akashi@...aro.org>
> Signed-off-by: Victor Shih <victor.shih@...esyslogic.com.tw>
> ---
>
> Updates in V11:
> - Drop the check mmc_card_uhs2_hd_mode(host->mmc)
> in sdhci_uhs2_set_transfer_mode().
>
> Updates in V10:
> - Use tmode_half_duplex to instead of uhs2_tmode0_flag
> in sdhci_uhs2_set_transfer_mode().
>
> Updates in V9:
> - Modify the annotations in __sdhci_uhs2_send_command().
>
> Updates in V8:
> - Adjust the position of matching brackets in
> sdhci_uhs2_send_command_retry().
> - Modify CameCase definition in __sdhci_uhs2_finish_command().
> - Modify error message in __sdhci_uhs2_finish_command().
> - sdhci_uhs2_send_command_retry() to instead of sdhci_uhs2_send_command()
> in sdhci_uhs2_request().
> - Use sdhci_uhs2_mode() to simplify code in sdhci_uhs2_request_atomic().
> - Add forward declaration for sdhci_send_command().
>
> Updates in V7:
> - Cancel export state of some functions.
> - Remove unnecessary whitespace changes.
>
> Updates in V6:
> - Add uhs2_dev_cmd() to simplify code.
> - Remove unnecessary functions.
> - Cancel export state of some functions.
> - Drop use CONFIG_MMC_DEBUG().
> - Wrap at 100 columns in some functions.
>
> ---
>
> drivers/mmc/host/sdhci-uhs2.c | 412 ++++++++++++++++++++++++++++++++++
> drivers/mmc/host/sdhci.c | 49 ++--
> drivers/mmc/host/sdhci.h | 8 +
> 3 files changed, 454 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/mmc/host/sdhci-uhs2.c b/drivers/mmc/host/sdhci-uhs2.c
> index 09b86fec9f7b..1f8d527424fd 100644
> --- a/drivers/mmc/host/sdhci-uhs2.c
> +++ b/drivers/mmc/host/sdhci-uhs2.c
[...]
> +
> +static void __sdhci_uhs2_send_command(struct sdhci_host *host, struct mmc_command *cmd)
> +{
> + int i, j;
> + int cmd_reg;
> +
> + i = 0;
> + sdhci_writel(host,
> + ((u32)cmd->uhs2_cmd->arg << 16) |
> + (u32)cmd->uhs2_cmd->header,
> + SDHCI_UHS2_CMD_PACKET + i);
> + i += 4;
> +
> + /*
> + * Per spec, payload (config) should be MSB before sending out.
> + * But we don't need convert here because had set payload as
> + * MSB when preparing config read/write commands.
> + */
> + for (j = 0; j < cmd->uhs2_cmd->payload_len / sizeof(u32); j++) {
> + sdhci_writel(host, *(cmd->uhs2_cmd->payload + j), SDHCI_UHS2_CMD_PACKET + i);
> + i += 4;
> + }
> +
> + for ( ; i < SDHCI_UHS2_CMD_PACK_MAX_LEN; i += 4)
> + sdhci_writel(host, 0, SDHCI_UHS2_CMD_PACKET + i);
> +
> + DBG("UHS2 CMD packet_len = %d.\n", cmd->uhs2_cmd->packet_len);
> + for (i = 0; i < cmd->uhs2_cmd->packet_len; i++)
> + DBG("UHS2 CMD_PACKET[%d] = 0x%x.\n", i,
> + sdhci_readb(host, SDHCI_UHS2_CMD_PACKET + i));
We are ignoring what we just read. Isn't there something we need to verify?
Moreover, the whole thing with i,j and the +4 thing above looks a bit
odd to me. I am not sure whether the above can be simplified, but I
leave that for you to have a second look at.
> +
> + cmd_reg = FIELD_PREP(SDHCI_UHS2_CMD_PACK_LEN_MASK, cmd->uhs2_cmd->packet_len);
> + if ((cmd->flags & MMC_CMD_MASK) == MMC_CMD_ADTC)
> + cmd_reg |= SDHCI_UHS2_CMD_DATA;
> + if (cmd->opcode == MMC_STOP_TRANSMISSION)
> + cmd_reg |= SDHCI_UHS2_CMD_CMD12;
> +
> + /* UHS2 Native ABORT */
> + if ((cmd->uhs2_cmd->header & UHS2_NATIVE_PACKET) &&
> + (uhs2_dev_cmd(cmd) == UHS2_DEV_CMD_TRANS_ABORT))
> + cmd_reg |= SDHCI_UHS2_CMD_TRNS_ABORT;
> +
> + /* UHS2 Native DORMANT */
> + if ((cmd->uhs2_cmd->header & UHS2_NATIVE_PACKET) &&
> + (uhs2_dev_cmd(cmd) == UHS2_DEV_CMD_GO_DORMANT_STATE))
> + cmd_reg |= SDHCI_UHS2_CMD_DORMANT;
> +
> + DBG("0x%x is set to UHS2 CMD register.\n", cmd_reg);
> +
> + sdhci_writew(host, cmd_reg, SDHCI_UHS2_CMD);
> +}
[...]
> +static bool sdhci_uhs2_send_command_retry(struct sdhci_host *host,
> + struct mmc_command *cmd,
> + unsigned long flags)
> + __releases(host->lock)
> + __acquires(host->lock)
> +{
> + struct mmc_command *deferred_cmd = host->deferred_cmd;
> + int timeout = 10; /* Approx. 10 ms */
> + bool present;
Why do we need a retry mechanism at this level? The mmc core sometimes
retries commands when it seems reasonable, why isn't that sufficient?
> +
> + while (!sdhci_uhs2_send_command(host, cmd)) {
> + if (!timeout--) {
> + pr_err("%s: Controller never released inhibit bit(s).\n",
> + mmc_hostname(host->mmc));
> + sdhci_dumpregs(host);
> + cmd->error = -EIO;
> + return false;
> + }
> +
> + spin_unlock_irqrestore(&host->lock, flags);
> +
> + usleep_range(1000, 1250);
> +
> + present = host->mmc->ops->get_cd(host->mmc);
> +
> + spin_lock_irqsave(&host->lock, flags);
> +
> + /* A deferred command might disappear, handle that */
> + if (cmd == deferred_cmd && cmd != host->deferred_cmd)
> + return true;
> +
> + if (sdhci_present_error(host, cmd, present))
> + return false;
> + }
If the retry is needed, would it be possible to convert into using
read_poll_timeout() for the above while loop instead? If so, please
make that conversion.
> +
> + if (cmd == host->deferred_cmd)
> + host->deferred_cmd = NULL;
> +
> + return true;
> +}
> +
> +static void __sdhci_uhs2_finish_command(struct sdhci_host *host)
> +{
> + struct mmc_command *cmd = host->cmd;
> + u8 resp;
> + u8 ecode;
> + bool breada0 = 0;
Nitpick: Maybe find some better variable names. Like error_code...
> + int i;
> +
> + if (host->mmc->flags & MMC_UHS2_SD_TRAN) {
> + resp = sdhci_readb(host, SDHCI_UHS2_RESPONSE + 2);
> + if (resp & UHS2_RES_NACK_MASK) {
> + ecode = (resp >> UHS2_RES_ECODE_POS) & UHS2_RES_ECODE_MASK;
> + pr_err("%s: NACK response, ECODE=0x%x.\n", mmc_hostname(host->mmc), ecode);
> + }
> + breada0 = 1;
> + }
> +
> + if (cmd->uhs2_resp &&
> + cmd->uhs2_resp_len && cmd->uhs2_resp_len <= 20) {
> + /* Get whole response of some native CCMD, like
> + * DEVICE_INIT, ENUMERATE.
> + */
> + for (i = 0; i < cmd->uhs2_resp_len; i++)
> + cmd->uhs2_resp[i] = sdhci_readb(host, SDHCI_UHS2_RESPONSE + i);
> + } else {
> + /* Get SD CMD response and Payload for some read
> + * CCMD, like INQUIRY_CFG.
> + */
> + /* Per spec (p136), payload field is divided into
> + * a unit of DWORD and transmission order within
> + * a DWORD is big endian.
> + */
> + if (!breada0)
> + sdhci_readl(host, SDHCI_UHS2_RESPONSE);
> + for (i = 4; i < 20; i += 4) {
Again we do sdhci_readl above but just ignore the data. I assume
that's deliberate, as we are probably just interested in the remaining
pieces.
Moreover, the whole thing with +4 things continues to look a bit odd
to me. I am not sure whether it can be simplified, but I leave that
for you to have a second look at.
> + cmd->resp[i / 4 - 1] =
> + (sdhci_readb(host,
> + SDHCI_UHS2_RESPONSE + i) << 24) |
> + (sdhci_readb(host,
> + SDHCI_UHS2_RESPONSE + i + 1)
> + << 16) |
> + (sdhci_readb(host,
> + SDHCI_UHS2_RESPONSE + i + 2)
> + << 8) |
> + sdhci_readb(host, SDHCI_UHS2_RESPONSE + i + 3);
> + }
> + }
> +}
[...]
> +
> +void sdhci_uhs2_request(struct mmc_host *mmc, struct mmc_request *mrq)
static void
> +{
> + struct sdhci_host *host = mmc_priv(mmc);
> + struct mmc_command *cmd;
> + unsigned long flags;
> + bool present;
> +
> + if (!(sdhci_uhs2_mode(host))) {
> + sdhci_request(mmc, mrq);
> + return;
> + }
> +
> + mrq->stop = NULL;
> + mrq->sbc = NULL;
> + if (mrq->data)
> + mrq->data->stop = NULL;
> +
> + /* Firstly check card presence */
> + present = mmc->ops->get_cd(mmc);
> +
> + spin_lock_irqsave(&host->lock, flags);
> +
> + if (sdhci_present_error(host, mrq->cmd, present))
> + goto out_finish;
> +
> + cmd = mrq->cmd;
> +
> + if (!sdhci_uhs2_send_command_retry(host, cmd, flags))
> + goto out_finish;
> +
> + spin_unlock_irqrestore(&host->lock, flags);
> +
> + return;
> +
> +out_finish:
> + sdhci_finish_mrq(host, mrq);
> + spin_unlock_irqrestore(&host->lock, flags);
> +}
> +EXPORT_SYMBOL_GPL(sdhci_uhs2_request);
Drop this, it's not used outside this module.
> +
> +int sdhci_uhs2_request_atomic(struct mmc_host *mmc, struct mmc_request *mrq)
This function is entirely unused. Did you actually test this with an
hsq enabled host? Or perhaps you have just added this for
completeness?
> +{
> + struct sdhci_host *host = mmc_priv(mmc);
> + struct mmc_command *cmd;
> + unsigned long flags;
> + int ret = 0;
> +
> + if (!sdhci_uhs2_mode(host))
> + return sdhci_request_atomic(mmc, mrq);
> +
> + spin_lock_irqsave(&host->lock, flags);
> +
> + if (sdhci_present_error(host, mrq->cmd, true)) {
> + sdhci_finish_mrq(host, mrq);
> + goto out_finish;
> + }
> +
> + cmd = mrq->cmd;
> +
> + /*
> + * The HSQ may send a command in interrupt context without polling
> + * the busy signaling, which means we should return BUSY if controller
> + * has not released inhibit bits to allow HSQ trying to send request
> + * again in non-atomic context. So we should not finish this request
> + * here.
> + */
> + if (!sdhci_uhs2_send_command(host, cmd))
> + ret = -EBUSY;
> +
> +out_finish:
> + spin_unlock_irqrestore(&host->lock, flags);
> + return ret;
> +}
> +EXPORT_SYMBOL_GPL(sdhci_uhs2_request_atomic);
> +
[...]
Kind regards
Uffe
Powered by blists - more mailing lists